B U SIN ESS IN N OVATIO N B RIEF

Maximize the
business value Place image here

of generative AI
Harness generative AI’s full potential
Executive summary
A surge of excitement across consumers and businesses—combined with Who is this content for?
broad, easy access to the technology—has led us to an inflection point
for generative artificial intelligence (AI). Today, all organizations should This innovation brief is designed for
evaluate the impact of generative AI and how they can leverage its business leaders seeking to better
transformative value.
understand generative AI—and learn
Most organizations recognize the importance of this moment and the how they can leverage it to improve
need to quickly develop a strategy to explore generative AI. But for many, business outcomes.
questions remain, including:

• What is generative AI?

• How is generative AI different from previous generations of AI?
• What are the main business use cases?
• Which customers are already using it?
• How should I start?
• What are the risks, and how can they be mitigated?

This business innovation brief provides an overview of generative AI,

outlining its capabilities, use cases, and business value. It also offers
valuable insights from Amazon Web Services (AWS) subject matter
experts, leveraging our extensive knowledge and experience in AI and
machine learning (ML) technologies.

2
Table of contents
Introduction: A new world of intelligence......................................................... 4

Understanding generative AI. . ........................................................................ 7

Business capabilities of generative AI.............................................................. 9

Business considerations for generative AI........................................................ 10

Executive insights on generative AI................................................................ 12

Common generative AI use cases by industry................................................... 13

How AWS can help you succeed with generative AI. . .......................................... 16

Next steps. . . . . . ........................................................................................... 19

3
I N T R O D U C T IO N

A new world of intelligence

Consumers and businesses alike are fascinated by generative AI’s ability to aspect of how companies and organizations operate.1 Gartner estimates that by
create sophisticated content, generate code, answer questions, and more—all 2025, 30 percent of outbound marketing messages from large organizations will
from simple natural language prompts, often within seconds. be synthetically generated.2

While a lot of attention has been given to how consumers are using generative Awareness of the technology is spreading at an unprecedented pace. According
AI, there is an even bigger opportunity in how businesses will use it to deliver to Fishbowl, which ran a survey among nearly 4,500 professionals in large
amazing experiences for their customers and employees. The true power of organizations in the US, 27 percent of professionals have already used
generative AI goes beyond a search engine or a chatbot and will transform every generative AI to assist with work-related tasks.1

“Rate of generative AI adoption in the workplace in the United States 2023, by Industry,” Statista, December 2022
4
1

2
“Beyond ChatGPT: The Future of Generative AI for Enterprises,” Gartner, January 2023
Seizing the opportunity
Organizations across industries are racing to seize the economic
opportunities that generative AI presents. If leading financial projections
prove accurate, the rise of generative AI is likely to usher in a new era of
the global economy. Global generative AI market

Market forecast to grow at a CAGR of 34.2%4

According to research by Goldman Sachs, generative AI

could increase global GDP by as much as 7%, or roughly
$7 trillion, over the next 10 years.3 USD 200.73B

These lofty financial projections are not driven by consumer interest

alone. The potential of generative AI to improve business productivity
and output accounts for just as much, if not more, of the excitement and
enthusiasm surrounding the technology.

For businesses of every size and across every industry, generative AI is a

revolutionary technology that is beginning to drive considerable value— USD 14.26B
and has the power to fundamentally transform the business landscape.

2023 2032

“Generative AI could raise global GDP by 7%,” Goldman Sachs, April 2023
5
3

4
“Generative AI Market,” Polaris Market Research, January 2023
The strategic imperative
Organizations across industries and around the world are looking
to leverage generative AI to reimagine customer experiences, boost
employee productivity, ignite creativity, and optimize business processes.

For most businesses, however, the path to achieving these

benefits remains unclear.

Many acknowledge the need to invest in generative AI—and to do

it quickly, lest their competitors gain the advantage. However, few
have succeeded in developing strategies for how they will adopt the
technology, where they will put it to use, or how they will achieve and
measure their results.

Read on to learn how your organization can start realizing the business
value of generative AI today—so you can keep pace with the market and
leapfrog your competition.

