0% found this document useful (0 votes)

351 views74 pages

TM Forum AI in Practice Benchmark Report wHhznq0

Uploaded by

stankh8281

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

351 views74 pages

TM Forum AI in Practice Benchmark Report wHhznq0

Uploaded by

stankh8281

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 74

TM Forum | March 2024

BENCHMARK

building an
AI strategy:
telcos put the
foundations
in place
Author:
Mark Newman, Chief Analyst

Editor:
Ian Kemp, Managing Editor

In addition, we create value with

intellectual property and long-
term research, led by the award-
winning Nokia Bell Labs. Tecnotree is a 5G-ready digital
CSG empowers companies to Business Support System (BSS)
build unforgettable experiences, Service providers, enterprises
and partners worldwide trust player, with AI/ML capabilities and
making it easier for people and multi-cloud extensibility. Tecnotree
businesses to connect with, use Nokia to deliver secure, reliable
and sustainable networks today leads the way on the TM Forum
and pay for the services they Open API Conformance with 59
value most. Our customer – and work with us to create the
digital services and applications certified Open APIs including 9
experience, billing and payments real-world open APIs, a testament
solutions help companies of any of the future.
to the company’s commitment to
size make money and make a excellence, and continuously
difference. With our SaaS striving to deliver differentiated
solutions, company leaders can experiences and services to both
take control of their future, and CSPs and DSPs. Our agile and
tap into guidance along the way ServiceNow helps Telcos accelerate
open-source digital BSS Stack
from our more than 5k-strong growth, maximize network
comprises the full range (order-to-
experienced global team. Want to investment, and reduce costs
cash) of business processes and
learn more about how to be a across the entire digital
subscription management for
change maker and industry ecosystem—all while delivering
telecom and other digital services
shaper like our 1,000-plus clients? seamless experiences. Be it
industries creating opportunities
Visit csgi.com to learn more. planning for 5G or fiber buildouts,
beyond connectivity. Tecnotree
managing data centers, launching
also provides Fintech and B2B2X
and delivering new services, or
multi-experience digital
automating network operations,
marketplace to its subscriber base
our purpose-built solutions help
through the Tecnotree Moments
make work easier across the entire
platform to empower digitally
At Nokia, we create technology service lifecycle with automation
connected communities across
that helps the world act together. and AI. Built for the fast-changing
gaming, health, education, OTT,
world of Telecom, the Now Platform
As a B2B technology innovation and other vertical ecosystems.
connects people and data for
leader, we are pioneering Tecnotree is listed on the Helsinki
greater productivity and innovation.
networks that sense, think and act Nasdaq (TEM1V).
by leveraging our work across https://www.servicenow.com/
mobile, fixed and cloud networks.

To learn more about our sponsors’ views on AI technologies and strategies watch these videos:

Aria AWS Cognizant CSG

Nokia Servicenow Tecnotree

4 inform.tmforum.org
BENCHMARK

the big picture

The AI journey started in 2016 or 2017 for some of the more progressive, ambitious operators, and in 2019 or
2020 for most others. But it is only in the past 12-15 months with the explosion of interest in Generative AI
(GenAI) that it has become a key focus for most boardrooms and the technology which communications service
providers (CSPs) are pinning their hopes on to transform their financial fortunes.

T
he purpose of this report is Two of the 20 operators that we
to understand where CSPs interviewed – China Mobile and
are on their AI journeys. It is SK Telecom – are building AI
based on briefings with AI businesses with a clear objective
decision-makers in 20 of the to monetize their investments. The
world’s largest and leading CSPs other 18 are investing in AI for their I’d say every meeting
globally. We believe they own internal usage, principally to I go to, nearly
represent some of the more drive efficiencies and improve
progressive CSPs in terms of their customer experience. Addressing
without exception,
investment in, and commitment the growing complexity of their there’s some
to, AI technologies. As such, they businesses – and specifically their component that
present a slightly distorted picture network operations functions – is
of the global status of CSP use of also a key driver.
discusses what we’re
AI. However, the challenges and All the operators that we spoke to
doing in AI – next
opportunities they speak about in are ramping up their investments year’s investment
relation to AI are common to the
sector as a whole.
in AI – be it building teams of AI planning, execution
experts or building platforms and
We also publish the results of a tools in readiness to start scaling
skills, people
survey in this research in the form AI deployments. development,
of charts throughout the report. “I’d say every meeting I go to, everything.”
The survey – which we also used nearly without exception, there’s
for an earlier report, Generative AI: some component that discusses Mark Sanders,
operators take their first steps – what we’re doing in AI – next year’s Chief Architect, Telstra
covers a broader cross-section of investment planning, execution
CSPs, with responses from 104 skills, people development,
executives across 73 CSPs globally.

5 inform.tmforum.org
the big picture

How effectively is your organization exploiting AI and machine learning?

We have made good progress in using AI

and machine learning in different parts of
55%
our business (e.g.customer experience or
network operations)

We have some trials and proofs of

concept running but we haven't 31%
deployed it in our business

We are at the early stages of researching

14%
the potential of AI and machine learning

TM Forum, 2023

everything,” says Mark Sanders, We expect this to change during What’s more, at a time when every
Chief Architect, Telstra. 2024 and, more likely, in 2025. person in the CSP organization is
However, it is important to draw a Somewhat ironically, this encouraged to be more customer
distinction between “traditional” enthusiasm for GenAI may end up centric, GenAI offers the promise
types of AI – what we refer to in boosting interest in predictive AI. of delivering insights and value
this research as predictive AI – While some use cases involve only from the millions of conversations
and GenAI. GenAI, many others require the and messages that pass between
two to be used in tandem. customers, customer service
CSP investment in predictive AI agents and online and mobile
has, until now, been relatively Furthermore, when it comes to
using AI for automation GenAI has platforms. Even the least
modest. Most operators have technology-savvy people in an
deployed a handful of use cases in its limitations, particularly in the
network and the drive towards organization can get their heads
areas such as energy saving, churn around what GenAI can do.
prediction, and next best action autonomous networks.
which uses AI and real-time data In speaking to CSPs about the
to target customer experiences. status of their AI journeys most
A realistic assessment
But they have lacked the skills, the conversations tend to end up AI is neither, on the one hand, the
tools and compelling evidence of focusing more on GenAI than answer to the challenges that CSPs
good returns on investment to predictive AI. Is this because CSPs face nor, on the other, a wildly
scale rollout. Furthermore, use believe there is far more potential overhyped technology that will fail
cases have been more “bolt-on”, than in machine learning and to have a meaningful impact on
rather ones that sit within an predictive AI? Or is it just because operator organizations. AI will be
overarching architecture or GenAI is the newer technology adopted by operators in the same
operating model. and has captured the imagination way that it will be adopted by
of boardrooms and their advisers? almost every other sector. But will
The focus on GenAI it enable CSPs to significantly cut
A key reason for GenAI’s appeal is
costs, to create new products,
Over the past year to 15 months its accessibility compared with
services and revenue streams, and
CSPs’ AI strategies have pivoted machine learning and predictive
to redefine the relationship with
to focus on GenAI. Indeed, in AI, and the fact that it uses the
their customers?
many of our conversations for this same inputs and provides the
report GenAI was the main topic same outputs as the tools we use The cost discussion is, perhaps,
of discussion. Most CSPs have in our everyday working lives. the most interesting one, and
drawn up well in excess of 100 use GenAI can take our conversations, lessons should be learned from
cases and are engaged in a the emails we write, the the experience of public cloud
process of aggressive documents we use, the online which was originally touted as a
experimentation. But very few research that we conduct and the cost-saving opportunity for CSPs.
have been put into production. presentations we make and add Operators are under pressure to
value to them. cut their costs because of revenue

6 inform.tmforum.org
the big picture

stagnation. In many cases AI- There are few examples today could result from the successful
enabled automation can help of AI use cases / deployments use and deployment of AI.
employees do their current jobs that have delivered significant In a “bullish” scenario total
more efficiently and more quickly. or clearly demonstrable value capex and opex for telecoms
But will operators be able to to the CSP business. Those operators globally falls by
capture these time savings and areas where CSPs have had 9.1% from $1.8 trillion to
convert them into the elimination early success include energy $1.64 trillion. A “bearish”
of different roles? Or will it simply saving, network planning / scenario sees a 2% decline
allow employees to focus their smart capex and churn to $1.77 trillion.
time on other tasks and functions prediction models.
that create more value? Ultimately, Despite the buzz around GenAI Recommendations
this will come down to very few use cases have been Operators need to develop
departmental budgeting, the put into production. Many CSPs easy access to data that sits
pressure to reduce capex and are hoping that 2024 will be across the entire organization
opex, and whether AI and the year GenAI-enabled use if they are going to deploy AI
automation are two of the tools cases become a reality. at scale.
that CSPs lean on to cut costs. Future predictive AI uses
If operators are to shape how
We should also not forget that AI cases are increasingly seen AI will change their businesses
will require operators to invest in as principally delivering they must fundamentally
new roles, new technologies and value to network operations embrace these new
platforms and, in some cases, new as CSPs seek to build technologies and make them
infrastructure. In addition, as we autonomous networks. integral to their operations. If
see in section 2, new skills tend to GenAI is most likely to see
they do not, they risk being
be more costly than old ones. wide-scale deployment in locked into technology choices
Below are some of the key customer-facing functions and architectures.
findings and recommendations and as a tool for driving CSPs must not repeat the
from our research. For a fuller greater productivity generally. mistakes they made in public
summary see section 9. Operator spending on AI (as cloud where many tried to
opposed to analytics) has compete with hyperscaler
Key findings been modest until now, but service providers. Their focus
There is nearly universal many CSPs are starting to in AI should be on the
recognition that AI will lead spend significantly more on deployment of use cases and
to a paradigm shift in CSPs’ hyperscaler services that the creation of platforms for
operating models. However, embed GenAI. delivering AI capabilities to
very few operators have a clear their internal organization.
As they embrace GenAI CSPs
picture of what that operating are becoming increasingly For operators to deploy AI at
model will or should be. reliant on hyperscale service scale and to industrialize
AI-ready access to clean, providers and the public cloud. deployments they need to
quality, usable data is the embrace machine learning
Very few operators are

single biggest challenge to operations (MLOps) and
building AI infrastructure.
the deployment of AI at scale. FinOps in the same way that
Rather, CSP investments are
the software sector was
Most AI deployments today
focused on developing use
transformed with the use of
are “bolt-on” use cases cases and building platforms
DevOps.
– specific, isolated deployments for teams across the business
rather than ones that represent to use different AI tools. In the first section we look at the
part of an overall architecture different approaches CSPs are
Our analysis of CSP capex and
and operating model. taking to build organizations able
opex presents two scenarios
to harness AI technologies.
for the potential savings that

7 inform.tmforum.org
BENCHMARK

section 1:

building an AI-ready
organization
Finding the right balance between creating central functions and capabilities on the one hand, and
democratizing access to AI tools on the other, is a major challenge for CSPs. As a result a variety of roles,
organizational alignments and governance functions are being developed to get this balance right.

O
perators had been The level of comfort with deployment of different AI
building and evolving embracing new technologies models and services across
their data analytics varies enormously. Every the organization. An AI
organizations for several years effort needs to be made to governance program needs
before AI emerged as a strategic make AI easy to use, even to be put into place to reduce
priority. AI functions and for teams with low technology the risk of incurring debt
organizations tend to sit on top capabilities. while, at the same time,
of these data analytics teams – Reusing successful use cases encouraging usage and
although in some cases they can be a major challenge. The experimentation.
are part of the same team – and central AI team / platform In this section we look at some
their size and remit tends to needs to be able to see where examples, from our research, of
be a function of how much successful use cases are being the different organizational
AI technology an operator is rolled out and facilitate their approaches that CSPs are taking
building and what, and how reuse across the organization. to harness AI.
much, it is buying.
Cost management risks Creating a CXO role for AI.
Any future organizational function become a serious issue if the As the profile of AI, and the
and structure needs to factor in democratization of access to expectations for what it can
the following considerations: AI is not accompanied by the deliver, have risen it has become
AI is a tool for innovation, and deployment of new FinOps increasingly important that there
anyone in the organization can practices for AI. is expertise and representation at
potentially be an innovator. There is a risk of incurring – or just below – board level. Two
As such, AI needs to be made technical debt through the of the CSPs that we spoke to for
available to every team in uncontrolled purchase and this report have appointed Chief
the company. AI Officers (CAIOs). Steve Jarrett,

8 inform.tmforum.org
section 1: building an AI-ready organization

previously Group Senior Vice Chief Digital Officer, Chema Alonso,

President, Data and AI, at Orange who created the voice-controlled
Innovation, was appointed Chief AI platform Aura in 2017, is
AI Officer of the French operator responsible for customer insights,
in December 2023. In his LinkedIn while Juan Manuel Caro Bernat is
profile Jarrett says his team “leads China Mobile Director of Autonomous Networks,
our ‘Data Democracy’ cultural and SK Telecom responsible for managing the
and technical transformation of whole lifecycle of networks and AI.
the company”. In his role Jarrett are investing
This splitting of responsibilities is
also leads the selection of key aggressively in particularly interesting given that
partners and the identification AI to deliver new the focus of GenAI is more towards
of potential acquisitions.
products and customer-facing rather than
BT has a Chief Data and AI Officer network functions, while predictive
– Deepika Adusumilli – who joined services. AI is increasingly a technology
the company in October 2023. used for network automation.
Adusumilli reports to Harmeen AI as a new business function.
Mehta, Chief Digital and China Mobile and SK Telecom are
responsibility, for example, for the
Information Officer. investing aggressively in AI to
company’s data architecture.
While a CAIO is responsible for AI, deliver new products and services.
Setting up a subsidiary
a Chief Digital Officer (CDO) looks The focus for both operators is
organization to develop AI. Many
after all “digital” technologies. In building their own Large Language
CSPs have set up subsidiaries in
CSPs this tends not to include the Models (LLMs) – and solutions and
recent years to build capabilities,
network itself which is the capabilities that sit on top of them
products and services in new tech
responsibility of a Chief Technology – and selling access to them to
and digital business areas. But
Officer or a Chief Network Officer. enterprises and other CSPs.
these have focused on building
SK Telecom is another CSP with a new products, services and China Mobile’s Jiutian business
high-profile AI role. Chung Suk- revenue streams rather than has been set up both to provide
guen is Chief Global AI Officer at delivering capabilities back into services internally to the
SK Telecom, but his role is the organization. organization and to deliver B2B
principally focused on GenAI and services. In October the operator
Proximus Ada is an affiliate of
building products that can be announced a Jiutian GenAI model
Belgian telecoms operator
leveraged internally and monetized which it says has “won support”
Proximus that is specifically
through partnerships with other from large enterprises including
focused on developing
CSPs and, potentially, software and China Ocean Shipping and China
cybersecurity and AI capabilities.
services vendors. Railway Construction Co. The
Its work is split between meeting
Rather than having a single CXO model has trained more than 2
internal Proximus requirements
responsible for AI, some CSPs million tokens – a basic unit of
and delivering services to B2B
divide responsibilities between text or code that an LLM uses to
customers. Proximus Chief Digital
executives. African telecoms group process and generate language –
Information Officer (CDIO),
MTN has an AI steering committee and has “knowledge” in eight
Antonietta Mastroianni, says Ada
co-chaired by the Group CIO and industries including telecoms,
was created to help find the
Chief Strategy and Transformation energy, and steel and transport,
requisite skills in cybersecurity
Officer, Chika Ekeji. says China Mobile.
and AI and to build a flexible
Chief Data Officer is also a role organizational culture that was “We believe that external success
which exists within many CSP different from the core business. is the real measure of success
organizations. Perhaps the [of Jiutian],” a China Mobile
Splitting the internal and
highest-profile Chief Data Officer spokesperson told TM Forum.
customer-facing AI functions.
in telecoms was Chema Alonso at Rather than building a central AI Establishing an AI center of
Telefónica, although he has now organization, Telefónica has excellence. In our survey 53% of
transitioned to Chief Digital decided to split responsibilities into CSPs said they had established
Officer with responsibility for AI in two: customer insights and a center of excellence for AI
customer solutions. Telefónica still innovation; and network, IT and (see chart on the next page). But
has a Chief Data Officer, but this is internal AI-driven transformation. the precise size, shape and role
now an internal role, with of these varies considerably.

9 inform.tmforum.org
section 1: building an AI-ready organization

How are you approaching the GenAI opportunity (choose all that apply)?
We have set up a
53%
center of excellence

We have identified families of use

cases and specific use cases within 59%
each family
We are working with partners /
technology firms to work on 45%
different use cases
We have different functions and
departments experimenting with 53%
GenAI across the organization

TM Forum, 2023

Dutch telco Vodafone Ziggo, for which the entire organization gains quickly the challenge is putting in
example, has an AI center of access to the LLM it is developing. place governance that remains
excellence which includes the The platform captures all the data relevant in the medium-to-long
company’s data science team. about how the LLM is being used, term and can be revised as new
Telefónica, meanwhile, has a global the applications and the inputs issues emerge.
centre of excellence within its and outputs, and this is then fed There are two types, or categories,
networks and IT group function back into the model to help train it of governance programs:
that is responsible for data and to become more accurate.
AI architectures, for the migration An external governance
The process of bringing program that is designed
towards a common data model, individuals and functions together
and for researching AI to protect individuals
to help drive better use of AI is and organizations outside
technologies and solutions. And having the added benefit of
Middle East telecoms group e& has the company
helping to break down divisional
a federated center of excellence siloes. It has the potential to help An internal governance
with each of the key departments empower the IT function and program designed to protect
/ functions having a representative. demonstrate its value as a driver employees and to ensure
An AI governance function sits on for business improvement by that AI is deployed
top of this with a mission to ensure interpreting and showcasing data successfully and sustainably
that use cases are reused across garnered from across the across the business.
different divisions. organization, particularly when Governance programs to protect
AI as a platform function. Some data from one department or people and organizations outside
operators have built – or are function is blended with data from the company tend to be a
building – platforms designed to another. In our survey, 53% of function of government laws
make AI accessible to different respondents said they have and frameworks. For example, the
parts of the organization. In the different functions and European Union passed an AI Act
case of GenAI, for example, a departments experimenting with in December 2023, to come into
platform may sit above the one, or GenAI across the organization effect in 2025 (see box on p.12).
many, LLMs that the CSP is using (see chart). And in the US, President Biden
to ensure that the best one – in AI governance imperatives issued an Executive Order on AI in
terms of cost, accuracy and so on October 2023. Both initiatives
– is used for a specific requirement. Many of the governance focus on protecting individuals,
For example, Vodafone has an AI requirements for AI sit within an societies and businesses from
platform which also provides existing data governance program. potentially harmful applications of
self-service tools and educational However, additional AI-specific AI, but the US Executive Order
materials for different teams to guardrails are needed to ensure also explores approaches required
build their own use cases. that AI tools and systems are, and to ensure that the country
remain, safe and ethical. With leads the development and
SK Telecom has what it calls an GenAI technology developing so
Intelligence Platform through exploitation of AI.

10 inform.tmforum.org
section 1: building an AI-ready organization

What are the main risks for your organization of using GenAI?
Shadow AI (i.e. non-authorized 52%
employee usage) 48%

Privacy and security (e.g. leaks 80%

of sensitive data via GenAI apps) 20%

Phishing and fraud (e.g. 64%

phishing emails or deep fakes) 36%

Lack of truth function (i.e. creating 74%

factually incorrrect answers) 26%

Misinformation and manipulation (e.g. 62%

creation of fake content, documents) 38%

Bias and fairness (because 66%

real-world data is often biased) 34%

Legal considerations 74%

(through use of public data) 26%

73%
Leaks of proprietary data
27%

Big risk Small risk

TM Forum, 2023

Our briefings found little evidence be advocates – individuals who operator Swisscom is
of concern that CSPs and most likely see governance as a experimenting with building its
organizations are failing to do hindrance and a frustration. If we own AI infrastructure and
enough, or to act quickly enough, had spoken solely to individuals developing internal expertise
to ensure they do not fall foul of responsible for security or which it can use to deliver new
current or future regulation. privacy we might have gained value and solutions in its IT
However, that may be a function a different impression. services business.
of the executives who briefed As the CDIO of one European The arrival of GenAI is also driving
us for the research. They tend to telco, in relation to AI governance, the requirement to improve
told us: “Sometimes I feel like they internal AI governance:
are slowing me down. I understand to help drive scale
it’s a good thing. But I feel like it’s
too much.” to help manage costs

Strict governance programs can to protect the organization

actually help CSPs to develop from the consequences of
using results that are not
Sometimes I feel technologies and capabilities
that they can monetize externally. accurate
like [governance is]
This is particularly the case to reduce the risk of technical
slowing me down. when countries have strict debt
I understand it’s a rules around data sovereignty.
to guard against the risk of
China Mobile reckons that the
good thing. But I feel LLM training model data
approaches it is taking to meet
like it’s too much.” being “corrupted”
AI laws may help it to develop
security technologies that it can to protect the organization
CDIO of a European operator provide to its own customers. against IP / copyright
infringement.
Switzerland is another country
with strict laws around data Our survey found that privacy and
sovereignty. There, incumbent security breaches represented the

11 inform.tmforum.org
section 1: building an AI-ready organization

main risk of using GenAI followed costs, the potential for technical
by “Lack of truth” (in other words, debt and a wasted opportunity to
accuracy), legal considerations and feed internal usage data back into
leaks of proprietary data (see chart the organization as training
on the previous page). materials to improve accuracy.
Managing the uncontrolled use of SK Telecom is a specific case,
Every development
AI tools represents a conundrum because it is building its own LLM team in my company
for CSPs and – in particular – for and needs to learn about usage and in SK Group has
the CIO function. Most CIOs across the organization in order to
recognise that different AI tools improve accuracy.
the opportunity to
and services are being procured, “Every development team in my
use our Intelligence
and deployed, across the company and in SK Group has Platform.”
company in an uncontrolled, the opportunity to use our
uncoordinated way. On the one Intelligence Platform,” says SK Chung Suk-guen, Chief Global
hand this meets a corporate goal Telecom’s Chung Suk-guen. AI Officer, SK Telecom
of using GenAI to drive innovation
and to experiment aggressively In the next section we look at how
and rapidly. But such a scenario operators are acquiring skills and
carries risks in terms of spiralling reskilling for an AI future.

Europe moves closer to setting AI governance rules

In March the European Parliament databases. Also forbidden are need to comply with EU copyright
approved the Artificial Intelligence motion recognition in the law and provide detailed
Act, designed to ensure safety workplace and schools, social summaries of the content used to
and compliance in the use of AI. In scoring, predictive policing, and AI train the model. Tougher
essence it sets out to ensure that that manipulates human behavior requirements will be imposed on
users can trust that the AI tools or exploits people’s vulnerabilities. models that pose a “systemic
they access are safe to use. The Obligations will also be risk”, expected to include
act now requires the formal introduced for other AI systems chatbots and image generators.
approval of ministers from EU deemed to be “high risk”. The EU says requirements will
member states, and other Examples include critical include “performing model
countries will be watching to see infrastructure, education and evaluations, assessing and
what happens in the EU before vocational training, employment, mitigating systemic risks, and
committing to their own acts. essential private and public reporting on incidents”.
The regulation establishes services – such as healthcare and Fines for non-compliance will
obligations for AI based on its banking – certain systems in law range from €7.5 million or 1.5% of
potential risks and level of impact. enforcement, migration and a company’s total worldwide
It bans systems that pose an border management, and justice turnover – whichever is higher –
“unacceptable risk”, but exempts and democratic processes (e.g. for giving incorrect information to
AI tools designed for military, influencing elections). Such regulators; to €15 million or 3% of
defence or national security use. systems must “assess and reduce worldwide turnover for breaching
Among the AI applications that risks, maintain use logs, be certain provisions of the act, such
threaten citizens’ rights and will be transparent and accurate, and as transparency obligations; to
banned, include biometric ensure human oversight”. €35 million or 7% of turnover, for
categorization systems based on When it comes to GenAI – deploying or developing banned
sensitive characteristics, and included in the category “General- AI tools. There will also be a new
untargeted scraping of facial purpose AI (GPAI) systems and European AI office that will set
images from the internet or CCTV the GPAI models they are based standards and be the main
footage to create facial recognition on” – all model developers will oversight body for GPAI models.

