A RTIFICI A L IN TELLIG E N CE:
Step-by-step
guide for
startup founders
A practical guide to help make
your AI vision a reality
I N T R O D U C T IO N
Forging ahead Generative AI could add
$2.6
TRILLION
When deployed with the right strategy, generative artificial intelligence (AI) has the potential
to change the anatomy of work. By streamlining work activities that absorb 60–70 percent of
employee time, startups can improve productivity down to the individual worker.1 This
can fast track product innovation, increase agility, streamline processes, boost revenue, and
to
allow for better, faster decision making.
$4.4
The lack of a single universally accepted playbook for generative AI success is keeping some
founders on the sidelines, unsure of how to take the next (or even the first) step on the AI TRILLION
journey. By democratizing generative AI, Amazon Web Services (AWS) is helping startups in economic benefits annually
at any stage overcome barriers to adoption and forge ahead with confidence. This eBook
outlines a proven path—from taking the first step to measuring results—with insights from
The impact of generative AI on
AWS best practices and its experience helping thousands of customers, including leading
productivity could add trillions of dollars
startups like yours, realize their own initiatives.
in value to the global economy, increasing
the effect of all AI by 15%–40%1
“The economic potential of generative AI: The next productivity frontier,” McKinsey Digital, June 2023
2
1
What are artificial
intelligence,
machine learning,
and generative AI?
AI is a way to describe any system that can replicate tasks that previously
required human intelligence. Almost all AI systems are built using
machine learning (ML). ML utilizes large amounts of data to create
and validate decision logic, forming the basis of an AI model. The AI
application then feeds input data into that model, and the model
outputs human-like decisions. The rapid advancement of ML, massive
proliferation of data, and easy availability of scalable compute capacity
can help businesses accelerate how they use AI and, particularly, how
they leverage generative AI.
Generative AI is a type of AI that can create new content and ideas,
including conversations, stories, images, videos, and music. Like most
AI, generative AI is powered by ML models—very large models that
are pretrained on vast amounts of data and commonly referred to as
foundation models (FMs).
3
Why generative AI?
Before diving into the steps of your generative AI journey, let’s explore
why any startup should begin its journey in the first place. According
to Goldman Sachs, generative AI could drive a 7 percent (or almost
$7 trillion) increase in global GDP and lift productivity growth by 1.5
percentage points over a 10-year period.2 Globally, we have reached an
inflection point where most customer experiences and applications will be
reinvented with generative AI.
Developing a strategy that includes generative AI initiatives is imperative
to ensure successful business outcomes. Even after completing the steps
outlined in this eBook, a business will need to regularly remind itself what
it’s working toward—staying focused on the precise business benefits that
can be unlocked by fully leveraging generative AI technology.
“Generative AI could raise global GDP by 7%,” Goldman Sachs, April 2023
4
2
Startups are already realizing the impact of: experiences. AWS customers have improved their own customer experiences
through the effective use of chatbots, virtual assistants, personalization, and
content moderation.
Ideation and developing new products faster
Generative AI applications can automate manual aspects of product
Transforming content creation for greater creativity
development, such as routine design, analysis, and writing tasks. It can also
generate dozens (and, in some cases, hundreds or even thousands) of ideas Startups, like game-industry-enabler Scenario, are using generative AI
from simple text prompts, all within seconds, allowing you to fast-track the to turbocharge real-world production processes across multiple types of
entire innovation process from ideation to go-to-market. creative content—including art, music, text, images, animations, and video.
The AI-powered capabilities for automating writing, media design, and
character modeling allow for unprecedented creative exploration and speed.
Boosting employee productivity
Generative AI is being leveraged for its transformative value to help AWS
Driving outcomes for real, competitive advantage
customers reach new levels of productivity. The latest advancements in
generative AI can be used to increase employee productivity. The technology Regardless of a startup’s size, field, and domain, generative AI should be
can help you accelerate your team’s performance to make the most of embedded into every business plan to stay relevant and competitive. AI21
headcount and limited resources. Labs (AI21) built a midsize state-of-the-art large language model (LLM),
deploying Amazon SageMaker to train the FM in under 20 days.3 This
midsize model would preserves the quality of text generation, making it
Improving customer experiences
nearly the same as AI21’s largest model at a much lower inference cost to
Startup founders can take advantage of generative AI to improve customer AI21 and its customers.
engagement, increase personalization, and attract new users through deeper
Now that we’ve outlined the why of generative AI,
it’s time to explore the how
The next sections will demonstrate the steps in the AI journey using AWS best practices
and those of AWS startup customers to exemplify the necessary changes that must take
place in order to successfully implement, deploy, and scale AI solutions.
