It’s All About the Data (But Not Just Any Data)

Modernizing data = AI maturity

Nearly all AI leaders have mature data It’s clear from our study that there’s an inextricable link between AI
modernization practices maturity and data management (what we call “data modernization”):
the work involved with ensuring the accessibility, reliability and
Percent of respondents who say they’re maturing or advanced in data management.
timeliness of data for AI and analytics. Nine out of 10 AI leaders say
Now
they’re in the maturing or advanced stages of data management,
In three years
while literally none of the AI beginners rate themselves that way (see
Figure 4). At the same time, having a proper IT architecture and data
100
modernization processes in place was the most important lesson
97%
91% learned for beginners (60%).
80 85%
74% What’s more, more than half of healthcare, banking and auto manufacturing
60
businesses – the industries with the highest percentage of leaders in
53% them – have already made significant progress in modernizing their data,
40
and an overwhelming majority expect to by 2023.
35%
20
20%
0%
0
Leader Advancer Implementer Beginner

Response base: 1,200

Source: ESI ThoughtLab/
Cognizant Figure 4
Further, the same industries
expecting particularly high growth
in AI maturity are the same ones that
expect the greatest gains in data
modernization (see Figure 5). In all,
businesses spend about 35% of their
AI budgets on data modernization,
or about $13.3 million per company. Industries expecting to accelerate their AI maturity will also
Beginners spend even more, or 44%. surge in data modernization
Percent of respondents who say they’re maturing or advanced in data management now
and in three years.
% Change
Media 212%
Energy 192%
Investment 167%
Life sciences 145%
Consumer/retail 126%
Manufacturing 117%
Insurance 111%
Technology 105%
All industries 97%
Telecoms 56%
Banks 55%
Automotive 55%
Healthcare 44%
Now

In three years 0 20 40 60 80 100

Response base: 1,200

Source: ESI ThoughtLab/Cognizant
Figure 5
Data that matters

Increasingly, businesses are finding the most easily Data sources will expand greatly in three years
accessible data sets aren’t enough to make the most
intelligent decisions. By 2023, we’ll see businesses pulling Data used for AI today Fastest growing over three years
Percent of respondents integrating each type of data Percent growth in the use of each type of data.
from wider and more diverse data sets for AI-driven into their AI applications.

insights (see Figure 6). IoT Psychographic

52% 109%
Competitive
Today, IoT, customer and internal information are the main Customer 46% 76%

types of data integrated into AI applications. In many cases, Internal 43% Geospatial 76%

this is simply because of the sheer volume of accessible Social media 21% Real-time 75%

data generated by sensors and customer interactions. But Real-time 20% Social media 60%

other forms of data are where the greatest insights often lie, Manufacturing 18% Macro 41%

particularly when such data is combined. Supply chain 18% Supply chain 35%

Competitive 17% Customer 31%

For example, we worked with a home goods store that wanted Macro 16% Manufacturing 24%
to know where people went when they left the store. By Local 15% Local 15%
combining geospatial data with other data, we discovered that Psychographic 8% IoT 14%
30% of people go to McDonald’s. That insight became very Geospatial 7% Internal 13%
valuable for cross promotions. 0 10 20 30 40 50 60 0 20 40 60 80 100 120

Response base: 1,200

Source: ESI ThoughtLab/Cognizant
Figure 6
Further, during the pandemic, global brands will Over the next three years, companies will double
see different regions of the country and the world their use of psychographic data and ratchet up
open for business at different paces. For one of their reliance on competitive, geospatial and
our clients, that has meant continually adjusting real-time data by about 75%. No AI program is
its product mix, product placement and product complete without “voice of the customer data”
sourcing as supply and demand conditions gleaned from sources such as social media and
change unpredictably, region by region. call center analytics. While much of that data is
generally lost or overly summarized, it’s essential
We’re using machine learning to refine our
to extract insights from that data and get them to
analytic models to predict the effects of
relevant business teams.
everything from ongoing infection rates to
regional weather conditions on future sales
and demand trends. This data includes medical
information from leading healthcare providers, as
well as historical internal data, such as same-store
sales. At each step, we’re using our agile analytics
methods to make sure we deliver the analytics
the business needs most as the recovery from the
pandemic unfolds.
Perishability and other data challenges

With data modernization comes challenges – most of which don’t ease Top 10 data challenges
up to any great degree as maturity increases (see Figure 7). This makes % gap between
Percent of respondents naming each as a top challenge.
sense, given that advanced AI maturity also means more scaling of 'leaders' and
'nonleaders'
pilot solutions, more use of diverse data sets and a greater shift toward Data integrity/quality 38% -5%
modernizing data (i.e., through the use of data lakes and the cloud) vs.
Governance/compliance 36% 5%
simply managing data (i.e., with databases and storage).
Data security 36% -1%
Compliance, for example, becomes more difficult as organizations scale
their AI solutions around the world, and cleaning and normalizing data Integrating data 36% -10%
becomes twice as difficult for leaders compared with non-leaders as Identifying trusted data 35% -2%
they leverage richer data sets. Identifying trusted data continues to be a
Identifying corrupt records 25% -6%
challenge as businesses turn to third-parties for external data, as does
ensuring data integrity as businesses work to ensure ethical algorithmic Size/frequency of data 22% 0%
decisions. In short, as AI gets used for more powerful business outcomes, Availability of data 17% -9%
the responsibility grows to meet ever higher standards.
Data silos in organization 16% -3%

Cleaning and normalization 10% 10%

0 5 10 15 20 25 30 35 40

Note: Multiple responses permitted.

Response base: 1,200
Source: ESI ThoughtLab/Cognizant
Figure 7
As businesses embrace a wider selection of data types, a
particular challenge is managing data perishability – ensuring
the data used for insights is current, accurate and relevant.
Unlike ERP and other structured data, more dynamic
data types – such as IoT, social, real-time, geospatial and
psychographic – come with a shorter expiration period. By
using machine learning, businesses can continually assess the
timeliness, accuracy and relevance of their data and analytic
models, testing millions of new models against real-world data
to continuously provide new scenarios and ranges of forecasts
for changing conditions.

As businesses embrace a wider selection of data types, a

particular challenge is managing data perishability – ensuring
the data used for insights is current, accurate and relevant.