12 inform.tmforum.org
BENCHMARK

section 2:

putting the right

skills in place
In the search for AI skills, most CSPs are at a disadvantage compared to tech companies, particularly when
seeking to hire younger talent. Tech companies typically offer better salaries, faster career progression and a
corporate culture that is significantly more agile.

P
reliminary results from a TM results; full, updated results will be Deutsche Telekom, e&, Singtel and
Forum survey of CSPs, which published in an upcoming report. SK Telekom in the Global Telco AI
asks about skills and culture, Alliance, which was set up in July
show that skills in AI / machine Insourcing and outsourcing 2023 to develop telco-specific
learning, data analytics and GenAI LLMs. The initiative is now
automation are in high demand. Some CSPs and alliances are being turned into a joint venture in
Only cybersecurity scores higher developing their own AI platforms, which each of the partners will
than AI when it comes to both so they need AI architects, invest. China Mobile is also
demand for, and difficulty in engineers, developers and developing its own AI systems
finding the appropriate, skills. The researchers. For example, at including an LLM (see sidebar
graphics below and on the next Mobile World Congress in on p.16).
page are a sampling of the survey February Softbank joined

Demand for skills within CSP organizations

Cybersecurity 69% 27% 4%

AI/ML specialist 64% 16% 20%

IT infrastructure (cloud,
62% 31% 7%
virtualization, containerization)
Data scientist/analytics 59% 30% 11%

Software engineer 51% 40% 9%

Systems architect 45% 40% 14%

Network engineer 41% 46% 12%

Platform engineer 37% 47% 16%

Automation specialist 36% 50% 14%

‘Netware’ engineer (hybrid role combining
software and network engineer) 35% 56% 9%

High Medium Low

TM Forum, 2024

13 inform.tmforum.org
section 2: putting the right skills in place

CSPs’ difficulty finding skills

Cybersecurity 63% 27% 10%

Data scientist/analytics 59% 36% 5%

AI/ML specialist 59% 38% 3%

Software engineer 45% 40% 14%

‘Netware’ engineer (hybrid role combining
software and network engineer) 43% 48% 10%

IT infrastructure (cloud,
virtualization, containerization) 38% 43% 20%

Platform engineer 37% 41% 22%

Systems architect 26% 56% 18%

Automation specialist 21% 67% 13%

Network engineer 20% 55% 25%

Very difficult Somewhat difficult Not difficult

TM Forum, 2024

Also at MWC, Korea Telecom explains. “Our employee doesn’t

announced it will recruit up to become their employee, but [the
1,000 AI and digital experts this CSP] gives that employee a badge
year in its quest to become an and that employee works with
AICT – AI and ICT – company. The them as part of their team.”
operator says it will “completely [CSPs] are looking
change KT’s DNA to centre on AI for insourcing of Reskilling and upskilling
by strengthening internal
education and internalising AI”.
cloud and AI experts Most telcos are also reskilling and
at the architect level upskilling to get the AI talent they
Other operators like Vodafone are need. In fact, this approach is
partnering with hyperscalers for AI – so, at a high level.” becoming necessary for all skills
platforms, but they still need skills that are difficult to recruit, and it
in AI operations (AIOps) as well as Ashish Yadav, Senior Director,
can be more cost effective than
analytics, automation, cloud and Engineering, Capgemini
hiring new employees. US telco
platforms. Vodafone is insourcing Verizon, for example, is retraining
talent by hiring full-time many of its senior developers in AI
employees. In 2021, the company and machine learning.
announced its intention to hire
7,000 new software engineers. Deutsche Telekom, MTN Group
CSPs’ definitions of insourcing can and Orange all have reskilling and
But many CSPs are also turning to vary, Yadav adds. While working upskilling programs to develop AI
systems integrators for help with with a systems integrator might and cloud talent internally rather
staffing. “[CSPs] are looking for seem like outsourcing, many than trying to compete for scarce
insourcing of cloud and AI experts operators view it as a form of skills on the open market. MTN,
at the architect level – so, at a high insourcing because the consultants for example, is partnering with
level,” says Ashish Yadav, Senior become like part of the CSP family. hyperscalers to help employees
Director, Engineering, Capgemini. “When our expert goes and sits develop new skills.
“They can live with outsourcing at with the customer as part of their
the engineer level, but at the Microsoft is MTN’s primary cloud
team, the customer considers this partner, which gives the CSP
architect level it’s insourcing and more like insourcing, even if it’s
especially for these key areas of “extensive access” to the cloud
sort of outsourcing,” Yadav provider’s learning platforms,
cloud, AI, cybersecurity.”

14 inform.tmforum.org
section 2: putting the right skills in place

What do operators need to do to give themselves the best

chance of effectively exploiting AI and machine learning?
70%
Recruit new AI and
28%
machine learning skills
2%

60%
Train existing staff in a
39%
range of different AI skills
1%

39%
Acquire specialist
37%
AI technology firms
24%

51%
Partner with AI specialists 40%
9%

This will have a big impact This will have a moderate impact This will have a small impact
TM Forum, 2023

according to Ayedime Amadi, GenAI specifically is forcing CSPs “Some telcos are setting up
Group Senior Manager, Enterprise to develop AI expertise internally, internal certification and
Architecture & Customer Channels, according to McKinsey & university-led training programs
at MTN Group, who also leads TM Company. The consultancy points to ensure their teams have the
Forum’s TechCo Organizational to two unnamed telcos, one that right skills and capabilities to
Design project (TCOD). “We have a continued to outsource its AI innovate and execute with the
very deliberate structure to ensure talent and another large company technology,” the company adds.
that people can learn and upskill that created a dedicated team of “For instance, the large telco
through the cloud partnerships,” 10 AI data scientists and engineers. created a badging system to
she says. TCOD is a framework to “In the time the first telco took to identify gen-AI-ready employees
help organizations address draft requirements for outsourcing who have completed the
operations, people and culture in gen AI use-case development, the company’s sessions on use, risk,
their digital transformation projects second built and deployed four and effective prompting
using the Open Digital Architecture. gen AI solutions,” McKinsey states. techniques given by its AI, legal
In our CSP survey for this report GenAI requires new skills like and risk experts.”
we asked what operators need to “prompt engineering” (Google On the next page we look at
do to exploit AI and machine defines this as “the art of asking China Mobile’s Jiutian AI platform.
learning. Training existing the right question to get the best And in the next section we look at
employees across a range of AI output from an LLM”). Operators operators’ approaches to data
skills was considered to have a also need to hire data engineers and how they are taking steps to
significant impact by 60% of and subject matter experts “who get it ready for an AI future.
respondents while 70% said the understand what data to collect
same for recruiting people with and how, and who can oversee
new skills. Partnering with AI daily quality reviews as new forms
specialists was seen as having a of data are generated and
big impact by 51% of respondents consumed by these systems”,
(see chart above). according to McKinsey.

15 inform.tmforum.org
section 2: putting the right skills in place

China Mobile’s Jiutian AI platform requires new skills and training

China Mobile is made up of 31 The team that supports China operations, according to Deng.
local operators serving nearly a Mobile’s AI efforts is growing “And they can develop their own
billion mobile customers. The quickly. What started as a AI applications based on our
company, which also operates a 20-person team about seven platforms to enable their day-to-
hyperscale cloud platform, is years ago now has more than 600 day operation routine,” she adds.
transforming from a traditional people, and the aim is for it to China Mobile offers courses
CSP into an “innovative reach 1,000 employees by the end focusing on AI project management
information services provider, with of this year, says Deng. and for data scientists who collect
a focus on AI technology Setting up the team and retaining data and apply labels so that it can
innovation”, according to Dr. Lingli employees has been challenging be used as a source for AI
Deng, who works as a director- because AI skills are in demand applications. The company also
researcher in the company’s R&D everywhere. “That is why we set offers training in LLMs and for AI
group responsible for developing up a special organization and are application developers, a role that is
its Jiutian AI platform. offering competitive salaries to further divided into algorithm
Watch this video to find out the market,” Deng says. “AI is our developers, engineers and
more about how AI is central strategic direction, and we need application architects.
to China Mobile’s network the top talent in the nation to
automation strategy: achieve that goal.”
China Mobile provides a group-
wide learning platform for
reskilling and upskilling employees
to take on new roles in AI
development and maintenance, AI is our strategic
and it partners with top direction, and we
universities in China to develop AI
China Mobile unveiled Jiutian in skills. The average age of the need the top talent
2019 as a platform to support Jiutian team members is under 30. in the nation to
the company’s transition to Employees in China Mobile’s AI achieve that goal.”
autonomous networks. The group often work with teams at
operator is aiming for its networks local operating companies to help Dr. Lingli Deng, Director-
to be highly autonomous by 2025. them implement AI pilots and Researcher, China Mobile
The AI platform is also accessible scale deployments. This includes
to external developers via open providing training and onboarding
APIs. In October 2023, China them as members of R&D
Mobile announced that it has projects, so that they can acquire
also developed its own LLM as the skills they need for daily
part of Jiutian.

16 inform.tmforum.org
BENCHMARK

section 3:

getting the basics right –

is CSP data AI-ready?
A regular theme in our conversations with operators for this report was the state of readiness of data to use
for AI applications and, in the medium-to-long term, to deploy at scale. For many, there are still considerable
challenges in terms of putting coherent data strategies in place in order to exploit AI technology to the full.

T
he CIO from one large data available for AI and machine leveraging unstructured data, a
European telecoms group learning (see chart below). subject which is explored in detail
told us during research for The only area where, based on the in section 6. Rather than having to
this report: “I always tell our CEO results of the survey, operators go through a slow, difficult
it’s data before AI.” say they still come up short is in process of cleansing,
Yet our industry survey suggests the use of unstructured data. This consolidating and labelling the
that CSPs are relatively happy is a particularly relevant topic data, there may be scenarios
with the progress they have made when it comes to GenAI, which where unstructured data can be
in their organizations in making potentially offers a short-cut to put straight into an LLM.

How successful has your organization been in making

data available for AI and machine learning?
We are making good progress 65%
towards the creation of a unified 13%
data model 23%

We have effective governance in

63%
place which helps ensure that data 12%
can be accessed by different teams
26%
and functions across the organization

There is good general awareness of 57%

the data that sits across different 17%
siloes within the organization 26%

We have made a good job of 41%

using unstructured data 24%
alongside structured data 35%

Where needed we have systems in 62%

place that allow data to be accessed 15%
in real time / near real time 22%

I agree with this statement I disagree with this statement

I neither agree nor disagree with this statement
TM Forum, 2023

17 inform.tmforum.org
section 3: getting the basics right – is CSP data AI-ready?

Internal pools of knowledge

• Customer data that sits within
BSS/OSS systems
• Operational data relating to the
performance of the network and
IT systems
• Financial and operational data / KPIs
about the operator’s business
• Technical manuals used by
External pools of knowledge
engineers to address issues and
build new products and capabilities
• Code that has been written • Customer data from third-party sources such as social
• Data about suppliers, partners and media sites (which tends to be unstructured data)
the supply chain more broadly • Information from standards, technical bodies and academic
• Data about distributors and channels research journals used for building new networks, IT
• Information (from HR) about people systems and customer products and solutions
working in the organization • Information from consultants and analysts about the
• The content of emails and messaging operator’s own business or the wider industry
platforms between employees, • Insights and data about a) other operators and service
customers, partners and suppliers. providers b) technology suppliers and partners c)
customers and potential customers and d) other companies
against whom CSPs might want to benchmark themselves
• Code from repositories such as GitHub

However, our analysis is that those business, from networks to if looking to implement GenAI
operators which are still at the service delivery and customer or any other type of machine
early stages of experimenting experience. Getting this part learning models.
with AI may be underestimating right is critical for all processes With telecoms, pools of
what is required – in terms of data around AI-based, data-driven knowledge exist both inside the
– to deploy AI at scale and with execution. operator organization and outside.
success. Those CSPs which think Lack of a common data
The graphic above defines some
they are in a relatively good place model. CSPs use data from of those aspects.
when it comes to readying their systems developed by
data for scale deployments of A data model is a design for
different vendors over many how to structure and represent
analytics and AI may come to a years. Without a common data
different view when they actually information so that it can be used
model, bringing together this consistently across disparate
start to build and run models. structured and unstructured systems. The problem is that most
data is a hugely time- CSPs work with tens or hundreds
Challenges to AI-ready data consuming process. Indeed, of vendors whose systems
CSPs face some specific many data science teams generate and/or consume data, all
challenges in getting data AI- spend more time cleaning up using different data models.
ready – all of which potentially data than they do analyzing it
lead to distrust of the accuracy of to provide actionable insights.
Telco data sources
data-led insights and actions. Lack of data context. Not fully

They include: knowing how, when or where CSP approaches to becoming
data was captured, and when data-driven organizations have
Lack of clean data. CSPs we
typically centered around
spoke to for this analysis it is useful or not, creates
reluctance to make data- operational function. They have
reported a lack of clean, organized their IT operations
consistent and useable data driven decisions. This is a
particularly significant around linear software processes
which can be applied across managed by separate teams:
different parts of the CSP obstacle for CSPs to overcome

18 inform.tmforum.org
section 3: getting the basics right – is CSP data AI-ready?

fulfillment, service activation,

assurance and so on. Each domain
Examples of CSP data types
owner has its own suite of data
tools, controls and processes. This Structured data Semi-structured data Unstructured data
divided responsibility and tooling
loosely organized data
often leads to the emergence of predefined data models no predefined data models
easy to search meta-level structure that difficult to search
data silos, data duplication and a could contain unstructured
lack of reusability of data across text-based data: HTML, XML, JSON text, PDF, images, video

the organization.
Stored in:
As the perceived value and usage
of data has increased, operators rows/columns, as: abstracts and figures, as: various forms, as:
have stepped up their investment relational databases relational databases applications
in data analytics. Telecoms and IT data warehouses tagged text format data warehouses/lakes
research firm Omdia estimates
that CSPs’ global spend on data Examples:
analytics services will increase
documents
from $2.3 billion in 2022 to $2.5 dates
e-mails and messages
phone numbers server logs
billion in 2025. But in reality, the transcripts
social security numbers e-mail sorted by folder
time and investment in collecting, image files
customer names tweets with hashtags
cleaning, transforming and storing open-ended survey
transactional information
data in the right format is often responses
disproportionately higher than
TM Forum, 2024
the time spent using that data.
However, the advent of AI-driven
unstructured data that sits to service orchestration, and
analytics and initiatives to tackle
within their organizations. This service assurance to billing. This
operational efficiency, drive service
needs to be tagged and cleaned consolidation makes it easier for
innovation and enhance customer
before feeding into LLMs. everyone involved in the AI /
experience, has accelerated the
Many CSPs still lack a cohesive machine learning lifecycle — from
need to leverage data in a more
strategy that enables data to flow data preparation, to model
transformative way:
horizontally across the building to application
AI / machine learning requires deployment — to share data and
organization in a single data
vast volumes of data to train collaborate effectively to derive
model. Essentially, that would
models valuable insights with speed.
mean acting as a single entity, and
Ensuring unbiased AI results not functionally separate in terms But bringing network, operations
requires diverse data sets and of where data is processed and and customer-centric data
a wide range of data types resides in the organization. together into a “single source of
Adding layers of data can truth” is a significant challenge,
improve accuracy of models Consolidating data not simply due to the rapidly
and the impact of applications rising volumes of data but also
Breaking down these operational because of the widely varying
Models must be continuously silos requires a holistic approach to formats in which it is created and
trained with the latest data, which plays out across stored (see graphic above).
information to maintain systems and processes. But it also
predictive performance, requires a cultural shift. CSPs Data may arrive in real time, so
particularly in dynamic already have initiatives underway must be harnessed for real-time
environments to move toward this, but there is predictions in use cases as diverse
not always a board-level strategic as fraud detection or product
Data must be available in recommendations, and AI /
drive to bring data capabilities
real time for critical business machine learning–powered
together, so progress is often slow.
functions, principally in the applications must be able to
network CSPs need to consolidate multiple handle the extra load as use cases
data types and structures from proliferate. If an application cannot
The advent of GenAI has given
many sources into a single scale, performance bottlenecks will
CSPs the opportunity to
repository, across ordering to diminish the value of using AI /
leverage the vast amounts of
fulfillment, network optimization machine learning. And to have an

19 inform.tmforum.org
section 3: getting the basics right – is CSP data AI-ready?

AI application that is relevant,

accurate and scalable, CSPs need Setting the right foundations to fully leverage data for AI
to make sure that their data is of
high quality and easy to access CSPs need to get essential building different departments and lines of
and share securely within the blocks in place to fully exploit business, all using a unified set of
organization and with partners. the data residing within their rules and a common approach to
organizations for AI applications. governance; or taking a flexible
A new data architecture approach by moving experts
Data compliance around and (de)centralizing
A data-driven architecture is key to based on changing requirements.
maintaining precision and There are extensive reputational
consistency across domains within and financial ramifications to
the CSP, using a common data violating data regulation – such Data skills
model to ensure that data flows as General Data Protection Another consideration is putting
smoothly through all systems and Regulation (GDPR) in Europe. in place the necessary data-
accurately feeds all automated Given that sensitive customer centric skill sets, often for very
workflows. Many CSPs have made data forms the backbone of specific disciplines. For example,
progress on this kind of automation CSPs’ core operations, they must some tools are geared towards
– indeed, TM Forum’s Open APIs ensure full compliance while business intelligence and not
and Open Digital Architecture are interacting with data. This could machine learning. CSPs need to
central to helping with this need – entail a full re-examination of be able to build teams that can
but end-to-end automation has how they collect, store and use new AI-based tools to
remained elusive because of use their customer data as leverage data at scale and
interoperability issues, architectural well as reinforcing diligence generate insights, whether to
fragmentation and customization. around compliance. drive operational and energy
efficiency, or enhance services.
A common, industry-agreed data
model would allow CSPs to buy Data culture Read our report to find out
products and services from The data strategy drives cultural more about exploiting data
different vendors knowing that change within the CSP and to drive future strategies:
they could re-use data across relates to the structure of the February 2024

systems, and it would help them data science capability. REPORT

partner and develop new services Approaches vary within CSPs: a numbers game:
more quickly. As part of TM from building a central team exploiting data to
drive future
Forum’s Open API and AI, Data & (likely in IT) under a chief digital strategies
Analytics Project collaboration officer; taking a decentralized
Author:
Richard Webb, Senior Analyst

Editor:
Sponsored by:

work, operators and suppliers are

Ian Kemp, Managing Editor

approach with data experts in

developing common data models.
CSPs urgently need to modernize
their data architectures and they they need a reference architecture meta-ecosystems. It also sets
want to play a key role in shaping that looks at data beyond SID out to tackle the question of
them – something of a shift, with compliance and runs not only on what a modern data architecture
vendors typically having heavily telecoms networks but also should look like for AI-enabled
influenced telco data architecture effectively enables them to telecoms operations.
modeling previously. The TM manage new business models. The Modern Data Architecture
Forum Information Framework In December 2023, the Forum for Telecom Operations Project
(SID) provides an information / launched a new data reference was launched in collaboration
data reference model along with a architecture for the telecoms with eighteen CSPs, including
common vocabulary for industry that aims to encompass Axiata, China Mobile, MTN,
implementing business processes. both new AI-enabled business Orange, Telefónica, Telenor,
It provides the basis for a models and network operations. Vodafone and Zain, in addition to
common data model and a The new project reflects network suppliers and systems integrators
common data dictionary and operators’ need to build AI-driven Accenture, Cloudera, Ericsson,
offers a foundation for function, autonomous processes, put data Huawei and Infosys.
application, component and API and AI at the core of operations, TM Forum Members can view or
development. But CSPs are saying and bring external data into their join the new project here.

20 inform.tmforum.org
section 3: getting the basics right – is CSP data AI-ready?

Drowning in data lakes

Evolving from data lakes to a data fabric
The process of creating data lakes
and data warehouses has taken Data analytics Data analytics Data analytics Data analytics Data analytics
place over many years, but it has
not given CSPs the capabilities to Data lake Data lake Data lake

deploy AI at scale across the Data lake Data lake Data lake
organization. The problem is that
data sits in different on-premises, Data, much of it sitting in BSS/OSS systems and in siloes across the organization

private and hybrid cloud

environments and is stored in
proprietary systems and formats.
That inevitably leads to duplication
GenAI wrapper
of effort because teams have
different approaches to data Data analytics Predictive AI GenAI

models and each one requires

bespoke massaging of the data to Data fabric
perform analytics and run AI
Data platform (public cloud)
models. To get an AI model into
production could take as long as
six or seven months, according to Data, much of it sitting in BSS/OSS systems and in siloes across the organization
the CIO of one south-east Asian
telecoms group we spoke to.
In addition to data fragmentation,
many CSPs struggle with a culture
GenAI wrapper
of data usage and data sharing
that makes it impossible for a Data analytics Predictive AI GenAI
single source of truth to exist.
“Right now we have this culture Data fabric
where we say ‘give us the data’, or
Data platform (public cloud)
‘give me a copy or dump of data’,”
said the SVP of data analytics and
AI at a large central European Data, much of it sitting in Unstructured data (customer/partner
BSS/OSS systems and in siloes interactions, internal knowledge,
telecoms group. “In the future we across the organization emails, messages etc.)
need to say no to this and we
need to have a single golden The graphic shows the progression from data analytics in an age of fragmented data sources sitting across
source of data.” multiple data lakes and warehouses, across on-premises, private and hybrid cloud environments and stored in
proprietary vendor systems and formats.
Building a single source of data is In the next stage operators move data into a single platform in a public cloud environment used for data analytics,
predictive AI and GenAI. GenAI is both a vertical pillar alongside analytics and predictive AI, and a wrapper around
an extremely complex task that all the three pillars, providing access to the different outputs.
has been beyond the capabilities The third stage demonstrates the potential for adding to the available pool of data within the organization by
of most CSPs until now, because accessing unstructured information and knowledge that sits within the business, its employees, its customers and
its partners.
of the fragmentation of the data.
TM Forum, 2024
But operators such as BT,
Deutsche Telekom and Telefónica
have taken steps to address this also known as ghost or orphaned important when it comes to
by moving all their data into the identities, when the same person identifying, and addressing,
public cloud. BT, for example, has appears as two separate people. potential privacy and security
moved more than 90% of its data To overcome such difficulties BT concerns around the data.
into Google Cloud Platform in the has built a “data fabric”. This allows Harmeen Mehta, Chief Digital and
past two years. the source systems data, and the Information Officer at BT, described
But this does not, in itself, create a permissions associated with them, the data fabric, in a briefing with us,
clean set of data. There are many to be brought into the public cloud as “one of the biggest things that
inconsistencies and inaccuracies environment, thereby allowing we have invested in” and what
in the data that is pulled together inconsistencies and inaccuracies to “makes us really unique from any
– for example, duplicate identities, be addressed. This is particularly other company”.