6
 M A C HI N E LE A R N I N G SI M P L E I N P U T S SI M P L E O U T P U T S

D E E P LE A R N I N G CO MPLEX INPUTS SI M P L E O U T P U T S

F O U N DAT IO N M O D E L S CO MPLEX INPUTS CO MPLEX O UTPUTS

Understanding generative AI
Before your organization can fully unlock the business value of generative AI, it’s map them to relatively simple outputs, for example, if the image contains a
important to have a fundamental understanding of how the technology works. cat or not. With generative AI, we can leverage massive amounts of complex
data to capture and present knowledge in more advanced ways—mapping
Generative AI is a term used to describe algorithms that can create new complicated inputs to complicated outputs, like summarizing a long document
content and ideas, including conversations, stories, images, videos, and music. and extracting the key insights.
Generative AI is powered by extremely large ML models that are pretrained on
vast amounts of data. These are commonly known as foundation models (FMs). Text-based generative AI systems use a specific type of FM called a large
language model (LLM). LLMs can perform a wide range of tasks that span
Traditional forms of ML allowed us to take simple inputs, like numeric values, multiple domains, like writing code, solving math problems, engaging in
and map them to simple outputs, like predicted values. With the advent of dialogue, and analyzing the information in documents to answer questions.
deep learning, we can take complicated inputs, like videos or images, and

7
Make data your differentiator Moments in generative AI history:

When you want to build generative AI applications that are unique to your business needs, Today’s FMs that are used to create generative
your organization’s data is your strategic asset. FMs can be customized and fine-tuned with AI applications are built atop a long history of
your organization’s proprietary data to deliver a more differentiated experience compared to AI innovation. Two of the earliest models with
an “out-of-the-box FM.” For example, a large grocery chain that tracks shopper preferences generative AI capabilities are the hidden Markov
can customize an FM to produce a better recommendation engine that is highly differentiated model (HMM) and the Gaussian mixture model
from competitors’ offerings. (GMM), both developed in the 1950s. HMMs use
known data to make educated guesses about
Organizations can also use customized FMs to easily create unique content that embodies unknown data (for example, predicting whether
their brand’s tone and style. For instance, a financial firm that needs to auto-generate a daily a card player is cheating based on their results).
activity report for internal circulation can customize an FM with proprietary data, including GMMs can examine a group of data (such as
past reports. The FM could then learn how these reports should read and what data was use a music playlist) and subgroups within that
to generate them – to deliver a report that better reflects the needs of the organization. data (for example, genres) to infer unknown
information (such as “this is a rap song”). Both
Now that you have a general understanding of how the technology works, let’s begin
are still used today.
exploring how you can put generative AI to work for your organization.

Text generation

Gaussian mixture model

Summarization

Information extraction
Pretrain Adapt

Q&A

Chatbot

Unlabeled data FM Broad range of general tasks

8
Business capabilities of Moments in generative AI history:

generative AI
Another early example of generative AI is
ELIZA, a chatbot (or “chatterbot,” as they
were previously known) developed by an
Organizations in many industries are using generative AI to enhance productivity and create MIT professor from 1964 to 1966. Like its
business value in many ways, including: namesake, Eliza Doolittle of Pygmalion and
My Fair Lady, the program grew more
sophisticated by “learning” from human
Code generation Virtual assistants interactions. ELIZA was most famously used to
Improve developer productivity by Enhance customer mimic the behavior of a therapist conducting
57% with AI coding companion experience with human-like an initial psychiatric interview, with the user
Amazon CodeWhisperer5 responses playing the role of the patient.

