Generative AI

Handbook

What to know and what

actions to take to prepare
your business for an era
of exponential innovation
 Table of contents

Introduction 3
Boundless potential at our fingertips

Landscape 4
How we arrived here, where we may be headed and meeting the natives

Basics 8
The strengths, weaknesses and unknowns of the technology

Applications 12
Known and expected use-cases of Generative AI in enterprise environments

Responsible AI 21
Deploying Generative AI solutions in a responsible and effective way

Readiness 26
Practical guidance for adapting and leading in a Generative AI world

2 The Generative AI Handbook

Introduction

Generative AI will reshape every industry. But when?

There’s little question that Generative AI has captivated models and operations. Many are now looking to scale early Still, through skills-building and laying responsible foundations
business interest since ChatGPT launched at the end of 2022. successes through broader initiatives. in 2023, companies equipped themselves for the next stage
Interest has only grown since that announcement and we of maturity in leveraging AI’s generative potential. 2024 is the
believe it will transform organizations through new levels of Our own research and client conversations this past year year to accelerate AI impact through focus and investment.
human-machine collaboration. Throughout this guide you’ll reveal enthusiastic curiosity tempered by thoughtful diligence
find statistics, predictions and perspectives to spur thinking on around these emerging capabilities. As enterprises look to The rules of engagement continue to rapidly evolve as
how to pragmatically apply this technology to innovate. transition experiments into scaled production-grade solutions, practical experience refines our thinking on the possible. By
understandable caution accompanies the excitement. working together, we can apply this technology practically
The sense of urgency is understandable given rapid and responsibly to increase productivity and deliver superior
advancements. However, while most companies have actively human-centric experiences.
explored Generative AI’s potential through proofs of concept
and early-stage experimentation this past year, Cognizant
research shows that many leaders (30%) believe meaningful
A handbook for realizing Let’s shape the future together with care, creativity and
purpose.
impact is still years away. Gen AI’s potential in
For most executives we engage, the question is not “if” but the enterprise
“how and when” Generative AI will transform their business

3 The Generative AI Handbook

Landscape
The human-like ability of Generative AI
to converse, consider and create has
captured imaginations. By understanding
how we got here – and the decades of
thinking that led us to Generative AI –
we can better predict what’s coming next.

4 The Generative AI Handbook

Landscape
Expert Advice

How we arrived here

An innovation boom 75 years in the making Since Alan Turing’s 1950 “Imitation Game” (Turing Test)
“ Generative AI sits within the
context of decades of research
into AI. We are now entering
proposal, we’ve imagined a future of computers with the era where this technology
In the summer of 2022 – well before Open AI’s ChatGPT human-like intelligence, personality and autonomy. Today, will start to fundamentally

”
exploded into the public discourse – Generative AI began we seem to be accelerating towards that disruptive future.
generating media buzz when a recently-fired Google
transform businesses.
Engineer claimed1 their LaMDA model might be sentient. If and when we reach sentient AI is anybody’s guess but Naveen Sharma
Generative AI presents a significant new milestone in this Head of Artificial Intelligence
True or not, this wasn’t an entirely surprising claim for journey, sparking new interest, innovation and discourse. and Analytics Practice
artificial intelligence. We’ve been expecting this. Here’s how we arrived at this moment.

1950s – 60s 1970s – Mid-90s Mid-90s – 2000s 2010s - Present

The Birth of AI The Two AI Winters The Rise of Machine Learning The Advent of Deep Learning

In the early days of modern While progressive advances With the internet and accelerated Significant breakthroughs in neural network
computing, the Turing Test like expert systems proliferate business digitization, data and Generative AI model development,
and the famed 1956 Dartmouth and cultural interest in AI availability and IT funding expand accomplishing previously impossible tasks,
Workshop launch the era of AI. grows, underwhelming to drive practical AI applications. alongside surge in big-tech investment. As of
The first AI programs form and results lead to two periods of Major advances in robotics and Q1 2024, the Crunchbase AI startup list has
optimism is high reduced funding and focus data-driven solutions grown to nearly 10,000 companies2

5 The Generative AI Handbook

 Landscape

Where we’re headed

Change (and more change) ahead

In an April 2023 interview with CBS, Alphabet CEO Sundar Pichai famously commented that today’s Generative AI will soon
impact “every product across every company.”3 With rapid consumer adoption and the increased competitive pressure created
by Generative AI across industries, Pichai’s predication is sure to hold true.