21 inform.tmforum.org
section 3: getting the basics right – is CSP data AI-ready?

In February, BT Group’s Digital for TM Forum members which In other words, they operate
unit became the 10th CSP to have set up projects to assess within their own spoke.
achieve Running on ODA status. how to make data available in real “You work in your spoke, but the
Running on ODA recognises that time and create a standardized source remains the same,” says
BT is now delivering products and way of implementing intent-based Omair Ahmed Khan, SVP, Data
services from reusable automation. Analytics / AI projects &
components, underpinned by Deutsche Telekom, meanwhile, Governance, Deutsche Telekom.
cloud-native and vendor-agnostic describes its new data “You can run your AI models on
solutions. BT’s Digital unit was set management strategy – which is this inference and you get your
up to simplify the operator’s broadly comparable to BT’s – as a outcome. But you don’t get a
architecture and establish a “hub and spoke” approach. copy of the data.”
data-first, AI-ready infrastructure. Anyone in the business can use In the next section we take a look
Indeed, CSPs need to infuse AI the data in the hub, with their own at what operators are building and
into their future (open) AI tools, but they are not allowed buying as they put the foundations
architectures. This is a major focus to get their own copy of the data. for their AI strategies in place.

Creating standards through an AI-enabled architecture

TM Forum is currently working huge amounts of data in real AI technical debt and the
with members towards a fully time, which is facilitated by complexity and customization
AI-enabled Open Digital data streaming architectures, that’s so expensive to maintain
Architecture (ODA). In an as well as the use of in legacy systems.”
interview for Inform following the knowledge graphs. Another significant activity within
Forum’s annual Accelerate event, The challenge of how to create a the TM Forum collaboration project
Andy Tiller, Executive Vice data architecture which is fully work is exploration of GenAI
President, Products and Services, AI-enabled and make that part of applications. For example, the TM
outlined some of those plans. the ODA is a major new project, Forum AN Project team is working
The Forum has been working on says Tiller. It includes how to share to explore how GenAI affects
AI architecture, frameworks, data across multiple parties within autonomous networks. To date the
governance and standards for privacy constraints and regulation. drive has been to use predictive AI
about five years and has already The second key area being models and machine learning to
embedded AI capabilities into addressed is how you enable self-healing and
ODA such as extensions to the implement the reference optimization of networks, says
e-TOM Business Process architectures for intent-based Tiller, who notes that GenAI is good
Framework and Open APIs for closed loop automation in a at generating content but not so
carrying intent, as well as standardized way. good at making predictions and
reference architectures, maturity decisions. But, he adds, “together
models and governance best “I heard one of our members there is complementarity”, so
practices. Now planning work is describe the way AI is being used the team has found use cases
being carried out on how to fully as a lot of artisanal applications where GenAI can be used to great
AI-enable ODA. hand crafted and bolted on,” effect to help with automation of
says Tiller. “That’s great for now network operations.
Tiller outlines two key areas that because we need a lot of AI
are being addressed: innovation and there are lots of Watch the video:
The first is the data very good use cases. But we need
architecture. New uses of AI to start thinking about how we do
require data to be made it in a standardized way.
available in real time and “We want to create standardized
structured differently from the ways of implementing intent-
relational database tables based automation using ODA
used in traditional operational Components, Canvas and Open
data stores and data lakes. APIs. The aim is to avoid future
This includes tapping into

22 inform.tmforum.org
BENCHMARK

section 4:

Architecting, building,
buying and scaling AI
Many parallels can be drawn between AI / GenAI and the advent of public cloud: it is transformational
technology, it is dominated by hyperscale service providers, and operators need to work out precisely how and
where to deploy it. But there is a marked difference when it comes to approaching the question of what to build
and what to buy in relation to IT systems.

G
oing back 10 years or so, As operators embark, or continue, knowing where technical debt
many operators built on their AI journeys, their may accumulate can be difficult.
their own public cloud in approaches will be shaped by the As new technologies, systems,
an effort to compete with extent to which they embed approaches and solutions emerge
hyperscalers. Most have now exited openness and composability as in AI, some of the incompatibility
the business, although there are still core principles. Deutsche Telekom’s issues that might lead to technical
some operators – such as Deutsche Omair Ahmed Khan says that most debt will necessarily be addressed
Telekom, Orange and Proximus – of his company’s AI projects and solved.
offering sovereign cloud services. involve a combination of building Never has there been a time
CSPs have come to accept that and buying different elements. when CSPs need to be more alert
they cannot compete with “We have a combination of a build to the potential of a new set of
hyperscalers, except in specific and buy strategy, and the buy technologies while, at the same
circumstances where their part has never involved our time, arming themselves with
national identity lends them an buying a full-fledged turnkey the knowledge and the skills to
advantage. Whether CSPs can solution,” he says. “We would not, make sound decisions that allow
rival hyperscalers in developing for example, turn to a consultancy them to sustainably benefit from
new technologies is an interesting firm and say: ‘here’s the business new capabilities.
discussion and one which is problem, fix it’.”
particularly pertinent to GenAI As CSPs launch themselves into AI architectures
and LLMs. We discuss this the AI era it is essential that they
specifically in section 6. CSPs are still at the very early
remain true to their principles. The stages of trying to figure out how
At the same time, however, there risk is that, faced with pressure AI fits into their enterprise or
is a clear shift to CSPs developing from business leaders to deploy reference architectures. Indeed,
their own in-house technology solutions and to show results the feedback from the majority of
capabilities as they transition to quickly, they will set out on a path CSPs we spoke to is that it is too
open, composable architectures. that will lead to their accumulating soon to start creating a reference
This is a response to the technical technical debt in the future. architecture. There are a number
debt they have accumulated over At the same time, however, of factors at play:
20 years or more by investing in aggressive experimentation and
systems that cannot be easily AI is moving too quickly to
the creation of minimum viable build it into a reference
swapped out, reconfigured or products (MVPs) is, undeniably,
upgraded without costly architecture
a good thing. Furthermore,
customization.

23 inform.tmforum.org
section 4: Architecting, building, buying and scaling AI

CSPs need to make more deliver specific benefits in terms of In the early days of software
progress in terms of efficiencies, customer experience development there was a heavy
simplifying their existing IT improvements or new revenues. reliance on manual processes. The
estates and shutting down These are explored more fully in software lifecycle that developed
old systems before they can the next section of the report. brought in tools and capabilities
consider where AI fits Even if CSPs did have a clear vision including source code repositories
CSPs need to decide what and strategy for embedding AI and DevOps, where continuous
they want their new operating into their future enterprise integration / continuous
model to be before they architecture, most would likely deployment became possible. This
consider where AI sits in their struggle to implement it because all happened at different points in
enterprise architectures they lack the requisite skills, tools time and this is how software
and capabilities to deliver effective factories work today. Any
More work is needed to organization that creates software
improve the accuracy and results and a demonstrable return
on investment. Readying the at scale today runs a well-oiled
value of AI, where to find software development cycle.
insights and how to deliver required data in an easy-to-ingest
better outcomes before it can fashion is one of the biggest AI, on the other hand, has been
become a central pillar of an inhibitors. As we discussed in done in a bespoke way. There are
enterprise architecture section 3, the fragmentation of many different processes and
data – in terms of data models and stages that sit between identifying
CSPs lack the skills to embed data lakes – held and managed by the relevant data to the delivery of
AI into their enterprise different functions leads to the insight. To deploy AI at scale
architectures (see section 3). duplication of effort and a slower this process must be industrialized.
So if it is too early to start overall process. CSPs need machine learrning ops
embedding AI into a future Another inhibitor relates to the like software teams had DevOps.
enterprise architecture, what does processes that CSPs need to put in This involves the creation of data
that mean in terms of today’s place to go beyond what Verizon pipelines coming into a data
deployments? In practice, the Group CIO, Shankar Arumugavelu, ingestion platform where data can
focus of AI deployments to date describes as the “artisanal” nature be curated and enriched. Data
has been “bolt-on” use cases – of today’s telco AI. He likens modellers or data scientists can
opportunistic usage of AI-enabled today’s AI with the early days then do the feature engineering,
services and applications that of software development. and then once it is deployed they

When we asked operators whether they are building AI into their enterprise / reference architectures
it was clear that they are still, at best, at the early stages of formulating their strategies.

Chief Digital Officer, western European telco Chief Digital Officer, Scandinavian telecoms operator
“We do not currently have any plans to embed AI within “We are just in the process of setting our AI strategy for
our enterprise architecture. This is not because I think it’s the next two years. But there is no way we can say that
not right, but because our enterprise architecture is very AI will be part of the enterprise architecture by 2026. It
much telco focused and the type of skills are different. just moves too quickly.”
If I had the right people, yes, I would do it.”

Chief Architect, Australian telecoms operator Chief Architect, Scandinavian telecoms operator
“We are working on a new release [of our enterprise “AI is going to fundamentally change how we operate. It
architecture] which will have the beginnings of some AI will become an integral part of the operation. And that
principles…We want to make sure that, with no regrets, operating model will drive our new architecture.”
we get some of the foundations well and right…One
area we are currently reviewing is the use of knowledge
graphs and trying to balance out how human code
knowledge will work with machine learning knowledge.”

24 inform.tmforum.org
section 4: Architecting, building, buying and scaling AI

because of macroeconomic
factors. Using the AI factory, it was
able to build five new churn Softbank: The Japanese tech
prediction models in just three group has also partnered with
months saving the business Nvidia to build data centers
Two years ago the significant amounts of revenue. designed to host GenAI and
production cost of wireless applications on a multi-
“This would have taken a year
tenant common server platform.
AI was super-high. before we had the AI factory,”
The new data centers will handle
says the CIO.
It took six to eight both AI and 5G workloads, with
months to produce Softbank exploring 5G applications
The AI technology stack for AI factories, autonomous
an AI model. Now When you consider the overall AI driving, augmented and virtual
it takes a matter technology stack it’s useful to reality, computer vision and
of days.” look at where CSPs are building digital twins.
out solutions and where they are
CIO, south-east Asian buying from third parties.
telecoms group A handful of operators worldwide
are investing in building their own SK Telecom: South Korean telco
AI (computing) infrastructure. SK Telecom is pursuing AI-driven
Each of them is doing so as part of demand for data centers as part
can fix any issues with data drift its strategy to develop products of its wider AI ambitions. CFO
– the process whereby the and services for the B2B market. Yang-Seob Kim says the operator
predictive power of a model gets plans to “drive our data center
Asia is a particular hotbed for AI
worse over time – and perform business more strongly, focused
infrastructure investment as we have
any necessary fine-tuning. on next-generation AI data
outlined in stories on our Inform
centers and global expansion”.
This process of industrialization website. Developments include:
is made possible as operators
move their data into public clouds
and data becomes accessible in
NTT: Japan’s NTT is setting up a
real time. China Mobile: The Chinese
new company with Tepco Power
The CIO of one south-east Asian operator has bought hardware,
Grid to jointly develop and operate
telecoms group, describes the and built its own data centers
data centers in Greater Tokyo. NTT
process that his organization has containing graphics processing
Global Data Centers is investing
put in place to industrialize AI as a units (GPUs) and accelerators as
¥1.5 trillion (about $12 billion) over
“data factory”. This, he says, has part of its Jiutian LLM project. It is
the next five years to expand and
massively reduced the time and also training the data, and
upgrade its data center business
the production cost of AI. building LLMs, from a number of
globally, citing a worldwide
other different vertical markets.
“Two years ago the production opportunity to meet demand for
cost of AI was super-high,” says data usage linked to rising use of
the CIO. “It took six to eight GenAI, among other technologies.
months to produce an AI model. However, these operators are the
Jio: Reliance industries, the parent
Now it takes a matter of days. You exceptions rather than the rule.
company of Indian telco Jio, has
can run the entire cycle much The trend over the past ten years
partnered with Nvidia to build
faster and with less people.” has been for CSPs to quit the data
supercomputing infrastructure for
By industrializing the creation of AI center business. The exception is
AI. Reliance aims to provide AI
models, CSPs are in a much better those operators in countries
infrastructure to scientists,
position to use AI as challenges with strict privacy and security laws
developers and start-ups across
and opportunities present that create a demand for national
India as well as create AI
themselves in the business. The public cloud services. In
applications and services for its
south-east Asian CIO cites the Switzerland, for example, Swisscom
450 million Jio customers.
example of one operator whose is partnering with Nvidia to build
churn prediction model failed GenAI full-stack supercomputers.

25 inform.tmforum.org
section 4: Architecting, building, buying and scaling AI

Swisscom’s motivation is the comparisons between running use more broadly, there has been a
potential to build expertise in GenAI cases in a private or public cloud more recent spike in spend with
– and potentially products and – operators will favor public cloud Microsoft which has added GenAI-
solutions – for the B2B market. for experimentation and for driven services to its suite of
building MVPs. Azure services.
Public or private cloud? What’s more, many CSPs will likely However, the dependency on such
Deciding where to deploy AI is choose to buy LLMs from other a small number of hyperscalers
not a clear-cut choice for CSPs. specialists, including from – all of which inevitably deploy
Public cloud provides substantial alliances such as the Global Telco resources in larger countries with
computational resources, energy AI Alliance. the biggest markets for their
and specialized hardware required services – potentially places CSPs
for processing complex Hyperscaler relationships in smaller, less developed
algorithms and large amounts of countries at a disadvantage. This
As CSPs increase the use of AI is particularly true when it comes
data. However, hyperscalers must across their organizations it
address non-technical issues to securing the resources and
inevitably leads to a deepening of collaboration needed to work on
related to cost, economics, ties with the major hyperscalers
sovereignty and data privacy. large deployments.
– Amazon Web Services, Microsoft
Transit costs, for instance, may Azure and Google Cloud Platform “Mobilizing and committing to
become an issue if a CSP solely – and a significant increase in getting things done is very
uses a public cloud to process spend on their services. different [from just talking about
large volumes of data. it],” says the CIO of one African
“The biggest investment we have telecoms operator. “Our
It is difficult to make a case today made in AI so far has been in
for building private cloud for hyperscaler just doesn’t have the
Microsoft copilot,” says the chief scale to get things meaningfully
GenAI unless the operator is architect of one Asian CSP.
building its own LLM – which is done in a reasonable timeframe.
the case with China Mobile, “Our spend on Microsoft has We had a lot of things that didn’t
Softbank and SK Telecom in Asia, doubled,” says the Chief Digital work [in] the first three months.”
for instance, and Deutsche Officer of another CSP from In the next section we look at
Telekom in Europe. No operators northern Europe. some of the specific areas in
are deploying GenAI at scale While spend with all hyperscalers which CSPs are deploying AI
today, and until they do so – and it has increased as a result of the technology, ranging from
becomes possible to make growing usage of analytics and AI networks to customer experience.

Vodafone opts for partner approach after early open source building strategy

Vodafone’s early investments in its energies on building out use Vodafone is also partnering with
analytics and AI involved heavy cases rather than managing other industry big-hitters to
investments in open source and the platform. develop its AI strategy. In January,
building its own big data Vodafone builds most of its AI use the operator announced that it
platforms. But it became clear cases in-house, rather than would invest $1.5 billion over the
after two or three years that it engaging, for example, with a next 10 years in cloud and
had invested so heavily in professional services organization. customer-focused AI services
underlying platforms and These use cases are built on developed in conjunction with
technology that it could not scale tooling provided by Google on its Microsoft. The deal includes
and deliver real-life use cases. Vertex AI platform – a capability replacing multiple physical data
Furthermore, it was not keeping that lets organizations like centers with virtual ones across
pace, in terms of capabilities, with Vodafone train and deploy AI Europe. Another key area of
hyperscaler service providers. models and applications. It collaboration will see the companies
The operator then pivoted and combines data engineering, data applying Microsoft Azure OpenAI
struck a large strategic deal science and machine learning to deliver real-time, hyper-
with Google Cloud Platform. engineering workflows. personalized experiences across all
This allowed Vodafone to focus Vodafone customer touchpoints.

26 inform.tmforum.org
BENCHMARK

section 5:

Where operators
are applying AI
Despite the fact that CSPs have been working to build AI use cases over the past half a dozen years, most have
had limited success. There are many reasons for this, but foremost among them – as we have detailed in earlier
sections – are that they have been unable yet to make data AI-ready, they have lacked the skills and the tools to
build an “AI factory”, and most operators do not have the requisite cloud and software skills and cultures.

O
perators’ use of AI / GenAI There is a case for adding a fourth where the best opportunities lie for
today – and in future – category – AI for new products, AI and the fact that it is, perhaps,
currently can be distilled services and revenues – but there too early to make any bold
into three broad categories: are relatively few operators assertions about the value of
Employee efficiency, focused on this category yet. AI in new product development.
productivity and In our CSP survey we asked our Operators have compiled lists of
experience (internal) respondents where AI and machine hundreds of use cases for both
The network and network learning had the most potential to predictive AI and, more recently,
operations (internal) positively impact the telecoms GenAI, but most of these are still
business. Results were consistent being evaluated. In some cases
Marketing and customer with our overall assessment of CSPs have built MVPs and in
experience (external).

Where does AI / machine learning have the most

potential to positively impact the telecoms business?

Customer experience / customer 87%

relationship management 13%

85%
Operations / network operations 14%
1%

43%
New product development 46%
11%

High potential Medium potential Low potential

TM Forum, 2023

27 inform.tmforum.org
section 5: Where operators are applying AI

others they have gone into full in the short-to-medium term is in our survey believe there will be a
production. But this does not mean helping employees to do their growing demand for new roles
they have necessarily yielded jobs more efficiently and, both in the medium and long term
positive results. The artisanal arguably, more enjoyably. (see charts below and on p.29).
nature of AI deployments to date This will only change when They also expect that more jobs
means that many have been slow CSPs are persuaded that AI is will be created than will be lost as
and expensive to deploy. sufficiently robust, trustworthy a result of AI.
What follows in this section is and ethically acceptable to allow
an overview of where CSPs are it to operate the network, make Harnessing GenAI
applying AI in their strategies business decisions and interact Unlike other forms of AI, GenAI can
within the broad categories directly with customers. If AI and be of value to employees because it
outlined above. We have covered GenAI fulfil their potential, such can be applied to the basic tools of
many of the specific use cases outcomes may be possible in the productivity and communications in
in other reports, as indicated medium term. But it is not, every role. Furthermore, by making
throughout, which can be read currently, informing CSP strategy. GenAI copilot services available on
for further analysis. A more plausible scenario is that existing platforms and products,
AI will change jobs and result in a hyperscaler service providers are
requirement for new ones. For making it possible to extend AI tools
example, call center agents will be to all employees quickly and simply.
expected to spend less time Copilot services can be applied
Employee efficiency, understanding customers and to a range of different roles
productivity and experience interpreting their requirements – and functions within the CSP
AI will help them to do this more organization. And they can
BT hit the headlines last year
quickly – and more time proposing help with a series of tasks,
when it announced plans to cut
solutions and, potentially, cross- such as coding, writing and
55,000 jobs by 2030 – more than
selling and up-selling. Network generating images.
40% of its global workforce – with
engineers will spend less time
10,000 reportedly to be replaced Whereas traditional forms of AI
looking for data that explains why
by AI. But as we carried out have tended to be deployed in
the network has behaved in a
research for this report it became those parts of the business most
certain way and more time putting
clear from our conversations with comfortable with using new
measures in place to help ensure
senior executives that CSPs in technologies, some of the biggest
the problem does not recur.
general are not yet close to being potential for GenAI is within
able to use AI-enabled solutions The requirement for new roles is a functions that are least
to replace human functions. function of the extent to which technologically advanced but
Rather, the value of AI today and CSPs seek to build, rather than where there is a huge amount of
buy, AI use cases. Respondents to

How do you expect AI to impact the types and number of jobs in your
organization in the short-to-medium term (choose all that apply)?
It will result in a relatively small
number of new roles to leverage 65%
and manage AI

It will result in a large number of new

32%
roles to leverage and manage AI

It will result in a relatively small

number of job losses as a result 46%
of AI-driven automation
It will result in a large number
of job losses as a result of 8%
AI-driven automation

TM Forum, 2023

28 inform.tmforum.org
section 5: Where operators are applying AI

How do you expect AI to impact the number of jobs in your

organization in the medium-to-long term (choose all that apply)?
It will result in a relatively small
number of new roles to leverage 44%
and manage AI

It will result in a large number of new

47%
roles to leverage and manage AI

It will result in a relatively small

number of job losses as a result 31%
of AI-driven automation
It will result in a large number
of job losses as a result of 33%
AI-driven automation

TM Forum, 2023

knowledge and text-based Energy saving is another AI-

content. HR departments retain driven use case for networks that
huge volumes of information many CSPs have taken into
about employees, for example, AI in the network, network production. Savings, for example,
while finance and legal functions operations and OSS are achieved when operators gain
process contracts with customers, insights that inform them when
There are two main objectives of
suppliers and partners. GenAI can and where they can shut down
using AI in the network, network
deliver huge efficiencies across capacity on their networks.
operations and OSS:
such functions, although strong AI Indeed AI is being used
governance must be applied to Reducing cost and complexity
extensively by operators to inform
ensure that ethical, security and Delivering new network their sustainability strategies in
privacy issues are observed. experiences that leverage areas such as measuring energy
GenAI for coding has been held investment in fiber and 5G. consumption and implementing
up as an important use case, Reducing cost and complexity energy management information
although the operators we spoke involves using AI to reduce both systems. Read our report, The
to for our research cautioned capex and opex. Smart capex sustainable telco: engineering
against its potential as an uses analytics and AI to drive networks for net zero, to find
alternative to employees creating better informed and data-driven out more.
code. Deutsche Telekom is investment decisions for both
experimenting with using GenAI fixed and mobile networks. BT
where there are small code says it has made annual savings of
vulnerabilities in large £20 million through its Sweeper
applications. It has built a tool project which involves the use of
which automatically fixes code geospatial data to rethink how
created by humans. fiber networks are built. And
Vodafone, meanwhile, has Verizon is using computer vision
identified some value of using for its 5G urban network
GenAI “to remove some of the deployments to better design how
networks are built to make best Predictive maintenance is a
drudgery” associated with certain
use of available street furniture. growing “bolt-on” use case for
coding functions, according to
CSPs and an early example of
CTO Scott Petty. And Verizon is A TM Forum Catalyst is using AI / how AI could be applied to
looking at whether using GenAI to machine learning to automate and telecoms. It enables CSPs to
write code potentially gives it the optimize RAN and FTTH rollouts address maintenance issues in the
opportunity to free up some with a target of unlocking 15%- network before a problem arises.
senior developers and retrain 20% efficiency gains.
them in AI and machine learning.

