Democratized,

operationalized,
responsible:
the 3 keys to successful
AI and ML outcomes
Leverage the full power of artificial intelligence
and machine learning
Table of contents
PA R T I – T H E M A R K E T

Introduction. . ............................................................................................. 4

Overview.. . . . . . . ............................................................................................ 5

Go time. . . . . . . . . . ............................................................................................ 6

PA R T II – T H E K E Y S T O S U CCESS

Key #1: Democratization............................................................................... 8

Key #2: Operationalization........................................................................... 11

Key #3: Responsible AI................................................................................. 13

Successful outcomes start here...................................................................... 15

2
PA RT I – T H E M A R K E T

The rapid development of AI and ML

technologies is fueling ever-growing
investments and innovations

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

Why artificial intelligence and

machine learning matter more
than ever
Artificial intelligence (AI) and machine learning (ML) are among the most transformative
technologies of all generations. That helps explain why AI and ML adoption is on the agenda of
so many business and technical leaders. Organizations everywhere are investing in AI and ML
for a variety of reasons. Most notably, they fuel rapid innovation. They also drive competitive
advantage by adding intelligence to existing processes, automating time-intensive manual
tasks, and accelerating the creation of new products and services.

Although specific estimates on the growth of AI and ML vary, there is no debate about the
impact of these technologies on the global economy. Gartner predicts that the size of the AI
and ML market will reach $134.8 billion by 2025, considerably outpacing the growth of the
worldwide software market. Likewise, IDC forecasts that AI and ML investments will jump
from $85.3 billion in 2021 to $204 billion in 2025, resulting in a compound annual growth
rate (CAGR) of 24.5 percent for the 2021–2025 period. The effects of these numbers
are (potentially) staggering—both for the organizations that make the investments and for
those that do not.

Whether you are expanding your existing AI and ML initiatives or just getting started, a clear
reference guide can help you develop your strategy and help ensure successful outcomes. This
eBook outlines the three strategic pillars to success and provides practical recommendations that
you can apply to your organization.

4
O V E R V IE W

What AI and ML
mean for business
The language of AI and ML reflects distinctions between the technologies themselves.
Understanding these distinctions will help you know how to talk about the relevant technologies
with those both inside and outside of your project teams.

At the most basic level, AI is a way to describe any system that can replicate tasks previously
requiring human intelligence. Most AI systems look at probabilistic outcomes based on incoming
real-world data to predict the most likely scenarios.

AI systems may use any combination of techniques to arrive at predictive decisions, including
“if/then” statement logic. However, most AI systems are created using ML, which ingests large
amounts of data to create and validate decision logic (also known as a model).

Deep learning is the part of ML that uses a technique known as deep neural networks to mimic
the way biological neurons signal to one another. Deep learning is typically used in complex AI
systems, such as autonomous vehicles or natural language processing (NLP).

A subset of deep learning called reinforcement learning is a technique that enables the system
to learn in an interactive environment by simulated trial and error using feedback from its own
actions and experiences.

5
GO TIME

Considering the recent rapid growth of

artificial intelligence and machine learning

AI and ML have experienced highs and lows since emerging in the latter AI and ML models can leverage much of the world’s available data. These
half of the twentieth century, but the technologies have become firmly models are equipped to inform predictions and enable the inclusion of video-,
established and market-facing over the past decade. A primary factor audio-, image-, and text-based data in a range of applications—from contact
contributing to the recent upsurge in ML is the unprecedented utility of the center intelligence and document processing to personalization and content
cloud. Cloud computing brings together robust AI and ML services along with moderation. For example, think of the cancer specialist seeking patient
high performance computing (HPC), such as CPUs and GPUs, networking, insights from X-rays, MRIs, or even handwritten prescriptions. Neither apps nor
and storage for model training and deployment. Additionally, the modern analytics can reliably extract predictive information from this kind of data; the
cloud continually protects the data and applications to support rigid security, physician (and her patient) need advanced intelligence for that.
compliance, and privacy requirements. These and other cloud-enabled
building blocks make ML available to any organization. With the help of AI and ML technologies, decision makers can now uncover
patterns in unstructured data by utilizing AI and ML to decipher video and
The continued proliferation of unstructured data (which makes up over 80 image content. The tools are ready, and the time is right to open the door to
percent of the world’s data) has also helped ML gain converts and momentum. meaningful AI and ML innovation. Now, all you need are the keys to delivering
ML is unique in that it can use unstructured data to make predictions, unlike successful outcomes.
traditional analytics, which usually rely on structured data in a tabular format.
To support the explosive growth in data, cloud services provide efficient
ways to access, process, analyze, and store any amount of data. That makes it
possible to realize value from the massive volumes of data created daily.