Contact center analytics Conversational search

Summarize and extract insights Extract insights from all your
from customer calls corporate information

Personalization Content generation

Improve personalized Create text, images, videos,
recommendations and and music
generate tailored content

Design and creativity

Get suggestions, generate
prototypes, and explore
innovative concepts

5
Data collected from a “productivity challenge” conducted by Amazon during the Amazon CodeWhisperer preview 9
Business considerations for generative AI
As you work to identify the capabilities of generative AI that are most useful to When evaluating FMs used to create generative AI applications, look for
your organization—and develop a strategy for implementing them into your models that offer:
business processes—you will need to determine which FMs to use in creation of
generative AI applications. 1. Easy ways to build and scale generative AI applications with security
and privacy built in
You should also carefully consider the infrastructure you will be using to 2. Performant, low-cost infrastructure to train your own models and run
support your FMs. Your models will benefit from a cost-efficient infrastructure inference at scale
that meets your requirements for performance.
3. Generative AI-powered applications to transform how work gets done
4. Data as your differentiator

10
Responsible AI, security, and privacy Moments in generative AI history:

With their vast size and open-ended nature, FMs raise new issues in defining, measuring, and In 2014, the development of the first generative
mitigating responsible AI concerns across the development cycle, such as accuracy, fairness, adversarial network (GAN) marked one of the
intellectual property (IP) considerations, hallucinations, toxicity, and privacy. For example, biggest breakthroughs in generative AI. In a GAN,
looking at the issue of fairness, can we ask an LLM to assign male and female pronouns at the two models (a “generator” and a “discriminator”)
same rate in reference to a doctor? Does that still apply if the prompt describes the doctor as compete in a zero-sum game. The generator
having a beard? And should we do the same for other professions? What about the Women’s manufactures content that appears increasingly
National Basketball Association (WNBA)? You can see that simply defining fairness in the “genuine,” while the discriminator analyzes its
context of an LLM is challenging and requires new approaches and solutions. opponent’s techniques to better identify fakes.
This novel approach of using AI to train other
Generative AI technology and how it is used will continue to evolve, posing new challenges AI proved revelatory, while GANs themselves
that will require additional attention and mitigation. To tackle these challenges and foster unlocked a new era for digital imagery.
innovation, academic, industry, and government partners are working together to explore
new solutions and concepts to ensure that generative AI continues to evolve in a responsible, In the following section, we feature an AWS
private, and secure way. industry leader who shares his experiences
and strategies for cloud best practices,
Data privacy and security are also critical to scaling generative AI responsibly. When it comes cultural change, organizational agility, and
time to customize and fine-tune a model, organizations need to know where and how their transformation with generative AI.
data is being used. They need to be confident their private data is not being used to train a
public model and that customer data remains private. Organizations need security, scalability,
and privacy to be baked in from the start to be viable for their business applications.

Read the blog post Responsible AI in the Generative Era

Learn more ›

11
Executive insights on
generative AI
Like any new technology that becomes mainstream, generative AI has a learning curve.
Executives are asking about where generative AI fits or doesn’t, how to use it effectively, and
about the non-technical considerations.
Phil Le-Brun
History teaches us that we see profound, positive change only when people, processes, skills, Director,
and culture are addressed alongside technology. Based on these learnings and from the AWS Enterprise
thousands of customers we talk to, my advice for those intrigued by generative AI is simple. Strategy

First, be curious. Learn what generative AI is, why it has captured people’s imaginations,
and what problems it can solve. Dive deep into areas such as the security of your data when
customizing models. And encourage others to learn these things, too, rather than delegating
them to your IT team.
Joining AWS in 2019, Phil’s experience
Second, think big and work backwards from the customer. This is a standard way of implementing technology at scale—
thinking here at AWS! Really understand the opportunities in your business—whether including a 25-year career spearheading
it’s an opportunity to improve supply chain efficiencies, invent new services, or enhance digital transformation efforts at McDonald’s
customer service. Fall in love with the problem before you jump to a solution, looking for Corporation—has allowed him to learn a variety
areas that reduce costs, increase resilience, or improve your top line. And think big about the of practical lessons. He shares this knowledge
opportunity; small thinking becomes a self-fulfilling prophecy. with enterprises to help them achieve their
cloud-based technology goals, such as
Finally, start now. Most business initiatives take time to get traction, so start experimenting supporting organizational agility and increasing
quickly. You will learn more from this than the endless planning and waiting for the customer centricity.
hypothetical perfect time that is typical of many business technology adoptions.