29 inform.tmforum.org
section 5: Where operators are applying AI

Another key use case, applying example, provide guidance – the experimentation; and “next best
analytics and AI to transforming network requires much higher action”, which involves
OSS systems, has been a major levels of performance than is transforming next best offers from
focus for Vodafone. The operator widely possible now. In a static promotions to dynamic,
has replaced its OSS tools with AI conversation with TM Forum CEO, data-driven experiences.
and machine learning tools by Nik Willetts, KDDI Deputy GM and The main challenge with using AI
extracting the data from its OSS Executive VP, Tomohiro Otani, said in the marketing and customer
systems and connecting it into its that early efforts to build AI experience domain is that a large
Data Ocean – a company-wide algorithms using data from proportion of the data is
data analytics platform. The disparate network systems yielded unstructured. It comprises the
transformation project paid for around 70% accuracy. But the communication that takes place
itself in six months, according to operator has now managed to push between the customer and a CSP
CTO Scott Petty. accuracy up to between 95% and representative, or digital
100% by using a common data set. touchpoint, and takes the form of
Laying the foundations KDDI created a testbed to the written and spoken word.
Delivering new network generate data for AI, comparing Operators have now started to
experiences is an aspiration for algorithms for various use cases use GenAI to:
every CSP investing in 5G such as failure detection and root
cause analysis to pinpoint the most Create insights both for
networks. The basic premise is
effective algorithms for network contact center agents and for
that network capabilities and
operation. Data generated from customer experience,
experiences will become
different vendor systems – such as marketing and sales teams
differentiated based on the
customer, customer group, device element and network management Reconfigure customer
or application that they support. systems (EMS/NMS) – generated chatbots and transition away
But such differentiation implies a different “forms”, says Otani, so from rules-based approaches
level of complexity that could only KDDI needed to collect the data to ones that are based on
be supported by a massive processes into a common data set natural language processing
increase in operational teams or – bringing together lots of different and GenAI. Many CSPs are
the deployment of new AI-driven data sets to train the model. taking hybrid approaches.
intelligence capabilities. Watch the video to find out more: AI will become increasingly
The box on p.32 explains the capable of improving the customer
predominant AI-driven initiatives experience and replacing agents or
that sit within CSPs’ technology freeing them up to carry out other
(IT and network) functions, tasks. The graphic on the next
including network automation and page shows the stages of AI,
autonomous networks. In during which progressively less
considering the overall progress human intervention is required. In
that CSPs are making in using AI the third stage, in around 2040,
to deliver network automation – machines are predicted to become
and in pursuit of the ultimate goal smarter than humans.
of autonomous networks – a Read our report, Counter
realistic assessment would be that Marketing and intelligence: using AI to improve
the foundations are still being laid. customer experience customer experience, to find
These foundations comprise a out more.
company-wide data platform and Some of the earliest use cases for
AI were in specific areas relating TM Forum Catalysts in the
common data sets – as detailed in
to churn and customer “customer” domain include
section 3 – an enterprise
experience. More recently, GenAI Hand over to the virtual assistant:
architecture and a machine
is being seen as a tool to help Future contact center using
learning operations (MLOps)
drive wider transformation. GenAI, and EBITDA Mastery:
capability that enables operators
Remodelling revenue journeys
to deploy quickly and at scale. Those early use cases include
with GenAI and intent analysis,
While some parts of the business churn reduction through the
which explores how GenAI can
may derive value from AI models creation of churn models that
help give customers the services
that are not accurate – but, for reduce the cost of
and packages that best suit them.

30 inform.tmforum.org
section 5: Where operators are applying AI

Stages of AI

Stage 1 – 2020s Stage 2 – 2030s Stage 3 – 2040s

Machine learning Machine intelligence Machine consciousness

Solves specific problems Learns like a human Artificial intelligence that is smarter
Turns big data into usable Solves a variety of problems using than humans
information by detecting patterns association-style processing
and making predictions Teaches itself to solve No human intervention required
new problems
Requires human intervention Also known as artificial super
No human intervention required intelligence (ASI)
Also known as weak or soft AI, or
artificial narrow intelligence (ANI) Also known as hard or strong AI or Applications unknown
artificial general intelligence (AGI)
Examples: digital assistants,
recommendation engines, chatbots, Can learn, generalize and
image and speech recognition plan for future

TM Forum, 2023 (source: Omnisperience)

expertise in AI to be able to offer Rival Korean operator SK Telecom

services and solutions to their has a personal assistant service
customers. For example, both called A. (pronounced A dot), which
AI for new products Swisscom and Telenor are the company says has garnered 1
and services partnering with Nvidia to build out million subscribers within just nine
AI infrastructure, with one of the months of its launch. The service
Asian CSPs lead the way when it
key motivations being to develop was revamped in September 2023
comes to leveraging AI to build
expertise and solutions in their IT to offer personalized entertainment
new products and services. Both
services businesses. services driven by ChatGPT. At
SK Telecom and China Mobile are
Korea Telecom has set up a MWC, SKT set out its “AI to
developing LLMs which – in
dedicated “AI/New Biz” services everywhere” strategy, which
addition to using them for their
division. It includes robotics, includes working to expand into
own internal requirements – they
logistics, AI-enabled contact diverse industries and use cases
plan to monetize through sales
centers and a myriad other including robots, media, data
and licensing agreements with
applications, including video analysis, airports and manufacturing,
other CSPs, with CSPs’
security, blockchain and “smart- machine learning and digital twins.
technology partners and by
partnering with industry verticals. space” services. As part of a new The next page defines key AI-
China Mobile, for example, is strategy announced at MWC, KT driven initiatives being applied to
building LLMs with healthcare and will launch an internal business IT and networks, while p.33 sets
other industrial verticals. Its vision innovation platform dubbed Gen. out TM Forum’s collaborative work
is to build solutions that integrate AIDU that will enable users to in AI and automation. In the next
GenAI capabilities into 5G “directly develop and use [GenAI] section we look at the impact of
connectivity offerings. APIs”. The company aims to GenAI on CSP strategies.
generate KRW1 trillion (about
CSPs with large IT services
$773 million) from AI-based
businesses are seeking to build
services by 2025.

31 inform.tmforum.org
section 5: Where operators are applying AI

Applying AI to networks: defining AIOps, autonomous

networks/operations and network automation

The categories AIOps, autonomous operations,

autonomous networks and network automation
represent four key AI-driven initiatives that sit
within CSPs’ technology (IT and network) functions.
There is overlap between them, largely as a function
of the convergence of enterprise IT and telecoms-
vertical technology.
The four categories all embed the terms automation
and / or autonomous. Automation is the use of
technology to perform tasks where human input is
minimized. The term “autonomous” means
independent and having the power to make its own
decisions. The two terms are different, but they
overlap because for a system to be autonomous it
needs processes within it to be automated.
AIOps is a term coined by the enterprise IT community
and which refers to the enhancement of IT operations
analytics using machine learning technology. Such
operations include automation and performance
monitoring. As such, it is a broad term that is applied,
within the CSP, to the functions that sit under the CIO.
Autonomous operations is a concept which,
according to TM Forum, “presents a new operating
model of an organization that automates capabilities
for learning, self-governance and self-adaptation of its
value streams in order to realize stakeholder value in a
closed loop fashion”.
Network automation, according to Cisco, is “the
process of automating the configuring, managing,
testing, deploying and operating of physical and virtual
devices within a network.
The vision of an autonomous network is one that
operates autonomously, with minimal human
involvement. Within the CSP context it represents an
aspiration, a vision for the network operations
function to use the concepts of intent and closed-
loop automation to replace manual functions.
Sanders. In addition to mapping out network elements,
TM Forum has identified five levels of maturity for this will need to encapsulate policies, behaviors,
autonomous networks between 0 and 5. High levels schematics and other dimensions of network operations.
of automation are required to get to Level 2, but to
progress to Level 3 and Level 4 operators will need to Our report, Leveling up: achieving level 3 autonomous
deploy AI at scale (see graphic). networks and beyond, sets out the progress
operators are making in moving up these levels.
Advancing beyond Levels 2 and 3 to Levels 4 and 5 will
require the use of AI. But AI will not, in itself, enable
September 2023 | www.tmforum.org

August 2023
August 2023

REPORT

CSPs to make this progression. They will need to encode

REPORT

levelingup:
leveling up:
network knowledge in a structure that machines can achievingLevel
achieving Level33
autonomousnetworks
networks
autonomous
reason with and to which they can dynamically adopt. andbeyond
and beyond

“This requires the development of a comprehensive and Author:

Author:
Mark Mortensen, Contributing Analyst
Sponsored by:
Mark Mortensen, Contributing Analyst Sponsoredsponsored
by: by:

robust ontology,” says Telstra’s chief architect Mark

Editor:
Author: Dr. Mark H Mortensen, Contributing Analyst, TM Forum Dawn Bushaus, Contributing Editor
Editor:
Editor: Dawn Bushaus, Contributing Editor, TM Forum Dawn Bushaus, Contributing Editor

32 inform.tmforum.org
section 5: Where operators are applying AI

TM Forum’s collaborative work in AI and automation

TM Forum, 2023

TM Forum members are working challenges related to security, the use of AI in IT and network
together in collaboration projects privacy, accuracy, performance operations. Meanwhile, Catalyst
and Catalyst proofs of concept to and scalability. co-innovation projects are turning
solve common AI challenges for the In the near term, TM Forum’s AI ideas into proofs of concept as
telecoms industry. Central to that is innovation agenda focuses on members explore applications of
a new Innovation Hub, hosted by gaining maximum benefits from machine learning and GenAI.
Jio in Mumbai, which opened on 1 AI, safely and cost-effectively, Members can keep control of AI
December 2023. Founding using existing IT and network deployments and avoid
members are Accenture, Deutsche systems and data. The medium- accumulating technical debt by
Telekom, Google Cloud, Orange, term goal is to exploit the full following TM Forum’s best
Reliance Jio, Telenor and Vodafone. potential of AI to reduce practice in AIOps and AI
Among the early projects will be operational costs and create governance, and prepare for the
the development of TMF Guru, a opportunities for growth future by upgrading IT and
GenAI/LLM conversational search by taking AI from a bolt-on network systems for AI with
capability to simplify and improve automation aid to an integral part TM Forum’s blueprint reference
how to find relevant material of IT and network system designs. architectures, ontology and Open
digitally. The project members are The Forum’s AI collaboration APIs. Many operators are
setting out to create an projects are creating standards adopting the Autonomous
architecture blueprint that and exploring best practice for Networks Levels Evaluation
outlines how GenAI technology AI-driven, intent-based closed Methodology to set targets for
can be seamlessly integrated into loop automation, building an network automation and track
the telco landscape, addressing industry-agreed framework for their progress.

33 inform.tmforum.org
BENCHMARK

section 6:

The impact of GenAI

Telecoms operators are not usually early adopters of new technologies, but the surge in interest in GenAI
following the commercial launch of ChatGPT in late 2022 has proved an exception. It immediately captured the
imagination of boardrooms, in part because even the least tech-savvy executives could see the power of the
technology. The world’s most pre-eminent consulting firms also helped ensure that the CEO community was wise
to its potential.

F
or technology leaders this seen as the year when operators for scale deployments. At the
enthusiasm about GenAI start to put GenAI into production same time, there is skepticism
was a double-edged sword. (see timeline on p.46). about the ability of the
On the one hand they were given In December 2023 we published operator community to build
the freedom to experiment our first piece of research on its own LLMs.
aggressively with the new GenAI, Generative AI: Operators There is no clear, coherent
technology; but on the other, take their first steps. This section strategy for collecting,
it put them under pressure to draws from that research but also consolidating, storing and
deliver results without the comfort includes new insights based on using the vast amounts of
blanket of standardization or the our operator briefings. They unstructured data in, and
guidance of traditional telecoms revealed that: around, the telecoms operator
technology partners. business and which does not
For a large number of CSPs,
However, there are signs in early GenAI is now the main focus currently sit in CSPs’ data
2024 that the first wave of of their AI strategy. lakes / oceans / fabrics.
excitement and aggressive GenAI exists both as a
experimentation is abating and While much of the usage, and
future usage, of predictive AI technology and capability to
that pioneering operators are use for specific use cases and
turning their attention to getting is in the network and network
operations, GenAI’s value is as a “wrapper” that sits around
their first use cases into data analytics, predictive AI
production. Many operators that seen as driving internal
productivity and customer- and GenAI that allows
we spoke to for our research individuals to consume the
believe that GenAI will, ultimately, facing functions.
output of different models.
be transformational. But they are There is a growing realization
well aware that they need to get of the telecoms – and In this section we explore some of
some use cases up and running – telecoms-function – specificity these issues in more detail and
and to clearly demonstrate required for GenAI to deliver point to how they are being
business gains – to start building the accurate results needed addressed by the operators we
momentum. This is very much spoke to.

34 inform.tmforum.org
section 6: The impact of GenAI

Building families of GenAI use cases

Customer operations Sales & marketing Network IT / software engineering
• Customer chatbot • Marketing collateral generation • Field service operations • Automated code generation
• Call center agent • Personalized customer / guided assistance and testing
documentation and coaching email scripts • Network / capacity planning • Automating repetitive tasks
• Website assistance • Social media automated • Network security testing (e.g. data mapping)
• Predictive and personalized responses • Post Mortem Creation • Detection of code security
services vulnerability
• Root cause analysis

Product innovation Internal knowledge, Business operations

• Carrier billing training & development • Legal / contract
• Personalized services • Evaluating new trends / • Fraud management
• Voice value-added services developments • Partner management
• B2B customer call services • Competitive analysis (e.g. roaming)
• Supply chain analysis • Human resources

For a more detailed introduction We have identified seven families of Global Telco AI Alliance, which is
to GenAI, the technology, the use cases for GenAI (see graphic developing LLMs for the telecoms
emerging value chain and some above). Most of the CSPs we spoke community. In a statement
of the broader challenges to have identified more than 100 providing an update on the
and opportunities, read our individual use cases in total – “We alliance at MWC in Barcelona in
earlier report: got to 120 and then we stopped February, SK Telecom CEO, Ryu
counting,” one CIO commented – Young-sang, said: “We as telcos
TM Forum | December 2023

although the groupings of families need to develop tailored LLM[s]

REPORT

of use cases may differ slightly for the telco industry to make
generative from one operator to another. telco operations more efficient,
AI: which is a low-hanging fruit. Our
operators Clearly no CSP can put more than
take their
first steps 100 use cases into production, ultimate goal is to discover new
however. As such, they are having business models by redefining
relationships with customers. The
Author:
Mark Newman, Chief Analyst

to prioritize, and based on our

Editor:
Ian Kemp, Managing Editor

discussions the following two Global Telco AI Alliance brings

approaches – or sets of priorities synergy to its members by
– are emerging: allowing them to achieve more
Where does GenAI sit? by working as a team.”
Use cases that involve
There is no suggestion that the Many use cases sit within the
capturing content and
organizational structure that CSPs first category. They include:
knowledge from customers
have put in place for predictive AI
and using that to deliver a Building customer-facing
should, in any way, be revised or
better, more tailored customer chatbots that embed GenAI. In
duplicated for GenAI. But at the
experience most cases this means taking
same time it is already evident that
Use cases across the entire an existing rule-based chatbot
the use cases for GenAI are very
organization – many of which and adding new capabilities and
different from those for predictive
are in parts of the business that functions using GenAI for more
AI. The two technology categories
are not normally early adopters difficult, customized queries.
are not interchangeable, although
in many cases one will of new technology – which will Complaint resolution.
complement the other. deliver productivity gains. Customers send emails /
Furthermore, the technology Many of these use cases will be messages and GenAI reads
stacks that underpin them, and the made possible by the use of them and either forwards
overall size and shape of the copilot services. them as trouble tickets or
investments required to leverage Indeed, these two approaches resolves the issues.
them, are very different. dovetail with the stated aim of the

35 inform.tmforum.org
section 6: The impact of GenAI

Service experience analysis. when its accuracy is significantly

One European CSP less than 100%.
interviewed for this research “So long as you have a human
used GenAI to analyze the being making the final decision,
seven million pieces of GenAI could be playing the role of
customer feedback received So long as you have a data miner, and also providing
over a year. It was able to get a human being innovative recommendations
detailed insights into the
nature of customer complaints
making the final instead of actually making final
decisions,” says China Mobile
broken down by service, type decision, GenAI Director-Researcher Dr. Lingli Deng.
of complaint, time of day, date could be… providing
and so on. This enabled it to
prioritize when and where to
innovative GenAI in the network
invest in delivering an recommendations While there are some use cases for
improved level of service. instead of actually GenAI in the network – for example,
in planning – CSPs are largely
Canadian operator Telus is making final skeptical about its application in
using GenAI to predict when
customers will have negative
decisions.” the day-to-day running of telecoms
experiences – for example, networks. This view is shared by
Dr. Lingli Deng, Director- Chung Suk-guen, Chief Global AI
when the network is likely to Researcher, China Mobile
go down because an engineer Officer at SK Telecom. “I’ve met a
is carrying out local repairs in lot of network experts in this
the area. industry that are trying to find
some real-time data monitoring or
Using GenAI as a means of root cause analysis use cases for
simplifying the customer According to Colt Technology
GenAI. But I don’t think it’s possible
experience. China Mobile is CEO, Keri Gilder, CSPs are on a
at this moment,” he says. “GenAI is
developing a feature on its retirement cliff edge. Many of the
not fast enough to handle tons of
mobile app which allows roles within technology operations
data in real time. And also, the data
customers to use natural teams require expertise on legacy
is not human readable.”
language to interact rather than systems and technologies. As
going through increasingly employees in these roles near Spanish telecoms group Telefónica
difficult, complex menus. retirement, operators need to find does not currently have any GenAI
a way to capture their expertise use cases for the network in
In the general category of practice, but it is evaluating a
and train new teams. GenAI can
productivity, and the use of number of different opportunities:
play an important role in this
copilots, there are likely hundreds
training or retraining process. The operator is training its
of different use cases across
the business specific to the work Call center functions face a LLMs with technical manuals
of individuals, small teams and different set of problems. Staff for different network systems.
departments. However, there turnover is high and agents often The LLM can then provide
are some “horizontal” use leave just when they are acquiring recommendations and next
cases that could be deployed the skills and knowledge to be best actions for field engineers.
across many different functions effective. Using GenAI to plug In trials, Telefónica has achieved
and departments. skills and knowledge gaps can a level of accuracy which it
help CSPs to provide a more considers to be promising but
Some of these are as simple as significantly below the
consistent quality of service
building knowledge and insights expertise of its own technicians.
from their call center teams.
from disparate online and offline
resources and communications What all of these examples and It is reporting detailed
channels. But one specific use cases have in common is that information back into the
example that is particularly they put a CSP employee between business about the network’s
relevant and attractive to CSPs the output of the LLM and the performance and where it
relates to knowledge transfer, final decision that it is helping to is facing technical issues.
learning and training. influence. In addition, the output This can be delivered using
of the LLM can be useful even natural language.

36 inform.tmforum.org
section 6: The impact of GenAI

And it is looking to use GenAI When do you expect that GenAI / LLMs will
to build a natural language
interface with the network – have a significant impact on your business?
for example, interrogating
different networks and
Over the next 1-2 years 57%
network functions about how
the network is performing.
A TM Forum catalyst, AI Agent Between 2 and 5 years 37%
empowering higher autonomous
levels, uses GenAI to consolidate
More than 5 years 6%
and orchestrate knowledge
accumulated in network
TM Forum, 2023
operations systems. It uses an AI
agent to perform tasks that were
previously carried out by experts.
Retrieval Augmented Generation In our survey 59% of respondents
From LLMs to SLMs (RAG) is the first attempt to draw said they were considering open
on knowledge from more specific, source models in the cloud for
The key to GenAI adoption within specialized resources – in this case LLM approaches, while 32% said
CSPs is delivering better results the telecoms operator business. It they were considering open
and transitioning from LLMs is the process of optimizing the source on premises (see chart on
which are trained on – and whose output of an LLM by referencing a the next page).
knowledge is based on – public knowledge base outside of the
internet data, to LLMs which data sources it was trained on
deliver results based on
Telecoms-centric LLMs
before generating a response.
knowledge and systems that sit Two key initiatives – which we
within an operator’s business. Fine-tuning is the process of have already outlined briefly in
taking a pre-trained model and this report – are aimed at
As the group CIO of one large North making small adjustments to its producing LLMs (or SLMs) for the
American CSP explains, the leading parameters. It can be seen as an telecoms industry: the Global
LLMs in use today – and specifically alternative to the costly process
OpenAI’s Chat-GPT LLMs – are of training a model from scratch.
designed to output content from a
vast library of knowledge that has While these processes can bring
nothing to do with telecoms. “These about significant improvements in
large language models can do the accuracy of a model, the fact
anything and everything. They can remains that for a CSP they still
involve relying on a model that uses
create a Valentine’s Day message;
knowledge from outside of its
These large
they can write code. But as
telecoms operators we need to step business. Verizon’s Arumugavelu language models
back and say, ‘do we really need all reckons that an alternative approach can do anything and
would be to build small language
that stuff?’”
models (SLMs) capable of
everything. They can
CSPs today are adopting a addressing a specific function within create a Valentine’s
number of different approaches to
take immense LLMs and train
the telecoms business – for example, Day message; they
billing. These would be much
them for their specific purposes. cheaper and quicker to roll out.
can write code.
The first approach has been to
The trend towards smaller LLMs is
But as telecoms
use prompt engineering. This is
the process of understanding how one that is playing out across all operators we need
to interact with an LLM to get sectors and verticals as to step back and say,
businesses figure out the best way
more accurate, relevant results. It
of leveraging GenAI. And the
‘do we really need
involves getting to know a
model’s strengths and process of creating SLMs is all that stuff?’”
weaknesses, experimenting, using becoming more viable thanks to
the growing availability of open Group CIO of a
examples and being specific North American CSP
about prompts and using source technologies.
contextual prompts.

37 inform.tmforum.org
section 6: The impact of GenAI

Which large language model (LLM) approaches are you

considering for your organization (choose all that apply)?