6
PA RT II – T H E K E YS TO SU CCESS

3 important keys can help

customers bring artificial
intelligence and machine
learning to more parts of
their business

7
KEY #1

Democratization
An article titled “Data Scientist: The Sexiest Job of the 21st Century” ran in the October 2012 “Recruiters are struggling to
issue of Harvard Business Review. Prescient for sure, but the article may have underestimated find data scientists that can
the intensity of the demand a decade later—a global need for experienced data scientists that
help them move their digital
continues to spike repeatedly amid the growing interest in ML applications. Case in point: a
295 percent annual growth rate makes data science the fastest-growing skill set on ZDNET, as
ambitions forward.”1
measured by technology recruiters. ZDNET, 2022

In response, many organizations are embracing a different approach. Instead of relying strictly
on the recruitment of data scientists, they are using tools to empower the people they’ve already
hired. By working with existing employees, they are expanding the capabilities and experiences
of workers while reducing the need to compete with other companies for the relatively few data
scientists who might be available at any one time.

This democratization of AI and ML tools and responsibilities is being widely embraced. It’s one of
the three keys to making ML accessible to the many in a scalable way. Organizations that create
a formal program for non-coders—such as line-of-business analysts supporting sales, finance,
operations, marketing, and other key lines of business—will be far more agile than those that do
not. In other words, democratization is good for business.

One way to democratize ML is through no-code tools for business analysts. No-code tools
that allow business analysts to make ML predictions even without ML experience are used for
common use cases, including churn prediction, lead conversion, price and revenue optimization,
delivery time predictions, inventory planning, and more. No-code tools are a way to automate
and eliminate dependencies on data science teams. They also give business analysts and data
scientists the option of working together and creating opportunities to collaborate and create
feedback loops to refine models further over time.

Hughes, O., “Employers Are Desperate for Data Scientists as Demand Blooms,” ZDNET, 2022
8
1
Then again, what’s good for business analysts is great for data scientists.
Using automated ML, or AutoML, data scientists can automate each step
of the ML workflow. It democratizes the process by removing the tedious,
iterative, and time-consuming work across the ML workflows—from
data acquisition to model operationalization—so data scientists can
spend less time on low-level details and more time on improving model
accuracy. AutoML tools handle sourcing and preparing data, engineering
features, training and tuning models, deploying models, and ongoing
model monitoring and updating. AutoML techniques ultimately reduce
the burden on data science teams, liberating experts from sacrificing their
valuable time on repetitive, low-value tasks.

Democratization also includes support for the developers creating and

maintaining critical business applications. While these developers will not
typically build, train, and deploy their own models, they will be able to use
pre-built models that can easily be integrated into their applications with
a single API to add AI capabilities like computer vision, speech recognition,
or personalization across a wide range of business applications.

Finally, AI and ML are becoming more readily available thanks to smart

processes and tools supporting all lines of business and different job roles,
such as AI-powered suggestions for next best action for customer support
reps or optimized code suggestions for developers.

Democratized AI automates more of the applications

we use to live, work, and play, allowing more time to
focus on high-value activities.

9
A clear path to democratized artificial intelligence
Organizations can democratize AI and ML across a broad range of functions and “The primary goal in demographic mapping is optimizing across accuracy and
roles with Amazon Web Services (AWS) solutions. For business analysts, Amazon scale,” reports Anindya Datta, CEO of Mobilewalla. “While this is generally
SageMaker Canvas generates accurate ML predictions and boosts collaboration difficult, we used Amazon SageMaker Autopilot to produce better models that
by sharing models and datasets with data scientists. improved our prediction accuracy by 137 percent.”

“With SageMaker Canvas, our business users can easily explore and build ML Developers can leverage a core set of AI services for text and documents,
models to make accurate predictions without writing any code,” says Marc chatbots, speech, and vision—built on 20+ years of Amazon experience in ML—
Neumann, product owner, AI Platform, BMW Group. The solution enables that can be used by any developer to easily add intelligence to any existing
business analysts to connect, access, and combine data, create unified business application; no ML skills required. In addition to these core services,
datasets, quickly and simply build ML models, analyze these models, and you have access to specialized and industry-specific AI services supporting AI
generate accurate predictions with just a few clicks. and ML use cases, such as forecasting, personalization, anomaly detection, fraud
detection, search, DevOps, industrial, and healthcare.
According to Davood Naderi, data science team lead at Industrial Applications,
Siemens Energy, “The collaboration is important because it helps us “Our developers now focus more of their energy on delivering differentiated
productionalize more ML models and ensure all models adhere to our quality capabilities and less time investigating problems in our production
standards and policies.” environment,” says Zak Islam, head of engineering, Tech Teams, Atlassian.