“AI21 Labs Accelerates Generative AI Model Adoption Using Amazon SageMaker,” AWS Customer Story, 2023
5
3
ST E P 1 “With tens of thousands of teams
relying on Coda AI to not only
Make data your differentiator finish tasks but make progress in
their work, it’s important to bring
Like with ML, data is an important piece of generative AI. While general LLMs can be used the best of security, privacy, and AI
“out of the box” for some use cases, when you want to build generative AI applications
quality to their enterprise data and
that are unique to your business needs, your startup’s data is your strategic asset. Data is
the difference between a generic generative AI application and one that truly knows your
user experience. Amazon provides
business and your customer in order to deliver a better, more differentiated experience. secure storage in S3 all the way
through serving AI responses
While some companies will build and train their own LLMs with vast amounts of data, many
with Bedrock, keeping our LLMs
more will use their organizational data to fine-tune FMs for their unique business needs or to
add context to prompts through Retrieval Augmented Generation (RAG). For example, you inside AWS, offering scalability
can use data from sources such as your data lake, database, and data warehouse to create and dependability. Coda AI can use
a chatbot that can provide technical support for your products or a model that provides this to create curated content and
marketing copy trained on your best-performing ads. insights—and all of it stays within
Because of this, success with generative AI requires relevant, high-quality data, which AWS’s secure platform.”
means that you need a strong data strategy in the cloud. According to McKinsey Digital,
David Kossnick, Product Lead, Coda AI
“...companies that have not yet found ways to effectively harmonize and provide ready
access to their data will be unable to fine-tune generative AI to unlock more of its
potentially transformative uses.”
The right data strategy for generative AI includes a comprehensive set of services to store
and query data at scale, breaks down silos so you have ready access to all of your data to
leverage for generative AI applications, and makes sure your data is secured and governed
throughout the lifecycle of building generative AI applications.
6
How Easebuzz did it
Easebuzz provides a comprehensive online payment solution platform for individuals “Our AWS data lake architecture has
and businesses. With its expansion, it became essential to extract value from its data to significantly added value to our platform. It
provide data-driven features to customers. To accommodate the number of logs ingested raises our ability to derive actionable insights,
daily, Easebuzz built a data lake on AWS.
optimize operations, and deliver a superior
Implementing a data lake reduced data storage expenses by 60 percent. Instead of payment experience to our customers.”
paying an hourly rate, the company now only pays for the data it stores in Amazon
Amit Kumar, CTO, Easebuzz
Simple Storage Service (Amazon S3). The data lake delivers analytics services to
customers, offering valuable insights into business performance and the ability to
identify emerging trends. Customers can obtain a highly detailed report that contains
three years of data in under three minutes. Report consumption has increased by 15
percent as customers are accessing on-demand daily dashboard reports 89 percent faster.
Read the customer story ›
7
ST E P 2
Finding the right business problem to address
One mistake startups often make in their AI journeys is to let their technical In a successful AI journey, you should start by creating a dream team to
team work in silos to build models as proofs of concepts rather than solve real address your specific business problem. This requires including both technical
business problems. With no specific business problem to solve, your team will and domain experts within your team. While the technical experts will take on
find it increasingly difficult to unlock the benefits of AI. This can stall or even the brunt of model creation, they need the field knowledge of domain experts
stop progress on AI initiatives. to define precise business challenges and identify the data most important to
finding a solution.
Here are some important questions you should ask before embarking on an
AI journey: Your team should also work through how to measure success. “Make sure
you have very crisp and clear metrics as you embark on the machine learning
journey,” Freshwater said. “Many times, your models are taking over for
1. Does it solve a real business problem? something existing, and you want to make sure that they’re actually better
and that you can measure it.”
2. Will the project benefit from AI?
3. Can it eventually be operationalized? For more on measuring the success of AI initiatives, refer to Step 5 in this eBook.
You can also leverage the AWS Generative AI Innovation Center to work
backwards from business challenges and go step-by-step through the process
The AWS AI Use Case Explorer is a business outcome-centric search and of creating AI projects and initiatives.
navigation site that enables users to find the right AI use cases and discover
relevant customer success stories.
8
ST E P 3
Upskilling teams
In parallel with creating a comprehensive data strategy, today’s startups
must focus on arming their team with the right skills to succeed in the
era of generative AI. However, startups across dozens of industries are
growing increasingly aware of an expanding skills gap—the separation
between technologies and the ability of internal technical and business
specialists to take full advantage of them. It’s an issue that should set off
alarm bells in light of recent research from the World Economic Forum,
which found that more than 75 percent of organizations plan to adopt AI
in the next five years.4
To help individuals train for the AI jobs of the future, AWS released
on-demand skills training to support those who want to understand,
implement, and begin using generative AI. Amazon has also designed
training courses specifically for developers who want to use Amazon
CodeWhisperer.