Open source in cloud 59%

Open source on premises 32%

A service model (i.e. buy from

49%
vendor with service agreement)

LLM partnership (i.e. partner with

vendor who supports training a 50%
proprietary model)

Building our own proprietary model 21%

TM Forum, 2023

Telco AI Alliance and China For example, one of the important AI platform – is to build LLMs
Mobile’s Jiutian. It is interesting to focus areas for SK Telecom is both for its own business and for
compare the two approaches and around unstructured data within different vertical markets. It has
assess other potential scenarios the network function. “When we already built a number of different
and opportunities for telecoms- looked at internal source data, LLM models: network, customer
specific LLMs. especially from the network care, traffic – all internal – medical
SK Telecom has put AI at the operations team, we found that a care, IoT, entertainment and smart
heart of its strategy. It has started lot of it was not text, it’s tables or city. It is also working with CSPs
building telco LLMs using graphs, a lot of visual data,” says outside China, providing them
technology developed by its own Chung Suk-guen. As such, SK with its own AI models and LLMs.
internal team, and by three Telecom has put considerable China Mobile sees its role – in the
third-party companies: OpenAI, effort into enhancing models to broader GenAI value chain – as a
Anthropic – founded by former process and understand this rich, developer of foundation models
employees of OpenAI – and multi-modal data. and as a partner to the systems
Konan Technologies. There is no suggestion, at this integrators and application
Much of SK Telecom’s work – and stage, that the members of the developers that build products for
the focus of the Global Telco AI Global Telco AI Alliance will only specific segments and businesses.
Alliance – is building telco LLMs use their own internally developed It includes BSS and OSS systems
that help CSPs to transform the LLM(s) for GenAI-enabled use in this application layer. The
relationship with their customers cases across the business. foundation model layer is trained
and to use this to build more However, by deploying centralized both on China Mobile’s own data
value to the business. One of SK platforms that capture the inputs and data coming in from other
Telecom’s outputs will be a telco and outputs from third-party industries and vertical sectors.
LLM-enabled customer chatbot. LLMs, these operators will be able China Mobile has a team of 700
to generate training data that people working at this foundation
This does not mean that the improves the performance of their model level and partnering
Korean operator is limiting its own LLMs. Within SKT there is with vertical industries on their
work on LLMs to the customer already AI governance in place own LLMs.
experience function. The company which encourages anyone using
is developing a range of LLM third-party LLMs to access them
technologies and capabilities
Where do vendors fit in?
via a central intelligence platform.
that address the specific needs As CSPs contemplate an evolution
of different functions of the The focus of China Mobile’s Jiutian from working with large LLMs to
CSP business. initiative – what it describes as an building or training smaller, more

38 inform.tmforum.org
section 6: The impact of GenAI

specific ones, it begs the question OSS/BSS vendors. It has a vision

whether they are best positioned of building capabilities within its
to leverage this “niche” opportunity. LLM that allow it to understand
As we have said, a case can be and integrate the APIs from
made for creating LLMs (or SLMs) different technology vendors.
for specific functions – for example, And then there are the many Perhaps it would
customer care or billing. But would partnerships among vendors and make more sense
each operator build its own SLMs
to take account of differences in
IT specialists, some of which are for a vendor in
already focused on solutions for
language, culture and choice the telecoms industry. Amdocs, its own specific
of vendors? Microsoft and Nvidia, for example, functional area
Perhaps it would make more announced a partnership last to create SLMs
sense for a vendor in its own November to create LLMs for the
specific functional area to create telecoms and media industries. for its own
SLMs for its own customers? Amdocs and Nvidia say they will customers?
Would competing vendors, customize enterprise-grade LLMs
anyway, be interested in running on Nvidia accelerated
partnering and creating a computing as part of the Amdocs
federated capability? Or perhaps amAIz GenAI platform. The
the best approach is the one companies are exploring GenAI In the next section we look at
adopted by China Mobile and SKT use cases which initially span whether it is possible to measure
– building foundation models and customer care, including the economic impact of AI, both
platforms and then partnering accelerating customer inquiry in terms of its effect on business
with other operators which can resolution by drawing information functions and on financial
then customise them for their own from across company data. On measures such as opex and capex.
particular requirements? the network operations side, they
SKT is already considering the are looking at how to generate
possibility of partnering with solutions that aid configuration,
coverage or performance issues.

39 inform.tmforum.org
BENCHMARK

section 7:

measuring the
economic impact of AI
It is not an easy task to measure the economic impact of AI on the telecoms industry – potential use cases are
wide and varied – and market value estimates vary greatly from different sources. What is certain, is that some
big numbers get touted around when forecasting the benefits of AI to the CSP business.

T
hose big numbers may As an industry the telecoms sector This is not to say that operators
be a function of: has consistently overpromised and have turned against public cloud.
underdelivered when it comes to Rather, they see different benefits
predicting the impact of new from cloud migration flowing into
Job cuts enabled by AI and technologies or new market the business. Those operators which
automation. BT estimates it opportunities. There is a near- are taking a cloud-native approach
could cut 10,000 jobs by consensus that AI could be the to cloud migration – rather than lift
2030 by leveraging latest overhyped new capability, and shift – see the benefits coming
digitization and automation not just in telecoms but also across through in terms of business and
New revenues resulting from
other vertical industries. But there architecture flexibility and scalability,
the launch of AI-enabled is also a sense that, because it is a the ability to make better reuse of
products. SK Telecom believes technology that has come from the technology that they buy or
it could be generating as outside telecoms, expectations are build, and a slow but steady
much as KRW25 trillion (about less prone to being exaggerated. commitment to simplification and
$18.5 billion) in AI-related the elimination of technical debt.
revenues by 2028 Mirroring the cloud? But can any CSP truly put a
Or, most likely, the aggregation As a useful exercise in setting financial figure on the benefits of
of a whole range of small expectations it is worth cloud migration? Indeed, has any
initiatives resulting in cost considering the impact of other CSP managed to quantify the
savings or revenue increases. breakthrough technologies in benefits of digital transformation
McKinsey estimates that GenAI recent years. The best example is more broadly?
could create up to $100 billion cloud adoption – private cloud In practice, new technologies help
in incremental value for the and public cloud (although public employees to do their jobs more
telecoms sector (see graphic cloud is, perhaps, a better point of quickly and efficiently and to start
on p.42). comparison). In the early years of doing new and different things.
Later in this section we look at public cloud adoption cost This is not to say that AI – or more
how opex and capex could be savings were most often touted specifically automation – will not
impacted if CSPs are able to use as the most tangible benefit. But change and eliminate roles. But
AI to generate cost savings. Our as operators lifted applications the most likely scenario is that
analysis shows the extent to which and workloads into the public new roles will be created as a
AI has the potential to create cost cloud many struggled to see result of the impact of deploying
savings across different areas of any of those cost savings the new technology and the
the telco business. coming through. changing dynamics of the
business more broadly.

40 inform.tmforum.org
section 7: measuring the economic impact of AI

We asked the CIO of one of the Measuring the benefits an explanation of what will
CSPs we interviewed as part of happen to the time that is
Most CSPs have rigorous processes
this research whether he thought saved once the solution is
in place for making the business
AI would “move the needle” in successful
case for investing in AI and
terms of the revenues, costs or the process of value creation
ensuring that the benefits are
profitability of the business. “I am a (taking out the cost)
realized. One of the operators we
$10 billion business,” he said. “If
spoke to has put together what it what changes (competencies
you were to ask me can we
calls a Benefit Realization Plan for and behavior) need to happen
generate 1% more revenues over
Intelligent Automation. This for the benefits to be realized.
the next three to five years from AI
identifies the financial and non-
– that is equivalent to $100 million When it comes to the benefits of
financial benefits for each use case.
– my answer would be probably AI, this particular operator has
not. But can I see $100 million of With a basic premise that a single identified a series of categories
savings over three or five years, solution results in time being outlined in the box below.
over three years? Yes possibly.” saved, the plan requires:

How one operator is measuring the benefits of investing in AI

One of the operators we spoke Increased revenues more satisfactory resolution from a
to for this report identifies the In addition to the launch of new call center agent equipped with
following financial and non- products and services, AI-enabled more knowledge about the
financial benefits for each use technologies could help an customer and previous
case when deploying AI. operator sell more existing interactions; or proactive guidance
products and services. A simple relating to new offers, price
Financial benefits example would be enabling a changes or network outages.

Time saving (quantifiable) CSP to sell more routers and There is a strong case for arguing
boosters to both consumers and that improved customer
For example, a new process could business customers by using AI experience drives greater loyalty
result in an 80% reduction in the technology to identify locations and lower churn. As such, a better
time taken by a team to fulfill a with poor coverage. customer experience could be
specific function. It is clearly much seen as delivering a financial
easier to realize the benefits of benefit – lower churn, which can
automation when they result in a Non-financial benefits
easily be measured in terms of
significant saving in one person’s Employee satisfaction retained revenues and lower
time than when they create small By eliminating repetitive functions customer acquisition costs.
– for example 10%-20% savings – – for example, the need to consult
in a single role. Time savings (small and difficult
multiple sources of databases and to quantify)
Cost avoidance (new hires) information, sometimes providing
conflicting data – AI can help An automated process might, for
New hires (in a legacy role) may example, result in a five-minute
be needed – for example, to employees spend more time on
productive, fulfilling tasks. Most time saving each time a task is
handle increased volumes during performed. But the process of
peak periods. Automating corporations run employee
satisfaction surveys, so this should generating quantifiable cost
processes may enable the existing savings from such a task would
team to handle these increased be relatively easy to track.
be difficult.
volumes. Cost avoidance could Customer satisfaction
also relate, for example, to cloud Sustainability
There are many AI use cases which
computing or software-as-a- could result in better customer An initiative to reduce energy
service (SaaS) where costs would experience – as measured by using usage – for example, by turning
normally be incurred as a result of Net Promoter Score (NPS). An off different parts of the network
the ingestion of poor quality data. improved customer experience during periods of low usage –
Automated updates and could result from an improved could be measured in terms of the
validation of data quality would digital interface – for example, a ensuing reduction in carbon
remove the need for such more intelligent chatbot; swifter, dioxide emissions.
additional spending.

41 inform.tmforum.org
section 7: measuring the economic impact of AI

Potential impact of generative AI on the telecoms industry

140-180 450-680

60-100 310-500

250-400 ~35-70%
incremental
~15-40% economic
impact
incremental
economic
impact

Advanced analytics, New generative AI Total use case-driven All worker productivity Total AI economic
traditional machine learning use cases potential enabled by GenAI potential
and deep learning including in use cases

Updated use case estimates from “Notes from the AI frontier: applications and value of deep learning,” McKinsey Global Institute, April 17, 2018.
Source: McKinsey Global Institute, The economic potential of generative AI, June 14, 2023.
TM Forum, 2024 (source: McKinsey & Company)

Middle East and Africa telecoms that can be derived from GenAI, research told us it had a
group e& has put in place a Value according to Chief AI Global target of generating $100
Realization Office whose remit is to Officer, Chung Suk-guen, is million in financial benefits
ensure that the company can improving both the employee and (EBITDA gains) from AI in
maximize the value of initiatives customer experience. 2023 (equivalent to 0.5%
and investments in areas such as of total revenues).
AI. It prioritizes different projects Crunching the numbers South Korean telecoms
based on their ability to deliver on operator KT Corp is targeting
six different values including A handful of CSPs have gone
public with big financial targets KRW1 trillion (about $750
revenue generation, efficiencies, million) in annual revenues
customer experience and time to for the benefits they hope to
derive from AI and automation: from its customized AI
market. After the use case is services by 2025.
implemented its success is Vodafone announced a €1
measured and monitored through billion cost-cutting plan in Strategy consultancy McKinsey &
techniques such as A/B testing and November 2022. The operator Company has produced some
multivariate testing – a technique says it has already generated extraordinarily bullish forecasts
for testing a hypothesis in which €500 million in capex/opex about the potential financial
multiple variables are modified. savings through network impact of “traditional AI” and
automation over the past GenAI on telcos. Some of the data
SK Telecom has one of the and analysis that relates to the
clearest, most ambitious three years.
impact of analytics and traditional
strategies for harnessing the BT is targeting £500 million AI is based on updated use cases
potential of AI. It is the founder of annual cost savings through first explored in 2018. The new
the Global Telco AI Alliance, its digital transformation analysis is based on the impact of
whose five members – the others program. But it also has its GenAI which, it says, could create
are e&, Deutsche Telekom, Singtel eyes on a bigger, £1 billion $60 billion to $100 billion in value
and Softbank – aim to develop prize courtesy of investments – it doesn’t say over what period
multilingual LLMs for their in new technologies such as AI – through the deployment of new
businesses. But it is also one of and automation. use cases and $140 billion to $180
the operators that is least keen to One North American CSP billion through general productivity
target efficiency gains from its AI that we spoke to for this gains (see graphic above).
investments. The biggest value

42 inform.tmforum.org
section 7: measuring the economic impact of AI

Telcos have already cut costs using GenAI

Customer service 16% 39% 32% 7% 6%

Network 34% 40% 15% 8%

2%
IT 47% 35% 13% 3%

2% 1%
Marketing & sales 50% 30% 15% 3%

1% 1%
Support functions 87% 6% 6%

Cost reduction attributed to GenAI in different domains, % 0 1-5 6-20 21-50 50+ Don’t know

GenAI CXO survey 2023, n=130. Question: What is the impact (% cost reduction) attributed to generative AI in the different domains? Percentages consider answers
only from respondents claiming to have achieved impact and to have had at least some use cases in execution; figures may not sum to 100% because of rounding.
TM Forum, 2024 (source: McKinsey & Company)

McKinsey surveyed 130 CSPs to Potential impact on capex/opex of using AI

gauge sentiment about the
potential for GenAI to reduce Cost category, opex/ % of total global
costs in five different parts of the Total $ value billions
capex item telco revenues
CSP business: customer service,
Global operator revenues 100% 1,779.80
network, IT, marketing and sales,
and support functions. Opex
Respondents saw the biggest
cost savings in customer service Network operations 14.1% 251.5
and network, with 39% and 40%
Network infrastructure 8.7% 154.7
of respondents, respectively,
anticipating savings of between Utilities 3.1% 54.1
1% and 5%, and 32% and 15%
respectively expecting savings Content 6.6% 117.5
of between 6% and 20%
R&D 0.7% 12
(see graphic above).
Sales & marketing 11.6% 207.1
The impact on capex/opex
General & administrative 10.6% 188.5
Drawing inspiration from the
McKinsey survey it is interesting to Cost of Devices 8.8% 157.5
look at how opex and capex would
Depreciation and
be impacted if CSPs are able to amortization (D&A) 18.7% 332.8
use AI to generate cost savings expenses
(see table right). Our starting point
Total opex 82.9% 1,475.70
for building out such analysis is a
breakdown of capex and opex for Capex
the global telecoms operator
business produced by research Property, plant & equipment
16.8% 299.6
(PP&E)
firm MTN Consulting. For each
category we have calculated its Software licenses 1.6% 29.08
size as a percentage of total global
operator revenues. Total capex 18.5% 328.7

For each cost item we made our Total opex plus capex 1,804.30
own judgment about the extent
TM Forum, 2024

43 inform.tmforum.org
section 7: measuring the economic impact of AI

Potential (long-term) impact of AI on capex/opex

Impact Bullish target % Impact Bearish target % Impact
Opex / capex item
of AI reduction ($ billions) reduction ($ billions)

Network operations 20.00% 50.3 5.00% 12.6

Network infrastructure 0.00% 0.0 0.00% 0.0

Utilities 10.00% 5.4 2.00% 1.1

Content 2.00% 2.4 0.50% 0.6

R&D 0.00% 0.0 0.00% 0.0

Sales & marketing 10.00% 20.7 2.00% 4.1

General & administrative 5.00% 9.4 1.00% 1.9

Cost of Devices 0.00% 0.0 0.00% 0.0

Depreciation and amortization

0.00% 0.0 0.00% 0.0
(D&A) expenses

Property, plant & equipment (PP&E) 5.00% 73.8 1.00% 14.8

Software licenses 0.00% 0.0 0.00% 0.0

Total opex/capex saving 162.0 35.03

Saving as % of total costs 9.0% 1.9%

No impact Small impact Moderate impact Big impact Transformational

TM Forum, 2024

to which AI has the potential to Big impact Moderate impact

create cost savings. We created
Utilities and sales and marketing General and administrative
five separate categories:
will both benefit considerably expenses (G&A) is not a part of
transformational, large impact,
from AI. The majority of utilities the business where you might
moderate impact, small impact
costs are from energy required to normally expect a major impact
and no impact (see table above).
operate telecoms networks. But from the deployment of a new
AI is already making huge strides technology. However, GenAI in
Transformational impact in helping operators to optimize particular has the potential to
We believe that AI and automation their energy consumption. automate many processes and
will have a transformational impact Sales and marketing includes functions across the organization.
on network operations, the total customer relationship Furthermore, functions such as
cost of which has been increasing management (CRM) and customer HR and finance, which are
slowly in recent years as operators experience, which is the biggest necessarily document- and
have built out new networks and single focus of CSP investment in data-heavy, could see significant
technologies and traffic levels have GenAI. There are many other use improvements in productivity
grown. However, network cases across predictive AI and through the use of capabilities
automation and the slow evolution GenAI within the broader sales and such as copilots.
to autonomous networks will marketing category which enable Property, plant and equipment
transform network operations over operators to gain more granular (PP&E) capex mainly comprises
the next five to ten years. insights into customer sentiment the network itself, both passive
and intent and deliver more and active infrastructure. AI is
targeted offerings. already helping operators to better

44 inform.tmforum.org
section 7: measuring the economic impact of AI

plan and optimize their networks, categories we were able to assess

and it will also become central to the total overall saving to the $1.8
the functioning of the network. trillion CSP business. We identified
a bullish and a bearish percentage
Small impact cost saving for each category in
order to show a range of potential AI has the
The content category comprises
spending on content for services
cost savings. In a “bullish” scenario potential to
total capex and opex falls by 9.1%
such as operators’ pay-TV from $1.8 trillion to $1.64 trillion. A deliver capex/
businesses. There may be some “bearish” scenario sees a 2% opex cost savings
small savings available to
operators by using AI to better
decline to $1.77 trillion. of between
meet customer needs and tastes. Based on this analysis, AI has the $35 billion and
potential to deliver cost savings of
between $35 billion and $162 $162 billion to
No impact billion to the global CSP sector. the global
We were unable to identify However, these numbers should CSP sector.
potential benefits of AI to the not be taken as forecasts. Rather, it
categories of network serves as a context in which to
infrastructure, operating expense consider the potential cost-saving
elements (spectrum, benefits of AI and, at the same
interconnection and rental / time, a challenge to the industry to
leasing), spend on devices ensure that benefits are realized.
(smartphones, routers and so on),
In the next section we look at the
depreciation and amortization
future of AI and how it could be
(D&A) or software licenses.
applied to telecoms markets.
By allocating a percentage cost
saving for each of the five

45 inform.tmforum.org
BENCHMARK

section 8:

the future of
AI technology
Last year was a tipping point for the use of AI, not just in tech industries but across most verticals the world over.
By February, OpenAI’s ChatGPT had become the fastest growing app of all time, and by year end GPT4 had
started to erode Google’s dominance as a search engine. It had added image creation capabilities and become a
shortcut for millions of repetitive tasks.

I
n 2024 everyday life has already sudden boost to the progress of and colleagues, to the present day
changed due to this technology. AI, which has been on a steady boom in both adoption and
From students being asked to evolutional path for several development, AI has had a long
do their homework at school decades. We are now in the grip of and chequered history. Its
without access to devices, to huge global hype surrounding the progress has tracked the rise in
software developers refining code, technology, with exponential computational power as
there is no putting the AI genie growth predicted for several years. progressively more complex
back in the bottle. But where is computing hardware and more
this journey leading us? Developmental milestones advanced software systems have
The success of GenAI, and provided an increasingly more
From the coining of the term in capable environment for AI
specifically ChatGPT, gave a 1956 by John McCarthy of MIT systems (see graphic).

A selected timeline of artificial intelligence development

Alan Turing test. John McCarthy MIT computer IBM supercomputer Amazon launches ChatGPT moves into Interactive AI enables Artificial general
Originally called the develops scientist Joseph Deep Blue beats Alexa intelligent the mass market, bots to become more intelligence brings
imitation game, the programming Weizenbaum creates chess world virtual assistant in the capable of creating interactive, with systems with
test evaluated a language Lisp, which an interactive champion Garry Echo smart speaker. original content customers able to human-level
machine’s ability to becomes the most program that carries Kasparov. including text, images, speak to their AI intelligence and the
exhibit intelligent popular programming out dialogue in audio and video. capabilities directly ability to self-teach.
behavior equivalent language used in English language. using natural
to, or indistinguishable AI research. Eliza is later dubbed language.
from, a human. the first chatbot.

1950 1956 1958 1959 1966 1970 1996 2011 2014 2020 2023 2024 2028 2034 2045

John McCarthy – Arthur Samuel coins The first Apple integrates Siri OpenAI launches Telecoms use cases Quantum AI brings
together with Marvin the term machine anthropomorphic intelligent virtual GPT-3 GenAI tool for AI begin together the
Minsky, Nathaniel learning, describing a robot, the Wabot1, is assistant into iPhones. for automated mass adoption. next generation
Rochester and Claude computer that will built at Waseda conversations, of powerful
Shannon – coins learn to play checkers University in Japan. generating supercomputers with
the term artificial better than the human-like text the intelligence of AI.
intelligence in a paper person who wrote (GPT was first
for a conference the program. released in 2018).
that launched the
field of AI.

TM Forum, 2024

46 inform.tmforum.org
section 8: the future of AI technology

Sales & Strategy & Software

Operations Product R&D Supply Chain
Marketing finance engineering

Telecoms

Banking

Energy

Healthcare

Retail

Media & entertainment

Very high impact Very low impact

TM Forum, 2024

In the future, AI’s impact will be felt responses to simple commands. Review last September. “That’s
across many industries to differing However, the next phase is for why I’ve bet for a long time that
degrees. Telecoms will be one of those bots to become more conversation is the future
the verticals that will see the most interactive, with customers able to interface. You know, instead of just
significant impact compared to speak to their AI capabilities clicking on buttons and typing,
similarly sized, tech-heavy and directly using natural language you’re going to talk to your AI.”
closely regulated industries, as we and engage in a more human-like The implications of this
show in the graphic. conversation. development for telecoms and
In addition, some predictions Google DeepMind co-founder, other industries could be centered
being made in other industries Mustafa Suleyman, believes the around customer service
about the future of AI adoption future of AI assistants will be in touchpoints as we have seen
may have a bearing on the their ability to interact with for GenAI, but it could also be
telecoms industry’s own roadmap. humans as though they were much deeper from an enterprise
GenAI may well give way to another person. “The third wave IT perspective. As teams within
interactive AI as a focus of our will be the interactive phase,” a business work at computers,
collective interest; the rise of Suleyman told MIT Technology each interacting with a centralized
quantum computing may AI, it could have a multiplier effect
supercharge the AI evolutional on productivity.
arc; and artificial general Queries and requests to
intelligence (AGI) could be the colleagues could be eliminated,
point at which machines become reducing email traffic, with a quick
as intelligent as humans. question to a contextually aware
Artificial general corporate AI providing the answer
Towards interactive AI and a series of suggestions of
intelligence (AGI) how best to proceed. Within a
Generative AI’s powering of
chatbots is remarkable for its
could be the decade the whole nature of
ability to gather and contextualize point at which working in an office environment
could be changed. Indeed, having
data which can perform customer machines become human-parity intelligence built
service tasks better than an FAQ
table. Personal digital assistants
as intelligent as into these systems would be the
like Amazon’s Alexa have been humans. equivalent of adding additional
around for more than a decade, workers, which is the goal of
but they are still just fixed companies driving towards AGI.

47 inform.tmforum.org
section 8: the future of AI technology

Quantum computing Smarter than humans

The developmental timeline for AI In recent months several large
is on a collision course with that technology company CEOs,
of quantum computing, with the including Meta’s Mark Zuckerberg
two likely to start hitting their true and Nvidia’s Jensen Huang, have
potential towards the beginning
It is not difficult to speculated that AGI is less than
of the 2030s. As such they will imagine a future five years away. AGI is a
likely form key components of a where the BSS/ theoretical future branch of AI
quantum AI (QAI) which is greater that will attempt to create
than the sum of its individual
OSS and network systems with human-level
parts: Quantum computing will management intelligence and the ability to
provide the pure speed and system ecosystem self-teach. The ultimate aim is
number crunching, while AI will be for AGI software to be able
the intelligence in need of more
we know today is to perform tasks that it is not
powerful processing. totally replaced by necessarily trained or developed
Quantum computing is a future a single operations for, instead of being limited to a
specific scope. In this respect, AGI
mode of computing which platform powered will be the first true complete
leverages the properties of some
quantum particles and quantum
by a quantum artificial intelligence that solves
mechanics to produce computers AI engine. complex tasks with generalized
able to solve complex problems human cognitive abilities.
much faster than those of today. The societal impact of such
They are particularly suited to capabilities could be huge. With
certain kinds of problem solving, AGI set to perform tasks that we
such as running extremely complex analyze billions of parameters to do today, it potentially hails the
simulations or constructing drive ideal-world business and start of the era of AI systems that
dynamic optimization strategies customer outcomes. are smarter than humans.
for complex systems. As such, their This technological synergy may Computer scientist Geoffrey
potential use in telecoms network also be the inflection point for use Hinton, noted for his work on
management is much anticipated cases like the metaverse, which has artificial neural networks and who
for network digital twins, planning not lived up to its hype. QAI will left Google last year to speak
and dynamic network optimization. also likely form the next generation freely about the dangers of AI,
Adding AI to the quantum mix of computing infrastructure for told CBS News last year: “Until
brings contextual intelligence to many existing technologies. It is not quite recently, I thought it was
the brute force power of the new difficult, for example, to imagine a going to be like 20 to 50 years
wave of machines. The years of future where the BSS/OSS and before we have general-purpose
customer, IT and network data network management system AI. And now I think it may be 20
that telecoms companies have ecosystem we know today is totally years or less.”
stored, but do little with, will be replaced by a single operations
fuel for QAI, which will be able to platform powered by a QAI engine.