Ready to remove heavy lifting from the ML process? For ML teams (which AWS is also adding ML capabilities to other services and products. For
can include data scientists, MLOps professionals, and data engineers), there example, Amazon CodeWhisperer is an ML-powered service that helps
are many ways to leverage AutoML. Amazon SageMaker JumpStart offers improve developer productivity by generating code recommendations based
hundreds of built-in algorithms and pre-built ML solutions that you can deploy on developer comments made (in natural language and in code) within the
with just a few clicks. integrated development environment (IDE).

Amazon SageMaker Autopilot eliminates the work of building ML models.

You simply provide a tabular dataset and select the target column to predict, “Our developers now focus more of their energy on
and SageMaker Autopilot will automatically explore different solutions to find delivering differentiated capabilities and less time
the best model. You then can directly deploy the model to production with investigating problems in our production environment.”
just one click or iterate on the recommended solutions to further improve
model quality. Zak Islam, Head of Engineering, Tech Teams, Atlassian

10
KEY #2

Operationalization
With AI and ML adoption growing rapidly, even urgently, business and technical complicates management across teams of business analysts, developers, and
teams are challenged to build more with ML and to build it faster. In this rush data scientists and in coordination with existing software tools and processes.
to leverage the technologies, organizations rarely stop to set standard tools In these common scenarios, scaling up or down becomes exceedingly difficult.
and processes for ML development. As a result, different teams with different
The good news is that there’s a proven way to minimize the risks and
skills and requirements often use completely different and disconnected tools,
complications of ML while providing straightforward, repeatable practices for
making collaboration impractical, if not impossible.
teams—by operationalizing ML. Operationalization of ML provides the tools,
For example, an R&D team might be working on a computer vision application infrastructure, and operations support to scale.
with state-of-the-art algorithms and frameworks, while sales and marketing
Operationalizing ML starts with the data acquisition and modeling activities of
teams are building a linear regression model to forecast customer demand
the data science team being informed by a clear understanding of the business
on a locally stored spreadsheet. Or developers could be coding a mobile
objectives for the ML application and of all governance and compliance issues.
purchasing app and want to add a recommendation engine to make the
MLOps ensures that the data science, production, and operations teams work
customer experience more personal.
seamlessly together across a series of ML workflows that are as automated
Relatively few organizations utilize operational ML tools and practices—such as possible. Human intervention is incorporated as needed, ensuring smooth
as infrastructure, IDEs, debuggers, profilers, collaboration tools, workflows, deployments, data monitoring, and model performance tracking.
and project management tools—that can be connected securely. This reality

11
How AWS helps customers operationalize machine learning
Amazon SageMaker, which we already know is a powerful service to help democratize ML, is
equally suited for operationalization. It automates and standardizes every step of the MLOps
workflow to help projects scale without limits. Thanks to SageMaker, AWS customers are
running millions of models with billions of parameters and generating hundreds of billions “Being on AWS enables a faster
of predictions. development process for us.”
SageMaker also offers an end-to-end ML service for data labeling, data preparation, feature Kenji Takeuchi, SVP of Technology Solutions,
engineering, training, hosting, monitoring, and workflows. Your team can use a single visual Mueller Water Products, 2021
inference in Amazon SageMaker Studio. In comparison to self-managed ML environments, the
productivity of your data science team can improve by up to 10 times and model development
time is reduced from months to weeks. And all SageMaker capabilities are offered on fully
managed, low-cost, high-performance infrastructure in the cloud.

AWS customers are realizing massive scale (and savings) with SageMaker tools:

• Vanguard has fully automated the setup of its ML environments and now deploys
ML models 20 times faster
• AstraZeneca can deploy new ML environments in five minutes versus one month to generate
insights that improve R&D and accelerate the commercialization of new therapeutics
• NerdWallet reduced training costs by close to 75 percent, even while increasing the number
of models trained
• Zendesk reduced ML inference costs by 90 percent by deploying thousands of models per
endpoint using SageMaker multi-model endpoints
• Mueller Water Products cut the number of false alerts in half and maximized the potential
to identify true leak events

Operationalization also means that we need to deliver the breadth and depth of AI use cases,
including intelligent contact centers, intelligent document processing, content moderation,
personalization, intelligent search, fraud prevention, identity verification, predictive
maintenance, AI for DevOps, health AI, and ML-powered BI. AWS offers services for all these
use cases and more.

12
K E Y #3

Responsible AI
Simply stated, responsible AI is a critical element of any AI and ML project. Encompassing a
core set of concepts—fairness, explainability, security, privacy, and governance—responsible
AI mitigates risks through the transparent use of data and models. It can be used to enhance
model performance, enhance data protection, and enable bias detection and mitigation
mechanisms in ML systems to improve fairness.