As Generative AI permeates markets, it’s critical that adaptability be built into the technology and cultural fabric of
organizations. New, disruptive intra-industry and extra-industry use-cases will arise frequently in the coming years creating
continuous change to navigate.

1 2 3 4 5
A much larger New Generative AI models, Waves of regulation Generative video War for talent shifts
context window expanded AI features in and standards and AR/VR renaissance to war for innovation
Increasing context windows enterprise software World governments will adopt With significant advancement As 30% of work hours4 are
are critical for many Next-gen models are already and adapt regulations at in AR/VR technology expected to be directly impacted
enterprise use-cases and in development, including lagging pace as they address spearheaded by Meta, Apple by AI and resulting automation
will allow for larger, more open-source models with more rapidly evolving ethical, and Microsoft, compelling capabilities, productivity gains
comprehensive prompts flexibility and control. Expect economic and societal new applications backed will be felt by all. The war
to be passed to models. acceleration of new entrants and concerns. Organizations will by Generative AI will launch. for technology talent will be
This new access to vast innovation. Enterprise platforms formalize AI governance roles With conversational user reshaped as a war for technology
contextual datasets will are adding AI tooling that will with variable risk tolerance interfaces (i.e., chat, voice), innovation as organizations
open even more doors for AI drive further proliferation for use-cases new visual worlds will be seen differentiate with data

6 The Generative AI Handbook

Landscape

Meeting the AI natives

The forever digitally young the “AI-native company”.
How are early pioneers applying Generative AI?
Operating effectively in the era of Generative AI requires Being “born into” the Generative AI era is far less important
a reconstruction of the now decades-old digital maturity than exploration and adoption. Those organizations who • Expert assistants for professional work
narrative. We’re entering a post-digital era where pioneer AI -- and set the rules early to gain competitive Generative AI assistants being used in legal
every enterprise is digital and what defines leaders is market share from it -- will establish what it means to be software to augment services and accelerate work
their adaptability – which extends to their definition an AI native. Enterprise organizations, with their robust
of maturity, how they operate and what they sell.   proprietary data to build upon, have the advantage. • Media generation
Entertainment and creative studios using
The Generative AI era is one of continuous, perpetual Generative AI to create animation sequences,
change. The fundamental definition of computation and Early pioneers have common traits: social reels and more
chip design is changing with AI as rigid, linear and exact
computations are swapped for the abstract and inexact • Modern operations and architecture: Delivery • Market insights
logic that underpins neural network thinking. Companies and IT infrastructures that leverage data and Aggregating customer feedback (reviews, call
must now operate in similar ways. accelerate change transcripts, etc.) for Generative AI-driven insights

• Enabled AI culture: Definitive policies and governance • Text to software product generators
So, are there AI natives? with access and training for staff to benefit from AI Generative AI using component libraries, design
systems and code bases to build POC software “on-
As noted on in our Generative AI timeline, there has been • A focus on disruptive propositions: Already piloting AI demand”
an explosion of AI-centric startups born over the past experiences to hedge against new entrant threats
two years – these might be defined as AI natives. These • Interactive chat-bots
companies focus on AI and, presumably, they have AI built • Engaging suppliers to evolve services: Addressing Transforming the classic chat-bot, Generative AI
into their operations and culture as well as their product. Generative AI opportunities for mutual benefit powering conversational agents to answer and
explain
But just as our definition of digital maturity requires a • Simplifying not complexifying: See Generative AI as a
‘continuous change’ perspective, so too will our definition of way to transform value creation vs. ”just new tech”

7 The Generative AI Handbook

Basics
The fundamental strengths of Generative AI
perfectly mirror its unavoidable weaknesses.
The fundamental characteristics of the
technology provide insight into its disruptive
potential – and explain why adoption will
impact every part of the enterprise over time.

8 The Generative AI Handbook

Basics Expert Advice

Known strengths of Generative AI

The blurring boundary between human and machine work ideating on novel molecular structures or animating an entire
“ Generative AI provides completely
new capabilities to automate and
augment knowledge work. It is
new movie, this generation of AI composes complex patterns going to turbo-charge tasks that
Generative AI presents a fundamental change in our and data to create. require creativity and expertise,

”
understanding of what practical, immediately-accessible AI
such as design, engineering
can do. Chat-bots, candidate screening tools, summarizers As organizations come to understand the strengths and
and picture-makers might inspire us today, but soon AI will potential use-cases of Generative AI, they also begin to realize and quality assurance
shape the core of modern business. the fundamental requirements within their organization for
fully leveraging this technology. Pramod Bijani
It isn’t sentient but it sure does behave in human ways – and Head of Digital Experience and
Digital Engineering Delivery
that’s what’s so inspiring about this technology. Whether It all starts with data. Combine that data with business
finishing a sentence, writing the code for a component, processes and logic and Generative AI is transformative.