48 inform.tmforum.org
BENCHMARK

section 9:

Key findings and

recommendations
In this report we have focused on the opportunities and challenges to CSPs implementing strategies to deploy AI
technologies throughout their organizations. This section brings together key findings from our research and
offers some recommendations for those strategies going forward. Most of those findings and recommendations
are based on the interviews we conducted with 20 senior executives, with some also based on the results of our
survey of 104 executives from 73 operators carried out towards the end of 2023.

AI has had a galvanizing the business has a means of

effect on CSPs’ ability to capturing AI usage across the
bring together teams and organization to help feed into
AI’s impact on the functions from across the its own learning / training; and
CSP organization business to work on joint d) setting up skilling and
projects and initiatives. reskilling programs.
Key findings
We found very little evidence As CSPs aim to democratize
The Chief AI Officer (CAIO)
of operators targeting the access to AI tools and
role is beginning to emerge in
elimination of certain job roles capabilities across the
a few CSPs, but in most cases
and functions through the use organization, centers of
it is the CIO or the CTO who is
of AI in the medium term. excellence should evolve into
ultimately responsible for the
However, many CSPs do see centers of enablement.
organization’s AI strategy. It is
AI as playing a key role as By focusing on the use of AI

common for CSPs to introduce
and when technical and to address issues relating to
shared responsibilities
operational staff retire. staff retirement CSPs may
spanning technology and
business functions. Recommendations have the opportunity to
The roles of an AI center of
automate functions and
Skills in AI / machine learning,

excellence should incude: a) systems, and to reallocate
data analytics and automation
creating AI tools and guides work and responsibilities,
are in high demand. Only
to enable teams across the rather than replacing some
cybersecurity scored higher
organization to start roles. GenAI can also play a
than AI in our survey when it
developing their own use role in helping to train less
comes to both demand and
cases; b) establishing rules experienced teams on the
difficulty in finding the skills.
and guidelines (AI specialized skills required by
Many operators have set up
staff that are retiring.
Governance); c) ensuring that
AI centers of excellence.

49 inform.tmforum.org
section 9: Key findings and recommendations

easy options for using the in readiness for ramping up AI

public cloud for use cases deployments. Computational
that require large-scale costs have been modest.
access to network data. CSPs are only at the very early
Getting the right data
foundations in place Recommendations stages of considering how and
The use of unstructured where to use AI to build new
Key findings
“knowledge” within CSPs products and services. However,
Fewer than half the
AI will play a crucial role in
represents a huge opportunity
respondents in our survey solving the complex network
for deploying GenAI. However,
considered that they were and service operational issues
CSPs need to put the first
doing a good job making use that must be addressed to
GenAI use cases into
of structured data. This is deliver revenues from concepts
production – and to get good
precisely where GenAI can including NaaS, network slicing
results – before attempting to
play a crucial role in capturing and API monetization.
build a data structure and
and deriving value from the
strategy around such data. The cost of experimentation

knowledge that sits within,
Building a data fabric or a
with GenAI today is low,
and around, the edges of
data ocean – a public cloud because of the business model
their organizations.
environment where data can of LLMs which scales directly
Without easy access to data
based on usage (tokens).
be accessed quickly and easily
that sits across the CSP However, cost does become a
– is an essential foundation for
organization it will be consideration when operators
deploying AI at scale. For
extremely difficult for any start to deploy at scale.
those operators which continue
operator to deploy AI at scale.
to retain a patchwork of data Operators are significantly

Making data AI-ready is a
lakes across public and private increasing their spend on
major challenge for those cloud environments, the next public cloud service providers
CSPs which keep data in best option is to deploy which are now offering licenses
multiple data lakes across multiple use cases in a specific for copilot services.
the organization and in a functional area simultaneously. There is a risk of incurring

combination of public and
If operators are to fully leverage technical debt through the
private cloud environments.
AI across their entire uncontrolled purchase and
CSPs are at the early stages
organizations, and to run deployment of different AI
in terms of trying to figure models, gain insights and models and services across
out how their commitment to automate processes that sit the organization.
the principles of openness between different systems and Our analysis of CSP capex and
and composability in their IT functions, they will need to opex presents two scenarios
systems – and perhaps in build a taxonomy – or rather for the potential savings that
future in their networks an ontology – that defines how could result from the
– shapes their choices in these systems and functions successful use and deployment
terms of the AI models that relate to each other and the of AI. In a “bullish” scenario
they deploy in use cases dependencies between them. total capex and opex falls by
across the organization.
9.1% from $1.8 trillion to $1.64
The heavy dependence on
trillion. A “bearish” scenario
partnering with hyperscalers sees a 2% decline to $1.77
for GenAI services may create trillion. However, it is worth
problems when it comes to Benefits, KPIs and getting noting that when CSPs first
use cases that present a return on investment started to embrace public cloud
privacy and security issues services many thought it would
– which might be the case Key findings
lead to significant cost savings.
with many use cases involving The costs so far incurred by This is not how public cloud
the network. As such, CSPs CSPs in their AI deployments deployment has played out and
are exploring options for how mainly comprise buying there is every prospect that AI,
to use hyperscaler capabilities services / solutions for similarly, could fail to deliver
while keeping their data on specific use cases – such as anticipated cost savings.
premises. However, as things energy saving in the network
stand there appear to be no – and the teams they have built

50 inform.tmforum.org
section 9: Key findings and recommendations

Operators which own

GenAI although many of the As part of their evaluation
networks in many countries feasibility analyses we have process for putting LLM-
are being frustrated in their seen include head-count enabled use cases into
attempts to replicate similar savings as the key benefit. production – and to ensure
use cases across different Most CSPs will likely use
that costs do not exceed the
regions. In many cases, they multiple LLMs in order to benefits – operators need to
use different vendors for avoid lock-in, get the best carefully evaluate: a) the
different parts of their business value and ensure that they opportunity cost of waiting
in each country and they are benefit from new technologies until the accuracy of the LLM
unable to replicate a use case and capabilities. This will, gets close to 100%; b) the
from one region to the next. however, incur higher costs in development time to
Recommendations terms of training, fine-tuning implement use cases
and so on. commercially; and c) the
SPs should make every
C “service” cost of deploying
effort to learn approaches and Many CSPs are now looking
the use case, which can be
techniques for minimizing to build platforms that sit significant if it is deployed
spend on LLMs, even though above the LLM(s) and which at scale.
the costs they are incurring capture inputs and outputs to
today are modest. When use use this for their training. If CSPs are to use GenAI to
cases go into production LLM exploit the graphical and
There is a trend towards the
non-text data that comes out of
costs will soar. use of smaller, more their network organization, the
n AI governance program
A specialized LLMs, sometimes LLMs that they use will need to
needs to be put into place to referred to as small language evolve from unimodal – where
reduce the risk of incurring models (SLMs). However, it is they are only able to take one
technical debt while, at the still unclear whether these will type of data as an input and
same time, encouraging usage be developed by each operator, generate the same type of
and experimentation. by a group of operators or by output – to multimodal where
CSPs must start to deploy
technology vendors. they receive multiple modalities
FinOps for AI as they start to GenAI is not, as things stand,
of input and generate multiple
scale AI deployments and a technology and a capability types of output.
open access to AI tools to for replacing people. Rather, For a more detailed exploration
teams across the organization. it is a tool for making jobs of GenAI, and our findings and
SPs should embrace the
C easier, making better decisions recommendations for using the
principles of openness and and for creating time to technology, read our report:
composability in their perform more valuable tasks.
deployments. If they don’t, Most LLMs today are trained
TM Forum | December 2023

REPORT

decisions made today in terms on text. However, much of

of how they use AI within their the data coming out of the generative

organizations risks leading to networks organization AI:

operators
technical debt in the future. comprises chart and take their
first steps

graphical representations. Author:

Mark Newman, Chief Analyst

Recommendations
Editor:
Ian Kemp, Managing Editor

Operators should not create a

dependency on a single LLM.
GenAI’s impact on CSPs They need to avoid lock-in
Key findings and have access to the
latest versions.
An improved customer

experience and better CSPs need to capture all

productivity are seen as the the learnings, inputs and
main benefits of deploying outputs from LLMs across
GenAI. Operators are reluctant their organizations and feed it
to be drawn on the (opex) back into their training and
efficiencies that might be future strategies.
possible through the use of

51 inform.tmforum.org
additional features
& resources
Sponsored features

53 | No Shortcuts to Excellence: Achieving Precision and Efficiency with GenAI

56 | Is Your Data AI Ready?

59 | Telecom’s AI-Driven Transformation: Embracing a New Era

62 | Setting the standard and shaping the future.

64 | Putting GenAI to Work in Telecom

66 | Moving Towards Business Goal-Oriented AI In Telecoms

Additional resources

71 | TM Forum Open Digital Framework

72 | TM Forum research reports

73 | Meet the Research & Media team

52 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

No Shortcuts to
Excellence: Achieving
Precision and Efficiency
with GenAI

I
n an industry as complex as models means that inevitably, well-labeled, and enriched with
ours, integrating GenAI mistakes will occur. Moreover, relevant metadata. TMForum’s
technology requires a strategic while mainstream GenAI aLDM, a certified integrated data
approach where precision and technologies can provide model specifically designed to
operational efficiency are not just adequate solutions for a wide refine data management processes,
goals, but essentials. It requires range of queries, their efficacy will can be utilized by CSPs to optimize
CSPs to navigate a complex maze, be limited by the data on which their data management practices.
from the meticulous data the models rely on, as well as a This model facilitates the
preparation, and fine-tuning of lack of familiarity with the development of personalized
large language models (LLMs) to nuances of telecom use cases. GenAI solutions through its
the strategic deployment within For CSPs in particular, the key meticulous design and precision.
their core systems. In this process, to overcoming these challenges With comprehensive pre-mapping
there are no shortcuts - the lies in setting up the right data to both Amdocs platforms and
objective is to ensure that the infrastructure and adopting a common non-Amdocs systems,
investment in these LLMs is fully verticalized Large Language aLDM presents an efficient solution
justified by the value they deliver. Model (LLM). This is crucial for for CSPs aiming to enhance their
getting accurate, relevant, and data management practices. It
Building an accurate model timely responses. efficiently translates BSS/OSS data
While GenAI’s democratization Preparing CSP data for GenAI into a unified model, significantly
makes AI more accessible to involves identifying the necessary shortening the Time-to-Market for
everyone, the essence of these data, and ensuring it is clean, deploying GenAI solutions across

53 inform.tmforum.org
SPONSORED FEATURE

both current and future use cases. insightful and accurate. Providing
By improving data reliability and specific examples within the
quality, aLDM enables the accurate telecom context further
calculation of KPIs and supports sharpens this focus, guiding the
the creation of detailed customer technology to produce outputs
profiles. This advanced level of Integration with our that resonate with the sector’s
data management is crucial for systems is critical. unique requirements.
developing personalized GenAI
solutions that can adapt
Otherwise, we’re
Practical applications in
dynamically to individual going to be stuck in telco GenAI use cases
customer needs, thereby leading that small, narrow, The examples below demonstrate
to more effective and responsive
customer interactions.
tip-of-the-iceberg how a solution tailored for the
use case exploration telecom sector can effectively
In the absence of any viable empower CSPs to navigate
model capable of answering the area of just text- challenges unique to our industry.
specialized needs of the telco based and image- Redefining customer billing
domain, industry veteran Amdocs
created amAIz, a telco-grade
based use cases” enquiries: This use case
GenAI solution, designed to drive demonstrates GenAI’s ability to
Director of Network Analytics rapidly and accurately answer
innovation, automate processes and Performance, Tier-1
and optimize decision-making. By common customer questions,
European operator
integrating telecom-oriented data such as “Why is my latest bill so
models with authoritative high?” which would traditionally
taxonomies, the solution ensures demand interaction with a human
delivery of accurate and relevant operator. This is achieved by
outcomes for a broad spectrum of analyzing billing-related terms,
to benefit from higher-quality
CSP business requirements. aggregating amounts and
interactions. Moreover, by
detecting discrepancies. Amdocs
“Integration with our systems is incorporating GenAI into business
amAIz takes this capability a step
critical. Otherwise, we’re going and operational workflows, CSPs
further by incorporating Amdocs’
to be stuck in that small, narrow, can automate tasks, streamline
insights into BSS/OSS systems.
tip-of-the-iceberg use case processes, and provide customized
exploration area of just text- solutions that weren’t possible Optimizing data flow
based and image-based use without these integrations. management: This use case
cases” Director of Network demonstrates GenAI’s ability to
face the data engineer’s crucial
Analytics and Performance, Maximizing telco efficiency
challenge of identifying and
Tier-1 European operator through GenAI
mitigating revenue leakages,
The approach emphasizes Leveraging GenAI to enhance together with the associated risks.
seamless integration with CSP operational efficiencies is yet But issue recognition is just the
network systems, Business another important strategic goal, beginning. More importantly, it
Support Systems (BSS), and demanding the ability to leverage demands the ability to conduct
Operational Support Systems deep industry insights and in-depth analyses of the CSP’s
(OSS), thus minimizing disruption expertise, particularly when operations, from the intricacies of
and ensuring smooth deployment. tailoring interactions to the unique data flow to the complexities of
Integrating GenAI with core needs of our industry. This too is a system integrations and
telecom systems requires a central value proposition of infrastructure design. In this
thorough understanding of the Amdocs amAIz. The key lies in context, the application of GenAI is
data flow—specifically, what data creating precise prompts and not just about utilizing a tool; it’s
enters and exits the GenAI selecting the most fitting LLMs, a about harnessing deep, sector-
models—to ensure that all process reliant on familiarity with specific expertise, which
interactions with BSS/OSS systems the intricate details of the telecom encompasses an intricate
are both transparent and effective. domain. Central to this is having understanding of telecom business
This integration enables CSPs to data that is both relevant and processes and domain knowledge.
achieve enhanced operational contextually rich, enabling AI to
efficiencies and allows customers produce outputs that are both

54 inform.tmforum.org
SPONSORED FEATURE

Charting the future About Amdocs

The journey towards fully realizing
our industry’s transformative
potential is complex, echoing the
need for not just operational
efficiencies and data precision but
also for strategic investment and Amdocs helps those who build the
integration. To remain ahead of the future to make it amazing. With
curve, CSPs must ensure their our market-leading portfolio of
models are designed to software products and services,
understand the specific contexts we unlock our customers’
into which they are being innovative potential, empowering
deployed. This means using them to provide next-generation
well-defined and easily accessible communication and media
APIs that support GenAI model experiences for both the individual
integration with both operational end user and enterprise customers.
workflows and the BSS/OSS. Our approximately 29,000
Moreover, it demands significant employees around the globe are
investment in developing data here to accelerate service providers’
preparation and governance migration to the cloud, enable
functions to ensure outputs are them to differentiate in the 5G era,
both accurate and ethically sound. and digitalize and automate their
This is where strategic partnerships operations. Listed on the NASDAQ
like amdocs become key. Global Select Market, Amdocs had
Identifying a knowledgeable revenue of $4.89 billion in fiscal
partner who brings comprehensive 2023. For more information, visit
industry insights, a deep www.amdocs.com
understanding of data, and a
commitment to operational
excellence, is the natural first
step towards fully realizing the
GenAI opportunity.

55 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

Is Your Data AI Ready?

CSPs must make their data visible,
current, and available on-demand
to maximize new investments in
AI and exploit its customer
experience advantages.

Nearly all communications service providers (CSPs) see AI as a tool to make needed improvements in the quality
and performance of their service offerings and customer experiences (CXs). But most CSPs also admit that most of
their data is not ready for AI. Without governance controls, a unified data view, real-time access, an effective trust
layer, or the ability to deliver the right data and context on-demand, AI initiatives can be undermined. CSPs need to
get their data ready for AI.

Weak data can limit next five years. And yet, most CSPs
AI impact remain unprepared, from a data
perspective, to deploy AI in the
CSPs have struggled to grow
ways that would generate the best
revenue in new categories at a
results and returns on investment.
faster pace than their traditional
revenue streams have declined.
99% of AI
Through this experience, many Preparing data for AI adopters across
CSPs have also found their CXs – TM Forum reports that most CSPs all industries have
particularly in the digital realm –
are compared to and have fallen
– more than 60% – now have data
run into some kind
governance in place sufficient to
behind the world’s top digital support AI initiatives, but nearly of AI deployment
brands like Amazon, Rakuten, one-third do not. hurdle; 42% of
Apple, Google, and Revolut. Most
CSPs – 87% according to this report
At the same time, 65% of CSPs the time that
– believe AI can fundamentally
are progressing toward some hurdle was their
form of unified data model, which
improve their customer experiences
is also an important precursor to own poor data
to help close this gap.
AI deployment. Having these quality.
Consequently, nearly every CSP capabilities within a CSP
reports that they expect GenAI, enterprise is important to ensure
Large Language Models (LLMs), that AI benefits and returns can
and Foundation Models to impact be maximized.
their businesses significantly in the

56 inform.tmforum.org
SPONSORED FEATURE

And while these factors are not operations and customer

what sells AI, it is normal for large experiences alike. Yet, TM Forum
organizations to progress through reports that 40% of CSPs lack
these steps to prepare for AI. In Gain a complete and unified real-time data access and sharing
fact, recent surveys have shown data view capabilities. In the AI environment,
that 99% of AI adopters across all data must be delivered as needed.
Having a complete view of all data
industries have run into some kind Real time and near-real time
in play is crucial. A federated view
of AI deployment hurdle; 42% of capabilities are the baseline.
of enterprise-wide or BSS-wide
the time that hurdle was their own
data suffices – not all data needs
poor data quality. As a result,
to live in one place. The data view
addressing data quality and the
should include both structured
ability to sustain it over time is a
and unstructured data. TM Forum Manage trust
key underlying aspect of AI success
finds that nearly 60% of CSPs are
that should not be overlooked. Deploying a trust layer is central
not using both, though AI’s
Once these organizational to data privacy and security
strengths are fully realized when
structures and controls are in compliance with AI. The trust layer
both types of data are utilized in
place, a CSP should take several should categorize and mask data;
training and responding to
important steps to prepare its remove personally identifiable
customer queries.
data, processes, and people for AI: information (PII); minimize data
duplication; and ensure only
authorized data requests are
fulfilled. AI managers must
Ensure data is holistic, monitor for bias, toxicity,
Take a component-based contextualized, and current hallucinations, IP infringement,
approach rooted in ODA and unauthorized use of private
For AI to succeed, LLMs need
A component-based approach is data, and help others learn how to
context to learn. A BSS solution
highly valuable in AI initiatives for use AI and trust the results
should add context so the AI can
achieving incremental success. An derived from it.
understand ideas like customer IDs,
AI architecture blueprint, rooted in IMEI numbers, and other metadata
a componentized approach like that gives added meaning to data Aria Billing Cloud helps you
TM Forum’s Open Digital used in specific contexts. When prepare for an AI future
Architecture and Open APIs, data is holistic and current, AIs can Your billing data is essential – it’s
makes it easier to identify specific leverage context to answer live not a product - and it’s critical for
goals – such as improving a queries and requests with precise applications beyond operational
customer experience process or information. If a customer asks billing. In AI use cases, correct
metric – and to achieve successive about an advice-of-charge notice, application can yield a more than
gains while deploying AI. the AI needs the real-time data, 30% improvement in EBITDA
Componentization facilitates context, and customer view to through enhanced operational
step-by-step progress without understand and then resolve the efficiency, automation, and
needing to address all data, all query immediately. digitalization.
functions, and all systems at once.
Pre-integration with major cloud Aria offers Aria Billing Cloud, an
and IT platform providers is an open data and API-oriented SaaS
additional benefit of this approach billing solution designed for
that can add further data visibility Establishing real-time data composable architectures and
and componentized BSS access and sharing unified data environments. We
capabilities to the AI environment. understand that many CSPs
Delivering the right data with
grapple with the legacy of data
perfect timing is one of the major
silos, which complicates the
advances AI should bring to CSP

57 inform.tmforum.org
SPONSORED FEATURE

achievement of AI readiness - platforms such as Salesforce, Aria provides the only cloud-native,
factors like unified, trusted, and ServiceNow, and Microsoft, Aria is SaaS billing and monetization
governed data are particularly able to enhance ODA-compliant platform that is both robust and
challenging. BSS components with AI-ready scalable enough to help CSPs seize
Consequently, Aria Billing Cloud data, broadening the impact of AI new market opportunities. The
provides vital data tools, such as across CX and monetization. agility of Aria’s platform allows
Aria AI Datashare, which offers a To learn how to prepare your data CSPs to meet the demands of
suite of capabilities that: for AI, optimize your customer their customers by enabling them
experience, and maximize the return to quickly deploy a wide range of
Train CSPs’ Large Language monetization models across all
Models (LLMs) with metadata on your strategic AI investments,
contact Aria Systems today. types of products and bundles.
to provide a business context For more information, visit:
for their data www.ariasystems.com.
Label and categorize data
About Aria Systems
to enhance context and
control access
Manage trust, privacy, and
compliance for data both in
transit and at rest Aria enables enterprises to
Deliver current, contextualized automate complex usage and
data in real-time to subscription billing in an agile
components, processes, and market environment. Aria Billing
AI platforms on demand. Cloud is top-rated by leading
research firms, and innovative
Beyond AI, Aria facilitates seamless
enterprises including Adobe,
connectivity with major platforms,
Comcast, Experian, Subaru, and
offering CSPs robust alternatives
Telstra depend on Aria to accelerate
to legacy BSS solutions. With
ideation, become customer centric,
dedicated integrations with
and grow recurring revenues.