While there’s no one-size-fits-all solution to the AI skills gap, there are
proven methods that can maximize the abilities of your team, reducing
the need to make large investments in buying or borrowing pretrained
expert talent.
“The Future of Jobs Report 2023,” World Economic Forum, April 2023
9
4
How Slalom did it
As a purpose-led, global business and technology consulting company,
Slalom helps customers build practical end-to-end solutions that drive
meaningful impact. To continue delivering powerful solutions and drive
business growth, Slalom looked to AWS Training and Certification and
AWS Certified Machine Learning – Specialty, which validate expertise
in building and deploying ML models on AWS.
Slalom’s experience using the AWS Training program has advanced
its consultants’ professional development and growth, which has also
driven sales. In the six months after one learning cohort concluded in
2022, 50 percent of participants drove more sales than they did in the
six months before.
Read the customer story ›
“Working with AWS Training and Certification empowers
our team members to learn new, valuable skills that can
help them grow and advance their career journeys. We
try to align the needs of our consultants with where our
customers’ strategic objectives are headed.”
Jeff Pearson, Managing Director, Slalom
10
ST E P 4
Scaling beyond pilot projects
After the first few successful pilots, you must take the next step on the journey: sustainably
scaling AI across your business. This is both a technical and a cultural challenge. There are
several ways startups approach the cultural shift necessary to scale AI. Some might find
success by creating a center of excellence that rallies the community and continues to
push for new initiatives. Or, like Amazon, you can make AI an integral part of your yearly
planning processes, continuously bringing domain and technical experts together to
brainstorm and determine their next steps.
Achieving scalability requires businesses to help their developers use AI. Building models at
scale can be labor intensive, which can slow innovation. With Amazon Bedrock serverless
experience, you can get started quickly to privately customize FMs with your own data and
easily integrate and deploy them into your applications using AWS tools and capabilities
you already know (including integrations with features like pipelines to manage your FMs
at scale). Many startup customers are solving scalability with SageMaker to prepare data
and build, train, and deploy models to and get them into production faster and at lower
cost, enabling sustainable expansion of AI initiatives beyond pilot projects.
Many more businesses are scaling through AWS AI services, a set of pretrained and
managed services that can be used as building blocks to address common use cases,
including personalizing recommendations, improving safety and security, and increasing
customer engagement.
11
How Fibe did it
Fibe, one of India’s leading consumer lending apps, was looking to
optimize its digital lending platform for the growing Indian market. To do
this, the startup was looking to deeply integrate ML into its platform to
offer faster and more convenient services to its customers.
To enhance the loan application process, Fibe implemented Amazon
Textract, which utilizes optical character recognition (OCR) technology
to extract text and data from Know Your Customer (KYC) documents.
To reduce administrative complexity, Fibe implemented Amazon
Comprehend, a natural language processing service, to extract relevant
information from income statements and other application documents.
Using AWS Lambda, Fibe streamlines development and maintenance
processes. This service automatically handles code execution requests
and simplifies processes like scaling instances. With AWS support, Fibe is
spending less on IT, which directly enhances its profits.
Read the customer story ›
“Our goal is to utilize AI more effectively in order to
enhance the chatbot’s capabilities. We continue to rely
on AWS for innovative solutions. ... Our relationship
with AWS keeps our entrepreneurial spirit alive.”
Anil Sinha, CTO, Fibe
12
ST E P 5
Measuring the results
When measuring the results of AI efforts, the traditional “project ROI” viewpoint—where a
project has defined start and end points, a budget, and a return—is reductive and can be
detrimental to the initiative’s success. If the project doesn’t generate a positive return within
the given time frame, you and your team may decide to move on and miss out on critical
opportunities down the line. Startups alike must measure AI efforts based on what success
means for their business with regard to the processes being optimized. In addition, they must
view AI efforts as long-term investments, acknowledging that a true “return” may not be
realized for several years and after countless iterations.
When planning AI initiatives, it’s better to view the process through the lenses of agility,
competitive advantage, or risk tolerance rather than “expected” return. A founder will have
greater success if it disregards the question of “What will be my return on investment in X
months?” in favor of something more like “If we don’t invest in this now, will we fall behind
our competitors in X years when the technology matures?” While traditional ROI metrics
may not be the best approach, the business impact of AI initiatives can still be measured—it
just requires a different outlook. AI results can be measured through something resembling
a “value tree,” where the main trunk of the tree represents the traditional revenue return
and branches extending from the trunk recognize the value of other business outcomes. The
specific branches of the value tree will depend on the business, the industry, and the initiative.