12
Common generative AI use cases
by industry
In previous sections, we’ve explored the capabilities of generative AI that can benefit all
businesses and how to select the right models and infrastructure for your business’ needs. Now,
let’s dive a bit deeper and examine generative AI use cases that are unique to specific industries.

Healthcare and Life Sciences • Improving clinical engagement and patient experience: Generative
AI can be used to improve patient navigation, care coordination, and
• Accelerating innovations in pharmaceutical R&D: Life sciences the productivity of healthcare providers. Through conversational
organizations can expedite the development of new, more effective drug interactions, clinicians can access data from electronic health records,
candidates with greater efficiency. For example, researchers can predict the research publications, and medical policies more easily. Generative AI
structure of proteins to better identify biological targets, generate novel can also accelerate digital transformation and personalized medicine by
amino acid sequences, and identify docking sites for specific targets to facilitating intelligent document processing (IDP), providing personalized
create precision therapeutics. Ultimately, this will result in more targeted guidance on medical and pharmacological practices, and helping to design
and effective therapeutic options for patients. effective clinical trial protocols to ensure safety and efficacy for new drug
development.

• Summarizing health and scientific data: Researchers and clinicians can

leverage generative AI to reduce time to insights and quickly find the data
they need for more data-driven decision making. In healthcare, search
and summarization can be used to enhance efficiencies by, for example,
auto-generating chart note components and note summaries, simplifying
workflows, and automating administrative tasks. The applications in life
sciences are vast, including the identification of patterns in clinical trials,
the ability to summarize findings from hundreds of thousands of research
papers, and the identification of efficiencies across the value chain.

13
U SE C A SES BY I N D U ST RY

Financial Services Automotive and Manufacturing

• Improving experiences: Financial services firms can better serve customers • Improving product design: Manufacturers can use AI to optimize the
and employees by deploying chatbots that resolve problems faster and design of mechanical parts—or create entirely new material, chip, and part
personalizing products and recommendations, like leveraging a repository designs—improving quality and durability, lowering costs, and simplifying
of customer interactions with summarization and queries; AI also enables production.
the creation of knowledge articles from disparate internal sources. • Creating new in-vehicle experiences: Virtual assistants and personalized
• Increasing knowledge-worker efficiency: Knowledge workers at financial route recommendations can enhance experiences for drivers and
services firms can quickly draft investment research, loan documentation, passengers.
insurance policies, regulatory communications, RFIs, and business • Improving testing and maintenance: AI can improve product testing by
correspondence; achieve deeper insights into customer behavior; improve generating information missing from datasheets—and unlock new assisted
collaboration; and increase the business value of unstructured content. maintenance use cases to better maintain and service machinery, including
• Analyzing market sentiment: Through faster and more thorough analysis products in use by consumers.
of social media, news articles, and financial data, financial services firms can • Enhancing overall equipment effectiveness for factories: Digitize and
surface market commentary, identify opportunities sooner, and proactively capture historical machine maintenance data, repair data, equipment
mitigate risks. manuals, production data, and potentially data from other manufacturers
• Building new products and automating business processes: Financial to generate suggestions for maintenance, repairs, or equipment
services firms can create on-demand structured data products from parameters—and improve productivity, availability, and quality.
large unstructured data sources, and generative AI can enhance productivity
tools for end users and improve developer productivity with automatic
code generation.

14
U SE C A SES BY I N D U ST RY

Education Media and Entertainment

• Summarizing text: Students and teachers can create concise summaries • Expediting content creation: From concept design and storyboarding to
of research documents, lecture transcripts, and class notes to make them post-production workflows, media and entertainment (M&E) companies can
easier to search and browse. automate lower-level tasks to increase production speed and allow creative
• Improving automation: AI can transform information into sample test talent to iterate faster and realize the director’s vision.
questions, accelerate grading, measure student performance across a wide • Enhancing music: Artists can complement and enhance their albums with
range of factors, and provide personalized feedback and recommendations AI-generated music to create new genres.
to teachers and students. • Improving the media supply chain: Generative AI applications can aid or
• Personalizing learning environments: Educators can create personalized automate tasks like localization, content moderation, artwork creation for
learning pathways for student segments—or even individual students—and promotional materials, and even content restoration.
leverage simulations and virtual reality to make learning more engaging.