58 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

Telecom’s AI-Driven
Transformation:
Embracing a New Era

I
n the past decade, the To plunge deeper into AI/ML and reduce costs, and delight customers
telecommunications sector has reap the full benefits, operators through innovation. As a leader in
dipped its toes into AI waters. must undergo a digital overhaul: telco, cloud, AI, and generative AI,
Forward-thinking telecom harnessing cloud infrastructure, AWS looks forward to helping our
operators have either directly executing a data modernization CSP partners build the foundations
embraced AI and machine strategy, adopting a software- of a future-ready organization.
learning (ML) or indirectly defined mindset, and credentialing
benefitted from AI/ML through their workforce with new skills. Enriching Customer
their solution vendors. Early This article will analyze AI’s Experiences with AI-
applications like customer transformative potential across the Powered Journeys
sentiment analysis, marketplace telecom value chain, spotlighting
recommendation engines, next- AI provides opportunities for
cutting-edge use cases covering revolutionizing customer
best-action, and predictive customer experience and network
analytics for network optimization engagement through connected,
operations. These examples personalized journeys spanning
have yielded positive results. The featuring Globe Telecom and Cox
arrival of generative AI could touchpoints and channels. Early
Communications should convince AI/ML implementations had
accelerate widespread AI/ML communication service providers
adoption and must be harnessed limited scope within individuals’
(CSPs) to adopt an AI-centric fragmented customer interactions
by telco leadership as a force for approach to amplify efficiency,
change and transformation. with a telco brand. To transform

59 inform.tmforum.org
SPONSORED FEATURE

customer experiences, telecom personalized, compliant customer With the platform, Cox has
operators must shift from siloed, experiences, timely support, and shortened the average duration
disjointed customer data to a efficient problem resolution, for triaging network issues from
unified view of each user. enhancing satisfaction. hours to minutes. By slashing
AI and generative AI enable fault identification times, Cox
personalized, omnichannel Reimagining Network can accelerate the resolution of
journeys. Globe Telecom, the Operations with Generative complex field issues, improving
Philippines’ largest mobile AI: From RAG to network uptime and the
operator, is a prime example. Knowledge Graphs customer experience.
Facing fierce competition in the While improving customer As network management
prepaid market, Globe, with the experience reduces churn and leveraging generative AI matures,
help of Stellar Elements, an builds loyalty, reliable network we foresee the adoption of
Amdocs company, is leveraging AI services are the product customers knowledge graphs and graph
to tailor and enhance customer expect and CSPs provide. However, neural networks to enable real-
retention strategies. surging scale, complexity, and time network insights and
The cornerstone is the Connected traffic have rendered manual extreme automation. Knowledge
Customer Journey (CCJ) platform, network management outdated. graphs can dynamically mirror
part of the AWS Solutions for Planning, deploying, maintaining, network architectures, mapping
Telecommunications portfolio. By monitoring, and troubleshooting correlations between the myriad
centralizing engagement data — have grown highly complex. AI elements to feed critical data to AI
purchases, support cases, network and generative AI can enable systems. In turn, generative
and app usage, store visits, and intelligent management while models will automate complex
payments — 360-degree driving efficiency. tasks by discerning failure
customer profiles are created. patterns, forecasting capacity
AWS is partnering with CSPs to risks, and rapidly formulating
Further applying AI techniques like build next-generation intelligent optimal remedies. Knowledge
natural language processing and service health platforms that graphs continuously capture
cluster analysis uncovers personas, synthesize and analyze structured network state while generative
preferences, and behaviors. The and unstructured data — network models plan and execute
telco customer data powers AI telemetry, technician notes, and responses, together achieving
applications leveraging AWS’s customer chats — to diagnose and sophisticated closed-loop
continuum of generative AI models fix issues swiftly. automation.
in Amazon Bedrock and Amazon For example, Cox
SageMaker and graph neural This novel approach of combining
Communications’ service health knowledge graphs and generative
networks underpinned by Amazon platform, developed with AWS,
Neptune (graph analytics and AI ushers in a new era of intelligent
ingests data like cable modem networks that can be self-
serverless database). and set-top box network regulating and self-optimizing.
CCJ effectively mirrors customers’ telemetry, technician journal Operators will gain unprecedented
omnichannel experience with CSPs entries, customer interactions, and visibility into network operations.
like Globe in a real-time digital twin troubleshooting guides. The Issues can be rapidly detected and
built on a well-governed, secure, underlying platform uses retrieval resolved before customers are
and private AWS data lake. augmented generation (RAG) on impacted. Additionally, continuous
Foundation models, prompts, and the rich data set within AWS learning based on human feedback
training and context data are infrastructure, leveraging multi- can ingrain unique telco
secured within Globe’s account. RAG orchestration with AWS institutional knowledge while
This knowledge graph-powered Lambda. The solution also used optimizing network efficiency over
model enables accurate, real-time Anthropic Claude on Amazon time. The result will be a dramatic
predictions — churn, journey Bedrock to improve the factual improvement in reliability,
segmentation, next-best experience, accuracy of event data performance, and automation,
communications personalization, summarization. Amazon Titan enabling the next generation
and context-aware chatbots. embeddings and Amazon of intelligent, autonomous
OpenSearch were employed to network architectures.
By integrating journey mapping
accelerate semantic search for
with AI, CSPs can deliver hyper-
relevant information.

60 inform.tmforum.org
SPONSORED FEATURE

Realizing AI’s Full Telecoms adopting these Telecom is harnessing the robust,
Potential in Telecom technologies and mindsets will mature, proven, and smart
not only survive but thrive in the platform, to deliver innovative use
Thus far, AI has enabled tactical
AI-driven era, heralding a new cases that are helping telcos
gains for CSPs. However, AI and
chapter in the telco journey. transform, digitize industries,
generative AI can fundamentally
and reimagine the consumer
augment human capabilities and
About AWS experience. With the power of the
reshape products, processes, and
AWS Global Cloud Infrastructure,
business models.
that’s architected to be flexible and
To lay the groundwork for this secure, AWS for Telecom help CSPs
change requires that CSPs take their business to the next level.
embrace cloud-centricity, institute Through simplifying operations,
data modernization, and adopt a cloudifying networks, monetizing
culture of experimentation around Amazon Web Services (AWS) is
assets, and innovating at scale
AI applications. This strategic the world’s most comprehensive
using AI and generative AI services.
commitment, deliberate and broadly adopted cloud,
investment, and cross-silo offering over 200 fully featured
coordination can and will services from data centers globally.
transform telecoms into digital Millions of customers are using
leaders driven by efficiency, AWS to lower costs, become more
performance, and innovation. agile, and innovate faster. AWS for

61 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

Setting the standard

and shaping the future.

G
enAI is arguably the most strategy. CSG’s capabilities enable How AI can accelerate
transformative and game- CSPs to leverage end-to-end digital transformation
changing initiative the visibility that spans from the for CSPs
world has seen in decades. With engagement to the network layer
AI can power everything from
global spending on AI predicted and empowers a unified customer
product ideation, creation and
to exceed $301 billion by 2026, it experience. By embedding AI
hyper-personalized
is a transformation that across its entire BSS suite,
recommendations to real-time
transcends mere technology. It’s a including Configure, Price, Quote
network optimization and
strategic shift requiring a new (CPQ), Revenue Management and
assurance. One of the most
approach to deliver its promise. AI Customer Experience solutions,
compelling reasons CSPs
is not an end in itself; it is an CSG empowers CSPs to monetize
integrate AI into their BSS stack is
enabler. An enabler that requires any service to any customer on any
to leverage the wealth of data
communication service providers network to drive resilient and
therein. AI allows CSPs to leverage
to tackle a broad set of scalable growth with reduced costs
valuable BSS data across all areas
considerations to ensure they and accelerated monetization.
of their business. This data can
capture its full value at the scale From highly innovative product
then create impactful use cases to
and speed they require. Early ideation and creation, predictive
grow the value of their B2C and
adopters have reported significant pricing and discounts to
B2B markets.
improvements in profit growth personalized bill presentment and
(29%), innovation (35%), time to payments, CSG ensures every Faster time to revenue: AI can
market (31%) and customer subscriber, B2B or B2C, receives a create best-fit products and
retention (32%), according to a frictionless value-driven experience. services in real-time response to
recent IDC Infobrief. When you bring together the buying behaviors, preferences,
robustness of the CSG portfolio competitor offerings, market
As an early adopter of AI, CSG is at
with the power of AI, CSPs can trends, etc. AI could predict what
the forefront of this transformation,
protect and grow their core CSPs’ customers want before they
leveraging the power of AI both
revenues, make their operations know they want it. AI can also
internally and externally. CSG is
more efficient and reimagine ensure that these new products
ideally positioned to ensure CSPs
customer experience. and services are profitable. A
get the most value from their AI

62 inform.tmforum.org
SPONSORED FEATURE

recent IDC report states that 33% customer engagement with Being robust and responsive
of EMEA businesses surveyed say dynamically generated hyper- enough to handle the scale and
that AI projects and initiatives have individualized billing experiences speed demanded by 5G means
accelerated their time to market that analyze individual usage eliminating manual processes and
for new products and services. patterns and predict best value implementing automated, zero-
plans for increased ARPU and touch processing workflows. By
Proactive customer care: AI-
reduced churn. implementing truly cloud-native
based digital assistants powered
Operational Excellence: Field solutions, automation and AI,
by large language models (LLM)
technicians benefit from AI CSPs will be well-positioned to
can handle complex problems,
with instant access to network achieve their primary business
provide quick solutions, and assist
knowledge, topology data, objectives: reducing churn, cutting
human agents in resolving issues
and planning information. costs, and increasing revenues.
faster. These assistants can
This accelerates issue resolution Digital transformation will enable
personalize interactions by
and system installations, CSPs to be flexible, agile, and
analyzing customer data, billing
enhancing productivity and cost-efficient enough to capitalize
history and usage patterns and
customer satisfaction. on all network monetization
access real-time network data to
opportunities 5G promises and to
offer contextual, personalized Transition to Cloud-Native better address their customers’
recommendations when they can Solutions: AI is essential to bridge evolving needs. AI is at the core of
have the most impact. Not only the gap between legacy this transformation, propelling
can Gen AI give better context for technologies and cloud-native CSPs from telcos to data-driven
support agents, but it can solutions. Legacy systems are ecosystem enablers.
anticipate potential care issues inflexible and don’t offer the
(i.e., bill changes/bill shock) and agility 5G demands. Suppose
proactively communicate CSPs want to operate at the scale
About CSG
information to customers to and speed necessary for 5G
prevent a call from even coming monetization. In that case, it only
in. CSG Bill Explainer taps the works if they automate. If they
power of generative AI through want to handle the volume of data
Microsoft Azure OpenAI Service and the resulting complexity, it
to personalize consumer bill CSG empowers companies to
only works with cloud and build unforgettable experiences,
insights across thousands of automation. If they want to draw
complex billing scenarios. With Bill making it easier for people and
actionable insights from this data businesses to connect with, use
Explainer, CSPs can offer an and deliver better experiences for
intuitive billing experience that and pay for the services they
customers and employees, it only value most. Our customer
improves bill comprehension and works with AI, cloud and
reduces contact center volumes. experience, billing and payments
automation combined. solutions help companies of any
Hyper-personalizing marketing: Maximizing the Edge: By placing size make money and make a
CSPs can leverage AI to create charging and other elements of difference. With our SaaS
personalized promotions and the BSS at the network edge, solutions, company leaders can
campaigns, reducing the time CSPs can use AI to control which take control of their future, and
required for campaign design and data is processed and analyzed at tap into guidance along the way
execution and increasing market the edge, reducing processing from our more than 5k-strong
responsiveness. For CSPs, costs and improving customer experienced global team. Want
knowing how to use data and experience. to learn more about how to be a
acting with awareness and change maker and industry
relevance blended with Security and Fraud: CSPs can be
shaper like our 1,000-plus clients?
omnichannel engagement could proactive in detecting and
Visit csgi.com to learn more.
be the magic ingredient to unlock preventing fraud, deploying AI to
maximum customer lifetime value. prevent calls and SMS scams
Even the notoriously hard-to- across networks. AI can block
please B2B customers can be won potential threats by identifying
over with messaging tailored to spoof numbers and patterns in
their preferences. For example, text messages.
CSG Ascendon uses AI to drive

63 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

Putting GenAI to
Work in Telecom
Everyone’s talking about generative
AI. But what does it mean for telcos
and where do you start?

G
enerative Artificial information to create novel abundance of data on customer
Intelligence (GenAI) has content across various mediums behavior, usage patterns, and
emerged as a such as images, text, audio, video, network assets, Telcos are well-
transformative force, capturing and code. At its core, GenAI positioned to leverage GenAI for
the imagination with its ability to leverages large language models growth and innovation.
create realistic content. In the (LLMs) to comprehend complex
realm of telecom, where the queries and produce outputs Use Cases for GenAI
industry faces significant based on learned patterns. in Telecoms
challenges, GenAI presents an
opportunity for renewal and Network Management: GenAI can
Telecom Industry revolutionize network
growth. But how do operators at a Crossroads
begin to weigh the potential of management by automating
GenAI, and which use cases will The telecommunications industry incident responses, identifying
emerge that will inspire industry finds itself at a critical juncture. issues, summarizing incidents, and
first-movers to embrace this Despite years of growth, volatile optimizing incident resolution.
revolutionary technology? market conditions, economic This proactive approach can
downturns, and competition from significantly reduce the $20 billion
big tech have created challenges spent annually by mobile
Understanding GenAI for operators. The need for operators on network outages
GenAI represents a reinvention is evident, and GenAI and service degradations.
groundbreaking advancement in could be the catalyst for Contact Center: With GenAI,
artificial intelligence, distinguished accelerating transformative change. contact centers can streamline
by its capability to generate Telcos possess a unique advantage operations, reduce labor costs,
original content by learning from in adopting GenAI due to their and enhance customer
existing data. Unlike traditional AI extensive networks, vast customer interactions. Watch this video to
algorithms that analyze or act on bases, and crucial role in the global see how the technology can
available data, GenAI synthesizes digital economy. With an summarize conversations and

64 inform.tmforum.org
SPONSORED FEATURE

suggest next best actions for A Call to Action for Telecom About ServiceNow
human agents. As conversational
In a rapidly evolving landscape,
AI continues to evolve, intelligent
telecoms must act decisively to
virtual agents may handle
integrate GenAI into their
more transactional interactions
operations. The potential benefits ServiceNow helps Telcos accelerate
independently.
are vast – from cost reduction and growth, maximize network
Service Support: GenAI proves increased efficiency to improved investment, and reduce costs
invaluable in field service customer satisfaction. across the entire digital
scenarios, offering predictive ecosystem—all while delivering
To learn more about the
maintenance, intelligent seamless experiences. Be it
transformative power of GenAI in
scheduling, and real-time planning for 5G or fiber buildouts,
telecom, explore ServiceNow’s full
monitoring. Technicians benefit managing data centers, launching
article below and gain insights
from guided assistance during and delivering new services, or
into use cases, implementation
troubleshooting, installation, and automating network operations,
strategies, and the path to
maintenance, improving our purpose-built solutions help
sustainable growth in the AI era.
operational efficiency and make work easier across the entire
customer satisfaction. Read the full article to learn more service lifecycle with automation
about how ServiceNow helps put and AI. Built for the fast-changing
GenAI to work in Telecom. world of Telecom, the Now Platform
connects people and data for
greater productivity and innovation.
https://www.servicenow.com/

65 inform.tmforum.org
SPONSORED FEATURE

BENCHMARK

Moving Towards
Business Goal-Oriented
AI In Telecoms

A
rtificial Intelligence (AI) business goals. It’s crucial to map
has changed the game for AI initiatives to those goals and
businesses by enabling establish Key Performance
them to leverage automation and Indicators (KPIs) that directly
data-driven insights to boost correspond to their business
productivity. However, to fully objectives such as revenue
realise the potential of AI, it’s growth, customer retention,
important to strategically align operational efficiency, cost
and integrate AI-driven objectives savings, or improved decision-
with broader business goals. making. Continuously evaluating
To ensure the effective and modifying AI initiatives is
implementation of AI and its essential to ensure they remain
ability to deliver measurable value, aligned with evolving business
organisations must first define requirements.
and understand their core

66 inform.tmforum.org
SPONSORED FEATURE

Leveraging Data and AI for

Intelligent Decision-Making
Telcos and businesses are
continuously chasing a myriad of
goals which include:
Revenue Growth: Increasing Product or
Service
Market
Expansion
sales and revenue by acquiring Innovation

new customers, expanding into

new markets, or introducing
new products or services. Revenue Cost
Growth Reduction
Cost Reduction: Minimizing
expenses and improving
operational efficiency can
help enhance profitability. Intelligent
Market Expansion: Expanding Decision-
the company’s presence in Making
existing markets or entering
new markets.
Employee
Brand Development
and
Building and
Customer Acquisition and Awareness
Engagement

Retention: Increasing the

customer base and retaining
existing customers is critical Customer
Acquisition
for long-term success by CSR and
Retention
improving customer
satisfaction and loyalty.
Brand Building and Awareness:
Enhancing brand visibility
and reputation.
Product or Service Innovation:
Continuously innovating
and improving products or continuously evolving. Businesses on offline recommendations,
services is essential for have been increasingly adopting augmenting manual business
staying competitive. data-driven approaches to decision-making practices.
improve decision-making to
Employee Development and extract more value. The volume of
Engagement: Investing in
Moving Towards
data points influencing each Automated and Integrated
employee training and business has grown exponentially
development can improve
Business Processes
with interconnected business
productivity, morale, and Most telcos and service providers
operations impacting each goal.
retention rates. are transitioning towards
This makes it significantly more
automated and integrated
Sustainability and Corporate challenging to effectively process
business processes by embracing
Social Responsibility (CSR): and make intelligent decisions.
AI within their operations. While
Increasingly, businesses are The evolution of computing power AI adoption has been on the rise,
setting goals related to in big-data space, along with the its initial application has centred
sustainability and CSR, such growth of AI, has resulted in the around building models for
as reducing environmental development of new capabilities data-driven recommendations.
impact, supporting community to process a variety of data However, with the proliferation of
initiatives, or improving rapidly, draw insights, and predict platforms, technologies, and the
social equity. outcomes that can be used to ability to interweave AI within
While the overarching goals of provide intelligent business processes, integrating AI
businesses remain the same, the recommendations. However, is now at the core of driving
methods, approaches, and tools despite these advancements, intelligent outcomes.
used to achieve them have been many business processes still rely

67 inform.tmforum.org
SPONSORED FEATURE

Network Security
Threat Simulation
Customizable 5G
Services
Voice and Chatbot

Data-driven Personalization

Decision-making AI-Generated
Content for
Customer Support
Synthetic Data
for Testing and
Development

Automated Contract
Generation and
Review
Personalized and
Dynamic Marketing
Campaigns

Virtual Network
Design Assistants

Tecnotree Sensa Fabric platform empowering customer journey intelligently selects the most
goes beyond the traditional lifecycle stages with hyper- suitable implementation
AI-based Model Ops platform that personalisation. It cuts across strategies, maximizing success
allows you to build and run AI and the Tecnotree portfolio rates. Its intelligent automation
ML models. It offers a framework operationalizing AI across business capabilities further enhance
that seamlessly integrates model workflows through a governed and operational efficiency by
outcomes into existing business trusted AI lifecycle. The platform seamlessly connecting various
processes, enabling intelligent enables building of rich, contextual, interventions with bots, which
decision-making. personalized, and empathetic conduct simulations to inform
Moreover, the platform features a interactions with customers, and decision-making. This process is
unique and patented technology enables quick launch of new supported by a continuous
that continuously learns and customer experiences. feedback loop and learning
provides feedback to optimise Tecnotree Sensa is designed to mechanism, allowing for ongoing
the outcome towards a specific empower businesses by facilitating refinement and improvement of
goal. This unique capability data-driven decision-making for strategies. With the ability to
allows operators to prioritize optimal outcomes. It measures simulate outcomes and refine
overarching business objectives business outcomes and KPIs for results in real time, Tecnotree
over individual model outcomes, driving personalized experiences. Sensa stands as a pivotal tool in
ensuring alignment with broader Sensa defines clear business driving business success.
strategic outcomes. objectives and enables users to It’s important to understand that
generate applications and leverage leveraging Tecnotree’s proprietary
Tecnotree Sensa’s AI through a dynamic system that telecom standard data model using
Overview - Business continuously measures and the TM Forum SID data standard
Goal-Oriented AI optimizes the stated business can drive outcomes across different
goals through a framework of telco enterprise business personas
Tecnotree Sensa AI-enabled trust and responsible AI. with accuracy; improving KPIs from
platform allows enterprises to concept to market, lead to quote,
rapidly realize business value, Leveraging feedback from
intervention simulations, Sensa quote to cash, usage to pay, trouble

68 inform.tmforum.org
SPONSORED FEATURE

to support, churn to hyper-care or 1. AI-Generated Content for (SLAs), by inputting a few key
procure to pay. This Tecnotree Customer Support: Generative AI parameters. The same AI could
Sensa data model can also be could be used to create detailed assist in reviewing contracts to
extended to serve industry verticals and easy-to-understand support ensure compliance with
such as advertising, IOT, healthcare, documents, troubleshooting regulations and company policies.
e-commerce, Fintech and Real- guides, or instructional content 4. Personalized and Dynamic
Estate and so on such that it can that helps customers set up and Marketing Campaigns: GenAI can
also extend to 3 dimensional use manage their services and generate highly personalized
cases for graphic, video and devices. marketing copy and messages
augmented reality experiences. 2. Virtual Network Design that resonate with individual
Assistants: Large Language customers. It could tailor
Extending Tecnotree Sensa Models with generative promotions and advertising
Fabric with Generative AI capabilities could assist network campaigns to each customer’s
in Telecom engineers in designing and usage patterns, preferences, and
It is obvious that generative AI has optimizing network infrastructures feedback.
great potential to unlock new by suggesting configurations, 5. Synthetic Data for Testing and
growth and innovation channels for anticipating potential issues, and Development: Generative AI could
the telecom sector. To set the stage generating network design be used to create large datasets
for a time when intelligent systems options based on best practices. of synthetic data resembling
improve consumer experiences 3. Automated Contract Generation actual customer usage patterns,
and boost efficiency, CSPs should and Review: Leveraging which can be used for testing new
use GenAI. Currently, telecom Generative AI, the operator could services, applications, and
companies can position themselves automate the creation of custom software updates in a privacy-
to take full advantage of application contracts for enterprise clients, compliant manner.
cases for generative AI. including service level agreements

69 inform.tmforum.org
SPONSORED FEATURE

6. Voice and Chatbot approach ensures alignment with About Tecnotree

Personalization: Implement organizational objectives, fosters
GenAI-driven chatbots and voice efficient resource allocation, and
assistants that can generate maximizes the value derived from
human-like interactions, providing AI investments.
personalized support and upsell Moreover, focusing on business Tecnotree is a 5G-ready digital
suggestions based on the goals ensures that AI projects are Business Support System (BSS)
customer’s history and the purposeful and impactful, driving player, with AI/ML capabilities and
context of the conversation. tangible outcomes that contribute multi-cloud extensibility. Tecnotree
7. Network Security Threat to the bottom line. Whether it’s leads the way on the TM Forum
Simulation: Use Generative AI to through enhancing customer Open API Conformance with 59
simulate potential security threats experiences, streamlining internal certified Open APIs including 9
and generate attack scenarios for processes, or uncovering new real-world open APIs, a testament
training cybersecurity systems revenue streams, AI solutions to the company’s commitment
and teams, thus better preparing tailored to meet specific business to excellence, and continuously
for real-world challenges and objectives are more likely to deliver striving to deliver differentiated
enhancing network security. measurable results and generate experiences and services to both
meaningful returns on investment. CSPs and DSPs. Our agile and
8. Customizable 5G Services: Use open-source digital BSS Stack
Generative AI to create Furthermore, a system-driven comprises the full range (order-
customized solutions for business goal-oriented AI solution to-cash) of business processes
enterprise customers, generating can unlock new opportunities, and subscription management for
network solutions that best fit improve operational efficiency, and telecom and other digital services
their specific needs for 5G deliver enhanced value to industries creating opportunities
applications, such as IoT, AR/VR, customers, ultimately positioning beyond connectivity. Tecnotree also
and smart factory deployment. themselves for success in an provides Fintech and B2B2X multi-
increasingly digital and competitive experience digital marketplace
Summary landscape. The transformation to its subscriber base through
enables users to optimise the Tecnotree Moments platform
Communication service providers operational costs, while significantly
stand to gain significant to empower digitally connected
improving revenue monetisation communities across gaming, health,
advantages by prioritizing business capability with AIML-based insights
goal-driven AI initiatives as they education, OTT, and other vertical
and recommendations. ecosystems. Tecnotree is listed on
embark on their AI journeys. This
the Helsinki Nasdaq (TEM1V).