But they might be things like time saved through automated processes, new leads, markets,
and opportunities identified, customer service improvements, or increases in upsells.
Measuring the success of AI through a more holistic and long-term model will keep startups
focused on the best outcomes for their business future.
13
How AI21 Labs did it
AI21 Labs (AI21) is a leader in generative AI and LLMs. Initially, the
company released two models: one with seven billion parameters and
another with 178 billion parameters. But its team saw an opportunity
to offer customers an LLM of 17 billion parameters to bridge the gap
between the existing sizes.
AWS trained the FM in under 20 days using SageMaker—saving several
weeks of time compared with AI21’s previous training methods. “Because
Amazon SageMaker handles node failures, restarts elegantly, and
orchestrates large, distributed runs, the team working on pre-training
the model can focus on core tasks,” says Dan Padnos, vice president of
platform at AI21.
Read the customer story ›
“We have a really good relationship with the AWS
team—[its members] have gone deep into the
technical details with us and collaborated on
challenging tasks. Throughout the process, the AWS
team has been creative and has had awareness about
our challenges and goals.”
Dan Padnos, VP of Platform, AI21 Labs
14
Taking the next step with AWS
No matter where your startup is in its AI adoption journey, you can take the next step
with AWS solutions built on the most comprehensive cloud platform and optimized for
generative AI with high performance computing (HPC), security, and analytics. Featuring
the world’s broadest and deepest set of AI services, over 100,000 customers are running
their AI workloads on AWS. Generative AI can help you unlock new business value for
your startup. From reinventing customer experiences to enhancing productivity and
accelerating growth, generative AI holds the power to help you transform your business.
Generative AI with AWS, by the numbers
100,000+ 20+ YEARS
customers using AWS for of building experience
their AI workloads at Amazon
15
Improve customer experiences Enhance creativity and content creation
Chatbots and virtual assistants: Streamline customer Marketing: Create engaging marketing content, such as
self-service processes and reduce operational costs by blog posts, social media updates, and email newsletters, to
leveraging generative AI–powered chatbots, voice bots, save time and resources.
and virtual assistants to automate responses to
customer queries. Sales: Generate personalized emails and messages based
on prospect’s profile and behavior, improving response
Agent assist and conversational analytics: Enhance rates. Generate sales scripts or talking points based on the
agent performance to improve first contact resolution customer’s segment, industry, and product or service.
and augment tasks such as knowledge search, call
summarization, and problem-solving. Managers can extract Product development: Generate multiple design
valuable insights to improve customer experience, monitor prototypes based on certain inputs and constraints,
agent performance, and boost business performance. speeding up the ideation phase or optimizing existing
designs based on user feedback and specified constraints.
Boost employee productivity Accelerate process optimization
Conversational search: Improve employee productivity Document processing: Improve business operations
by quickly and easily finding accurate information and by automatically extracting and summarizing data from
summarizing content through a conversational interface. documents and insights through generative AI–powered
question and answering.
Code generation: Accelerate application development
with code suggestions based on the developer’s comments Data augmentation: Generate synthetic data to train ML
and code. models when the original dataset is small, imbalanced,
or sensitive.
Automated report generation: Generate financial
reports, summaries, and projections, saving time and
reducing errors.
16
Solving the biggest Challenge Solution
artificial intelligence Discouraging Developing a fault-tolerant culture
challenges failures
Siloed, Creating a modern data strategy that
Most startups have made investments and progress in their unprocessed data includes data lakes
AI journeys and are exploring the possibilities of generative
AI. But many find themselves hitting stopgaps along the way,
Finding the right Building blended teams that include both
worried that costs and complexities will grow too high as
business problems technical and domain experts
they progress.
Throughout this eBook, we explored the steps to forge ahead
The AI skills gap Adopting new organizational models,
and realize the full power of generative AI. To recap, let’s
processes, and team management
look at the biggest challenges we identified along the way,
philosophies
along with a brief recommendation of how your startup can
solve them.
Sustainably scaling Leveraging end-to-end tools like Amazon
beyond pilot Bedrock and SageMaker to build and scale
projects generative AI applications
To learn more about how you can overcome
obstacles and accelerate your AI journey, visit
Measuring the Forgoing traditional ROI metrics in favor
the AWS AI Resource Hub.
results of agility, competitive advantage, and risk
tolerance using the value tree model
Get started ›
Explore generative AI ›
©️ 2024, Amazon Web Services, Inc. or its affiliates. All 17
rights reserved.