Moments in generative AI history:

The 2017 introduction of a new type of deep learning model, the transformer, language processing (NLP), where understanding the full context of the input
set the stage for modern generative AI. Unlike older models, which break is critical. In 2018, OpenAI took the technology further by creating the first
down input data, process it, then put the pieces back together, transformers Generative Pre-trained Transformer (GPT). From there, OpenAI developed its
process the entire input all at once. This makes them ideal for natural GPT-2 engine in 2019—which it then used to power ChatGPT, introduced in
late 2022.

15
How AWS can help you succeed
with generative AI
You can unlock the full business value of generative AI for your organization with AWS. More than
Reinvent your applications, create entirely new customer experiences, drive unprecedented
levels of productivity, and ultimately transform your business.
100,000
Experience and expertise customers currently
use AWS for AI and ML
One of the key advantages of AWS lies in a rich AI heritage built over two decades of focused
investment. In fact, more than 100,000 customers currently use AWS for AI and ML.

Amazon, the driving force behind AWS, harnesses ML capabilities to power its ecommerce
recommendations engine, optimize robotic picking routes in fulfillment centers, and much
more. Further, ML informs Amazon’s supply chain, forecasting, and capacity planning.

Deep learning is also employed in the Amazon Prime Air drone delivery system and the
computer vision (CV) technology behind Amazon Go, the innovative retail experience that
allows customers to select items and leave the store without traditional checkouts. And Alexa,
which is supported by more than 30 different ML systems, helps customers with a wide array
of tasks billions of times each week.

With thousands of dedicated ML engineers, AI and ML are deeply ingrained in the heritage of
Amazon and AWS—continuing to shape the future.

16
Why build with AWS? Data as your differentiator

With AWS, it’s easy to use your organization’s data as a strategic asset
to customize FMs and build more differentiated experiences. Data is the
Organizations of all shapes and sizes choose to build generative AI and other difference between a general generative AI application and one that truly
AI and ML applications on AWS for many reasons. Here are some of the top knows your business and your customer. And with the most comprehensive
advantages of building on AWS, according to our customers: set of data and AI services, you can securely customize an FM on AWS with
your data and build a model that is an expert on your business, your data,
The easiest way to build and scale generative AI and your customers.
applications with security and privacy built in

Amazon Bedrock is the easiest way for customers to build and scale Generative AI-powered applications for the enterprise to
generative AI–based applications using FMs. Bedrock makes Amazon Titan transform how work gets done
FMs and models from leading AI companies such as AI21 Labs, Anthropic,
AWS is building powerful new applications that transform how our
Cohere, Stability AI, and Meta accessible via an API. Customers using Bedrock
customers get work done with generative AI. Boost productivity with
can leverage the benefits of AWS, which is architected to be the most flexible
purpose-built conversational agents that streamline coding in the enterprise
and secure cloud computing environment available today. Agents for Amazon
with Amazon CodeWhisperer, simplify business intelligence with Amazon
Bedrock is a fully managed capability that makes it easier for developers to
QuickSight Generative BI, and improve clinical efficiency for healthcare
create generative AI applications that can deliver up-to-date answers based
organizations with AWS HealthScribe. With security, privacy, and responsible AI
on proprietary knowledge sources and complete tasks for a wide range of
at the forefront, easy customization, and integration into your existing
use cases.
data sources and applications, enterprises can quickly take advantage of
generative AI without the heavy lifting.
The most performant, low-cost infrastructure for
generative AI