70 inform.tmforum.org
BENCHMARK

tm forum open
digital framework

A blueprint for intelligent operations ﬁt for the 5G era

The TM Forum Open Digital Framework (ODF) provides a migration path from legacy IT systems and processes
to modular, cloud native software orchestrated using AI.
The framework comprises tools, code, knowledge and standards (machine-readable assets, not just documents).
It is delivering business value for TM Forum members today, accelerating concept-to-cash, eliminating IT &
network costs, and enhancing digital customer experience.
Developed by TM Forum member organizations through our Collaboration Community and Catalyst proofs of
concept, building on TM Forum’s established standards, the Open Digital Framework is being used by leading
service providers and software companies worldwide.

Goals of the Open

Digital Framework
The aim is to transform
business agility (accelerating
concept-to-cash from 18
months to 18 days), enable
simpler IT solutions that
are easier and cheaper
to deploy, integrate and
upgrade, and to establish a
standardized software model
and market which benefits
The framework comprises TM Forum’s Open Digital Architecture all parties (service providers,
(ODA), together with tools, models and data that guide the their suppliers and systems
transformation to ODA from legacy IT systems and operations. integrators).

Open Digital Architecture Transformation Tools Learn more about

Architecture framework, Guides to navigate digital member collaboration
common language and transformation
If you would like to learn
design principles Tools to support the
more about the Open Digital
Open APIs exposing migration from legacy
Framework, or how to get
business services architecture to ODA
involved in the TM Forum
Standardized software Maturity Tools & Data Collaboration Community,
components Maturity models and please contact George Glass.
Reference implementation readiness checks to baseline
and test environment digital capabilities
Data for benchmarking
progress and training AI

71 inform.tmforum.org
tm forum research reports

June 2023 | inform.tmforum.org

REPORT REPORT REPORT REPORT

REPORT BENCHMARK REPORT

Digital
BSS Transformation

for B2B generative

the telco
Tracker 7
making
operators sustainable revenue cutting waves:
diverge on the
AI: telco: growth: complexity the future for partner
path to cloud
with automation
and AI
Open RAN
technology
ecosystems:
operators navigating the time for
take their maze of scope operators to who does what
3 emissions place new bets and can CSPs
first steps compete for more?
Author:
Dean Ramsay, Principal Analyst, TM Forum

Author: Editor:
Authors:
Dawn Bushaus, Contributing Editor
Author: Mark Newman, Chief Analyst Ian Kemp, Managing Editor, TM Forum
Author: Author:
Ed Finegold, Contributing Analyst Mark Newman, Chief Analyst Dean Ramsay, Principal Analyst Author:
Joanne Taaffe, Editor in Chief, Inform
Editor: Ed Finegold, Contributing Analyst
Editor: Editor: Ian Kemp, Managing Editor
Editor: Editor:
Dawn Bushaus, Contributing Editor Ian Kemp, Managing Editor Ian Kemp, Managing Editor Ian Kemp, Managing Editor
Sponsored by:
Sponsored by: Dawn Bushaus, Contributing Editor
Sponsored by: Sponsored by:

TM Forum | March 2023 March 2023

REPORT BENCHMARK REPORT

counter a roadmap for establishing links:

intelligence Asia-Pacific platform models
in the Open
using AI to
telecoms API economy
improve
growth
customer
experience
Authors:
Mark Newman, Chief Analyst, TM Forum
Dawn Bushaus, Contributing Analyst, TM Forum
Author:
Editor:
Teresa Cottam, Contributing Analyst Ian Kemp, Managing Editor, TM Forum

Dawn Bushaus, Contributing Editor

Authors:
Sponsored by:
Mark Newman, Chief Analyst, TM Forum
Dawn Bushaus, Contributing Analyst, TM Forum
Sponsored by: Joanne Taaffe, Editor in Chief, Inform

Editor:
Ian Kemp, Managing Editor, TM Forum

September 2023 | www.tmforum.org September 2023 | www.tmforum.org September 2023 | www.tmforum.org

February 2024 February 2024 August 2023
August 2023

August 2023 | www.tmforum.org

REPORT REPORT REPORT

REPORT
REPORT

closing levelingup:
leveling up: wholesale
REPORT

Reigniting
the loop a numbers game: Cybersecurity strategies: REPORT

cybersecurity strategies: wholesale

REPORT

CSPs aim to exploiting data to risk management moves

achievingLevel
achieving Level33
autonomousnetworks
networks
changes:
changes:
wholesale
telecoms growth
firmly into the telcomoves
spotlight autonomous
automate service
orchestration and drive future risk management
firmly into the telco spotlight
andbeyond
and beyond changes:
rethinking
rethinking
for new
support systems
support systems
fiber operators
a Playbook
for CEOs

assurance strategies for new fiber operators

rethinking support systems

mainframe
Author: sponsored by:
Patrick Donegan
Principal Analyst
HardenStance
for new fiber operators
Author:
Mark Newman
Sponsored by:
Authors:
Sangeet Paul Choudary, Platform Thinking Labs
Nik Willetts, CEO, TM Forum
Author: Sponsored by: Chief Analyst Anthony Rodrigo, CIO, Axiata
Author: Sponsored by: Author: Sponsored by: Editor: Mark Mortensen, Contributing Analyst Sponsoredsponsored Dean Ramsay, Principal Analyst, TM Forum
Author: by: by: Editor:
Dean Ramsay, Principal Analyst Richard Webb, Senior Analyst Dawn Bushaus, sponsored by:
Mark Mortensen, Contributing Analyst Dawn Bushaus, Author: Sponsored by:
Contributing Editor sponsored by: Mark Newman
Editor: Contributing Editor
TM Forum Chief Analyst
Editor: Editor: Author: Patrick Donegan, Principal Analyst, HardenStance Author: Dr. Mark H Mortensen, Contributing Analyst, TM Forum Author: Mark Newman, Chief Analyst
Dawn Bushaus, Contributing Editor
Editor:

modernization:
Ian Kemp, Managing Editor Ian Kemp, Managing Editor Editor: Dawn Bushaus, Contributing Editor, TM Forum Sponsored by: Editor: Dawn Bushaus, Contributing Editor, TM Forum Editor: Dawn Bushaus, Contributing Editor
Dawn Bushaus, Contributing Editor September 2023 | URLEditor:
TBC
September 2023 | URL TBC Dawn Bushaus,
Contributing Editor

September 2023 | URL TBC

charting a course
standout IoT
REPORT REPORT
REPORT REPORT
to cloud native

how software-as-a-service transforming

CSPs take
key steps to
strategies: at the mainframe
edge:
how telcos are
is reshaping BSS: modernize
network inventory
innovating in a
crowded market modernization: Author:

business support racing to a flexible, taking opportunities

Rahul Gupta, Senior Analyst

systems customer-focused future beyond connectivity

charting a course to cloud native
Editor:
Ian Kemp, Managing Editor, TM Forum

Author: Sponsored by: Author: Sponsored by:

Author: Mark Mortensen Sponsored by: Author: Sponsored by:
sponsored by: sponsored by:
Editor: sponsored by:
Editor:
sponsored by: Sponsored by:
sponsored by:
sponsored by: Editor: Dawn Bushaus Editor:
Author: Ed Finegold, Contributing Analyst Author: Teresa Cottam, Contributing Analyst Author: Dr. Mark H. Mortensen, Contributing Analyst Author: Mark Newman, Chief Analyst Author: Mark Newman, Chief Analyst Author: Rahul Gupta, Senior Analyst
Editor: Ian Kemp, Managing Editor Editor: Ian Kemp, Managing Editor Editor: Dawn Bushaus, Contributing Editor ISBN: Editor: Ian Kemp, Managing Editor Editor: Ian Kemp, Managing Editor Editor: Ian Kemp, Managing Editor
ISBN: ISBN: ISBN:

REPORT COVER

October 2022 | www.tmforum.org September 2022 | www.tmforum.org August 2022 | www.tmforum.org August 2022 | www.tmforum.org August 2022 | www.tmforum.org Key Words June 2022 | www.tmforum.org
Bridging boundaries with common standards
Working together
Handing over common standards to each other
Logistics chains knowledge
Cooperation REPORT
REPORT REPORT

DIGITAL OPERATIONS
MATURITY:
autonomous
networks:
into the future:
DIGITAL
DIGITAL ECOSYSTEMS
ECOSYSTEMS
can telcos
achieving from the settingnew
setting new standards
standards
business value to forfor integration
integration
from transformation
Author: Author:
Ed Finegold,Analyst,
Ed Finegold, Contributing Contributing Analyst, TM Forum
TM Forum
Authors :
Editors: Editors: Dean Ramsay (Principal Analyst)
sponsored by: Ian Kemp,
Ian Kemp, Managing Managing
Editor, Editor, TM Forum
TM Forum Author:
Sponsored by:
sponsored by:
Annie Turner,Analyst,
Annie Turner, Contributing Contributing Analyst, TM Forum
TM Forum Annie Turner
Editor: Authors: Annie Turner, Contributing Analyst
Editor: Sponsored by:
sponsored by: sponsored by: Ian Kemp Dean Ramsay, Principal Analyst
Author: Ed Finegold, Contributing Analyst Dawn Bushaus sponsored by:
Author: Ed Finegold, Contributing Analyst Author: Dean Ramsay, Principal Analyst Editors: Ian Kemp, Managing Editor
Author: Teresa Cottam, Contributing Analyst Editors: Ian Kemp, Managing Editor ISBN: 000 Authors and editors: Sponsored by:
Sponsored by:Sponsored by: Editor: Dawn Bushaus, Contributing Editor Editor: Ian Kemp, Managing Editor Dawn Bushaus, Contributing Editor ISBN: 978-1-955998-27-7
Editor: Dawn Bushaus, Contributing Editor Annie Turner, Contributing Editor Author: Dr. Mark H. Mortensen, Contributing Analyst
Image used on tile xxx
Editor: Dawn Bushaus, Contributing Editor
xxx CALL TO ACTION
Report Cover

March 2022 | www.tmforum.org February 2022 | www.tmforum.org February 2022 | www.tmforum.org November 2021 | www.tmforum.org

next generation
next
generation
CEM
now!
the Telefónica way

Author: Mark Newman, Chief Analyst, TM Forum

sponsored by:
Editor: Annie Turner, Contributing Analyst

Author: Ed Finegold, Contributing Analyst Author: Dr. Mark H. Mortensen, Contributing Analyst Author: Dean Ramsay, Principal Analyst sponsored by: sponsored by: with the support of:
Editor: Dawn Bushaus, Contributing Editor Editor: Dawn Bushaus, Contributing Editor Editor: Ian Kemp, Managing Editor

72 inform.tmforum.org
BENCHMARK

meet the Research & Media team

Report Author: Report Editor:

Mark Newman Ian Kemp
Chief Analyst Managing Editor
[email protected] [email protected]

Principal Analyst: Editor in Chief, Inform:

Dean Ramsay Joanne Taaffe
[email protected] [email protected]

Senior Analyst: Commercial Manager:

Richard Webb Tim Edwards
[email protected] [email protected]

Head of Operations: Digital Media Coordinator:

Ali Groves Maureen Adong
[email protected] [email protected]

Global Account Director: Marketing Manager:

Carine Vandevelde Ritika Bhateja
[email protected] [email protected]

Report Design:
Intuitive Design UK Ltd
[email protected]

Published By: All rights reserved. The Forum would like to thank the sponsors and
European Office advertisers who have enabled the publication of this fully independently
TM Forum, Uncommon, 34-37 Liverpool Street, researched report. The views and opinions expressed by individual authors
London EC2M 7PP United Kingdom and contributors in this publication are provided in the writers’ personal
Phone: +44 207 748 6615 capacities and are their sole responsibility. Their publication does not imply
that they represent the views or opinions of TM Forum and must neither
US Office be regarded as constituting advice on any matter whatsoever, nor be
181 New Road, Suite 304, Parsippany, NJ 07054 USA interpreted as such. The reproduction of advertisements and sponsored
Phone: +1 862-227-1648 features in this publication does not in any way imply endorsement by TM
ISBN: 978-1-955998-80-2 Forum of products or services referred to therein.

73 inform.tmforum.org
BENCHMARK

to learn more about TM Forum’s work

on autonomous networks and AI,
please contact Andy Tiller

GENAI - Assessment ID 77961
100% (5)
GENAI - Assessment ID 77961
7 pages
TM Forum GenAI - Driving Intel in NLM Report
No ratings yet
TM Forum GenAI - Driving Intel in NLM Report
32 pages
Netcracker Digital Platform Strategic Overview 2023
No ratings yet
Netcracker Digital Platform Strategic Overview 2023
22 pages
TM Forum Report Building An AI Strategy
No ratings yet
TM Forum Report Building An AI Strategy
58 pages
Telecom Revenue Assurance Solution
No ratings yet
Telecom Revenue Assurance Solution
25 pages
WEF Artificial Intelligence in Telecommunications 2025
No ratings yet
WEF Artificial Intelligence in Telecommunications 2025
29 pages
Landscape Analytics
No ratings yet
Landscape Analytics
110 pages
Make Money With Ai PDF
No ratings yet
Make Money With Ai PDF
182 pages
Technology and Future of Work
No ratings yet
Technology and Future of Work
145 pages
Comcast Cable Swot Analysis
No ratings yet
Comcast Cable Swot Analysis
7 pages
Remarketing Unit Securities and IRS Guidance
No ratings yet
Remarketing Unit Securities and IRS Guidance
3 pages
CatalogONE Ebook Amdocs
50% (2)
CatalogONE Ebook Amdocs
23 pages
Neon Soft Telecom Process Management Company
No ratings yet
Neon Soft Telecom Process Management Company
14 pages
Lecture 1
No ratings yet
Lecture 1
58 pages
Battery OpenCloud Report 2024
No ratings yet
Battery OpenCloud Report 2024
43 pages
Arize U - Intro To ML Observability
No ratings yet
Arize U - Intro To ML Observability
13 pages
Feb 23 Webinar PPT Slides Collated
No ratings yet
Feb 23 Webinar PPT Slides Collated
46 pages
Only The Small Bones - CP Harris
No ratings yet
Only The Small Bones - CP Harris
285 pages
The Battle of Market Positions in The MVNO Field Has Started
100% (2)
The Battle of Market Positions in The MVNO Field Has Started
23 pages
How Telcos Could Use Gen AI To Revitalize Profitability and Growth - McKinsey
No ratings yet
How Telcos Could Use Gen AI To Revitalize Profitability and Growth - McKinsey
19 pages
Bain Brief Selling-The-Cloud
100% (2)
Bain Brief Selling-The-Cloud
12 pages
N-Gram Models For Language Detection
No ratings yet
N-Gram Models For Language Detection
14 pages
Ativa Slide Presentation - Infovista Detail
No ratings yet
Ativa Slide Presentation - Infovista Detail
34 pages
Manual Penggunaan PlaySMS Rev2
No ratings yet
Manual Penggunaan PlaySMS Rev2
121 pages
Conv AI Brochure v2
100% (1)
Conv AI Brochure v2
7 pages
Comcast
No ratings yet
Comcast
3 pages
Autonomous Networks Whitepaper
No ratings yet
Autonomous Networks Whitepaper
14 pages
Aruba Clearpass Competitive Landscape: Beating Cisco and Forescout
No ratings yet
Aruba Clearpass Competitive Landscape: Beating Cisco and Forescout
1 page
Enea Mobile Industry Trends 2022 MR v1
No ratings yet
Enea Mobile Industry Trends 2022 MR v1
34 pages
Full Stack Observability Aag
No ratings yet
Full Stack Observability Aag
3 pages
Harnessing The 5g Consumer Potential
No ratings yet
Harnessing The 5g Consumer Potential
16 pages
Ai in The 5g Network White Paper
No ratings yet
Ai in The 5g Network White Paper
6 pages
SVS402-R2 - Building APIs From Front To Back
No ratings yet
SVS402-R2 - Building APIs From Front To Back
56 pages
AWS Partner Network (APN) Marketing Toolkit
No ratings yet
AWS Partner Network (APN) Marketing Toolkit
46 pages
Use Cases in ML Telco PDF
No ratings yet
Use Cases in ML Telco PDF
10 pages
LUMAScape List Searchable
No ratings yet
LUMAScape List Searchable
11 pages
Dynamic SIM Allocation™: Current Processes Are Inefficient
No ratings yet
Dynamic SIM Allocation™: Current Processes Are Inefficient
2 pages
Mobile Value Added Services
No ratings yet
Mobile Value Added Services
30 pages
AI & Automation in Telecom
No ratings yet
AI & Automation in Telecom
8 pages
Introductory Guide To External Data Monetization: Frameworx Exploratory Report
No ratings yet
Introductory Guide To External Data Monetization: Frameworx Exploratory Report
27 pages
Network Infrastructure Assessment PDF
No ratings yet
Network Infrastructure Assessment PDF
7 pages
The Total Economic Impact of Ibm Netcool Operations Insight (NOI)
No ratings yet
The Total Economic Impact of Ibm Netcool Operations Insight (NOI)
19 pages
Value Drivers of Telecom Industry
No ratings yet
Value Drivers of Telecom Industry
3 pages
Analysys Mason - European 5G Conference January 2022
No ratings yet
Analysys Mason - European 5G Conference January 2022
9 pages
Comcast Ims Network v6 PDF
100% (2)
Comcast Ims Network v6 PDF
19 pages
Acing 5G Pricing: How To Price 5G For Optimal Monetization
No ratings yet
Acing 5G Pricing: How To Price 5G For Optimal Monetization
4 pages
IMS - IP Multimedia Subsystem
No ratings yet
IMS - IP Multimedia Subsystem
21 pages
Automation As A Service - RFP Presentation 27112014 v0 8
No ratings yet
Automation As A Service - RFP Presentation 27112014 v0 8
54 pages
The Road To Telco Cloud
No ratings yet
The Road To Telco Cloud
19 pages
Innovations in Telecom Industry
No ratings yet
Innovations in Telecom Industry
6 pages
CSP Masters Copilot Technical Part 01. AI Fundamentals
No ratings yet
CSP Masters Copilot Technical Part 01. AI Fundamentals
133 pages
Global and Indian Telecom OSS - BSS Market
No ratings yet
Global and Indian Telecom OSS - BSS Market
5 pages
Mobileum - CaseStudy ROAMING CEM
No ratings yet
Mobileum - CaseStudy ROAMING CEM
4 pages
MVNO Critical Success Factors GF 25 Sep 09 V1.0
No ratings yet
MVNO Critical Success Factors GF 25 Sep 09 V1.0
27 pages
Salesforce Certified Agentforce - 22
No ratings yet
Salesforce Certified Agentforce - 22
5 pages
CFTE AI Literacy Whitepaper
100% (1)
CFTE AI Literacy Whitepaper
42 pages
500 System Operations Guide
No ratings yet
500 System Operations Guide
196 pages
Battlecard - Microsoft System Center Vs IBM v11
No ratings yet
Battlecard - Microsoft System Center Vs IBM v11
8 pages
IBM Telecom Solutions Labs
No ratings yet
IBM Telecom Solutions Labs
4 pages
Netvertex Use Cases
No ratings yet
Netvertex Use Cases
66 pages
Alcatel Lucent GeoLocation
No ratings yet
Alcatel Lucent GeoLocation
9 pages
Key Questions For Telecom CTO
No ratings yet
Key Questions For Telecom CTO
4 pages
OPEX Reduction in Telecom Industry: Qarib Raza Kazmi SP-2010/M.M/205 Telecom Business Management
No ratings yet
OPEX Reduction in Telecom Industry: Qarib Raza Kazmi SP-2010/M.M/205 Telecom Business Management
41 pages
Consumer Goods Use Case Examples GenAI
100% (1)
Consumer Goods Use Case Examples GenAI
24 pages
Title: AI in Art and Music Generation
No ratings yet
Title: AI in Art and Music Generation
8 pages
The Globalization of Copyright Exceptions For AI Training
No ratings yet
The Globalization of Copyright Exceptions For AI Training
47 pages
Gen Ai Report - v5
No ratings yet
Gen Ai Report - v5
36 pages
WEF Artificial Intelligence 2023
No ratings yet
WEF Artificial Intelligence 2023
24 pages
AI Governance Comprehensive Sept 2024
No ratings yet
AI Governance Comprehensive Sept 2024
358 pages
Gen AI Report - v8
No ratings yet
Gen AI Report - v8
66 pages
Harmonizing Minds and Machines
No ratings yet
Harmonizing Minds and Machines
13 pages
Globant Mit Report GenAI SDLC
No ratings yet
Globant Mit Report GenAI SDLC
20 pages
English Preparation Guide Aie 202501
No ratings yet
English Preparation Guide Aie 202501
18 pages
New Text Document
No ratings yet
New Text Document
1 page
Human Centered Ai Guidance K 12 Edition 2
No ratings yet
Human Centered Ai Guidance K 12 Edition 2
37 pages
Hackathon PPT Template
No ratings yet
Hackathon PPT Template
15 pages
Ai Agents For Oracle Cloud Erp
No ratings yet
Ai Agents For Oracle Cloud Erp
9 pages
Corporate Fact Sheet (En)
No ratings yet
Corporate Fact Sheet (En)
1 page
Kiran GenAI ML Engineer
No ratings yet
Kiran GenAI ML Engineer
4 pages
AI 1 Tools
No ratings yet
AI 1 Tools
18 pages
Skillible LMS Login Process
No ratings yet
Skillible LMS Login Process
33 pages
Why Many AI Startups Are Consultancies Posing As Software Businesses - The Information
No ratings yet
Why Many AI Startups Are Consultancies Posing As Software Businesses - The Information
6 pages
Rajen Resume
No ratings yet
Rajen Resume
3 pages
CHE482 F24 Course Outline - Group Design Project
No ratings yet
CHE482 F24 Course Outline - Group Design Project
7 pages
A Vaibhav Resume
No ratings yet
A Vaibhav Resume
1 page
Hidding The Ghostwriters - An Adversarial Evaluation of AI-Generated Student Essay Detection
No ratings yet
Hidding The Ghostwriters - An Adversarial Evaluation of AI-Generated Student Essay Detection
14 pages
Ai Law and Ethics
No ratings yet
Ai Law and Ethics
16 pages
Cloud Computing
From Everand
Cloud Computing
Dr. Nirvikar Katiyar
No ratings yet
The Today and Future of WSN, AI, and IoT: A Compass and Torchbearer for the Technocrats
From Everand
The Today and Future of WSN, AI, and IoT: A Compass and Torchbearer for the Technocrats
Dr.Chandrakant
No ratings yet
CPA USA Information Systems and Controls: The Complete Syllabus Guide
From Everand
CPA USA Information Systems and Controls: The Complete Syllabus Guide
Azhar ul Haque Sario
No ratings yet