Responsible AI is an integral part of the complete AI and ML lifecycle, extending from initial
design and development to deployment and, ultimately, ongoing use. It is an iterative
process that requires ongoing testing and auditing for potential bias and accuracy. While
most companies have begun their journey to responsible AI, the majority (94 percent) are
struggling to operationalize across all key elements of responsible AI.2

“94% (of companies) are struggling to operationalize

across all key elements of responsible AI.”
Accenture, 2022

So, how do organizations transform responsible AI from theory into practice? They begin by
educating the next generation of ML leaders to elevate fairness and mitigate bias by bringing
more diverse perspectives to the table and providing resources to promote education and
training. Responsible AI also requires a multidisciplinary effort by technology companies,
policymakers, community groups, scientists, and more to tackle new challenges as they arise
and work to share best practices and accelerate research.

“From AI Compliance to Competitive Advantage,” Accenture, 2022

13
2
Build more responsible, inclusive AI with AWS Amazon SageMaker Model Monitor automatically detects and alerts you to
inaccurate predictions from models deployed in production.
AWS is committed to developing fair and accurate AI and ML services and
providing customers with the tools and guidance needed to build AI and Check out three essential resources to enable more responsible AI:
ML applications responsibly.
• The Responsible Use of Machine Learning guide provides considerations
As you scale your use of AI and ML technologies, you can leverage AWS and recommendations for responsibly developing and using ML systems
resources to help implement responsible AI across the entire AI and ML across three major phases of their lifecycles: 1) design and development,
lifecycle. AWS services help you better detect bias in datasets and models, 2) deployment, and 3) ongoing use. Read the guide ›
provide insights into model predictions, and better monitor and review model • Work with experts in responsible AI within our AWS Professional Services
predictions through automation and human oversight. organization to create an operational approach encompassing people,
processes, and technology that maximizes benefit and minimizes risk. The
You can mitigate bias and improve explainability with AWS purpose-built
engagement includes the development, deployment, and operationalization
services. Amazon SageMaker Clarify helps you mitigate bias across the ML
of responsible AI principles. Learn more ›
lifecycle by detecting potential bias during data preparation, after model
• Continuous education on the latest developments in ML is an important
training, and in your deployed model by examining specific attributes.
part of responsible use. AWS offers the latest in ML education across your
Similarly, SageMaker Clarify provides greater visibility into model behavior,
learning journey through programs like the Machine Learning University
both overall and for individual predictions, so you can provide transparency
(MLU), Training and Certification program, AI & ML Scholarship program,
to stakeholders, more deeply inform humans making decisions, and track
and AWS Machine Learning Embark program.
whether a model is performing as intended. Monitoring is also important
to maintaining high-quality ML models and ensuring accurate predictions.

AWS is committed to developing artificial intelligence and machine

learning in a responsible way, helping our customers put responsible AI
into practice and spurring research and continued development in this
area. Our work to build more responsible, inclusive AI is just beginning.
Learn more ›

14
Successful outcomes start here
More than 100,000 customers have chosen AWS for AI and ML to create new customer
experiences, optimize their businesses, augment their employees’ ingenuity, help improve
the quality of their products, and so much more. That’s because AWS supports you no matter
where you are on your AI and ML journey—with the solutions you need to scale without limits.

Connect with the experts from AWS

• AWS Professional Services is a global team of experts that can help you realize your desired
business outcomes for the AWS Cloud
• Amazon ML Solutions Lab provides a risk-free environment in which to identify and test
ML solutions and consider your organization’s best ROI-producing opportunities

Collaborate with an official AWS Partner

• AWS Partners are uniquely positioned to help AWS customers accelerate the journey
to the AWS Cloud

Construct it yourself with proven solutions

• AWS AI services let you easily add intelligence to applications—no ML skills required
• Amazon SageMaker empowers users to build, train, and deploy ML models for any use case
with fully managed infrastructure, tools, and workflows
• AI Use Case Explorer helps you discover the top AI use cases, customer stories, and
implementation paths based on your business objectives
• AWS Solutions Library offers solutions built by AWS and AWS Partners for a broad range
of use cases
• Amazon SageMaker JumpStart provides SageMaker solutions for the most common use cases

15
The visionary Harvard Business Review article that portrayed the role of the data
scientist as “the sexiest job of the 21st century” closed with a universal message
aimed at cautious circa-2012 business leaders waiting on the sidelines for the
technology to mature. “If companies sit out this trend’s early days for lack of talent,
they risk falling behind as competitors and channel partners gain nearly unassailable
advantages. Think of big data as an epic wave gathering now, starting to crest. If you
want to catch it, you need people who can surf.”3

Fast-forward to today. The data is unquestionably bigger, and the wave more epic—
but now it’s a democratized, operationalized wave that all can surf responsibly.

From the world’s largest enterprises to emerging startups, more ML is built on AWS
than anywhere else.

Learn more about how your business can deliver successful

AI and ML outcomes ›

3
Davenport, H., and Patil, DJ, “Data Scientist: The Sexiest Job of the 21st Century,” Harvard Business Review, 2012
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.