For years, AWS has invested in developing silicon that delivers the highest levels
Further reading on responsible AI:
of performance and cost optimization for AI and ML workloads. The results—
AWS Trainium and AWS Inferentia—deliver the lowest costs for training AWS responsible AI resource hub ›
models and running inference in the cloud. AWS has also developed Amazon
Elastic Compute Cloud (Amazon EC2) instances to help you take advantage eBook: Democratized, Operationalized, Trusted:
of these capabilities. For example, Amazon EC2 Trn1 instances powered by The 3 Keys to Successful AI Outcomes ›
Trainium save you up to 50 percent on training costs,6 while Amazon EC2 Inf2
instances powered by AWS Inferentia2 deliver up to 40 percent lower cost
per inference.7

Over other comparable Amazon EC2 instances

17
6

7
Compared to prior generation AWS Inferentia–based instances
AWS generative AI services
Facilitate your generative AI applications with a range of AWS technologies, including:

AWS Trainium: Train models faster with up to 50% cost savings8 using this
ML model accelerator
Amazon Bedrock ›
AWS Inferentia2: Run high-performance FM inference with up to 40% lower
Build and scale generative AI applications with FMs.
cost per inference using this accelerator 9
Bedrock supports a variety of FMs, including:
Amazon CodeWhisperer: Enjoy 57% faster application development10 while
• Amazon Titan: For text summarization, generation, classification, helping to ensure security with this AI coding companion, which is at no cost
open-ended Q&A, information extraction, embeddings, and search for individual use

• AI21 Labs Jurassic-2 Multilingual LLMs: For text generation in Amazon QuickSight Generative BI: Transform traditional multistep
various languages business Intelligence (BI) tasks into intuitive and powerful natural language
• Anthropic Claude 2: LLM for conversations, question answering, experiences with Generative BI capabilities in Amazon QuickSight
and workflow automation based on research into training honest
Amazon SageMaker: Build your own FMs with managed infrastructure and
and responsible AI systems
tools to accelerate scalable, reliable, and secure model building, training,
• Stability AI Stable Diffusion: Generates unique, realistic, high- and deployment
quality images, art, logos, and designs
• Cohere Command + Embed: Text generation model for business Amazon SageMaker JumpStart: ML hub that provides access to algorithms,
applications and embeddings model for search, clustering, or models, and ML solutions so you can quickly get started with ML. With
classification in over 100 languages SageMaker JumpStart, ML practitioners can choose from a broad selection
of publicly available FMs. ML practitioners can deploy FMs to dedicated
• Meta Llama 2: Pretrained and fine-tuned LLMs for natural language
SageMaker instances from a network-isolated environment and customize
tasks like question and answering and reading comprehension
models using SageMaker for model training and deployment.

AWS Trainium delivers up to 50 percent cost-to-train savings over comparable Amazon EC2 instances
18
8

9
AWS Inferentia delivers up to 40 percent cost per inference over comparable Amazon EC2 instances
10
Data collected from a “productivity challenge” conducted by Amazon during the Amazon CodeWhisperer preview
Next steps
Now that you have a better understanding of generative AI, what it can do, Collaboration with experts is highly recommended to ensure you consider
and its potential business benefits, the next step is to clearly define your factors such as data availability, data quality, and ethical implications related
objectives and identify use cases for leveraging it. It’s best to start with smaller to generative AI. Furthermore, infrastructure considerations should not be an
experiments and simple, precise goals. Once you’ve achieved some quick wins, afterthought, as they can significantly impact costs, scalability, and energy
you can begin scaling your efforts upward and outward. consumption. Engaging with AWS experts can provide valuable guidance
throughout the decision making process and stages of implementation.

The time is now

The dramatic rise of generative AI brings us to a tipping point. FMs grow
more sophisticated and powerful every day. For organizations, it’s the
power to transform business by creating entirely new customer experiences
and driving unprecedented levels of efficiency and productivity.

All of which leads to an indisputable fact: To compete in this new era of

profound technological advancement, every organization needs to consider
making generative AI a part of its innovation road map.

With the most cost-effective cloud infrastructure for generative AI; a host
of AI products, services, and solutions; and years of trusted AI expertise,
AWS can help turn the promise of generative AI into results for your Partner with AWS to accelerate your
organization. generative AI journey today.

Get started ›

© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
19