1 2 3 4 5 6
Complex process Data augmentation Predictive Driving efficiency Real time Multimedia
automation and completion analysis and supporting optimization generation
Core business processes Generative AI systems Generative AI is knowledge work Generative AI is capable Generative AI is capable
that, in the past, have not work with data to capable of analysing AI makes knowledge of monitoring processes of both consumption
been open to automation provide first-line complex, structured or work more efficiently by and outputs to proactively and creation of rich
due to complexity and analysis, classification, unstructured data to accelerating and expanding identify opportunities for media across text, audio,
variability can now be sanitization and more, identify patterns and on ideation, distilling data improvement, prescribe video and images,
managed and reshaped free from human error trends to form actionable to find insights, rapidly and even implement unlocking powerful new
by AI and at scale recommendations drafting and more changes possibilities

9 The Generative AI Handbook

Basics

Known weak points of Generative AI

To err is human. For AI, it’s by design.

The cyclical evolution of AI over the past 75

AI, contextual understanding is often difficult
to achieve “out of the box,” especially with
consumer tools like ChatGPT. This has been the
1 2 3
years has been marked by periods of waxing source of much of Generative AI’s criticism. Hallucinations Data quality The explainability
enthusiasm and waning pessimism. As new and AI safety problem
advances promised new opportunities, Its strengths shape its weaknesses When AI produces
institutions and businesses have jumped in unreliable and Because data shapes When AI provides
and invested heavily in the technology. When Generative AI wants to answer us. It’s built erroneous outputs, it AI’s knowledge base, any a decision, there is
outcomes haven’t met expectations, though, to respond to our prompts – no matter their erodes data-driven inadequate data inputs inherent uncertainty in its
the AI space has experienced disillusionment complexity – and often provides answers strategy, reduces will create bias and limit certainty. The larger and
and stagnation. that, in a sense, acknowledge this fact. customer trust and limits accuracy, fairness and more complex a model,
ChatGPT allows us to refresh responses. Image operational efficiency. decision-making. the harder to ascertain.
By its nature, AI is unpredictable and this is only generators like OpenAI’s DALL-E or the popular
further highlighted by the Generative AI boom Midjourney both return multiple images to any
and the human-like creativity of its outputs. single prompt. These tools understand that Expert Advice
they may be wrong.

“
Like humans and on many tasks, Generative AI
is capable of working flexibly towards a goal or The creative muscle of Generative AI can
Overcoming GenAI’s limitations, we are achieving
target output rapidly and creatively. Also, like be striking, but it is definitely not magic. very strong results in hybrid systems where
humans workers, context matters. Generative AI’s capabilities are fundamentally generative and evolutionary AI models are
based on reference data and training.
combined to play to their strengths. This will be

”
Whether its brand values, ethical
considerations, situational knowledge, historical AI adoption creates new categories of risk that a critical foundation for successful adoption.
learning, consumer needs or anything else, require focused assurance at the enterprise
human workers are expected to understand level. Organizations that engage in this Babak Hodjat
AI CTO
the context of their work and this can impact transformative technology with this in mind will
the output of their efforts. With Generative gain the most from the AI era.

10 The Generative AI Handbook

Basics

Generative AI’s open questions

Waiting is not the answer These questions largely mirror the ‘weaknesses’ discussed earlier and
their answers will be shaped through trial and error, learning and time.
As enterprise decision-makers consider their paths
forward in this new AI-enabled world, questions are What’s certain is that readying the organization to navigate this AI-
being asked that are as complex and difficult to enabled world is critical for future business performance – exploring
navigate as the technology itself. How do we know our these questions is a key part of that readiness.
AI isn’t hallucinating? What ethical boundaries does
this system present? How can we rely on answers, if we We’ll explore near-term steps to take to address these concerns within
can’t explain how our systems reached them? the Readiness section later in this guide.

How can we ensure What legal implications How can we ensure How will this impact our
ethical usage? should concern us? predictable output? brand or public perception?
Significant concerns for misuse and Concerns for copyright, IP infringement Concern over hallucinations, poorly Concern over customer preference
harm, bias in poorly trained systems and regulatory issues when dealing structured or insufficient direction in against Generative AI in key touchpoints
and other negative outcomes of use with protected data, privacy business-impacting scenarios and dilution of brand experience

Cognizant Answer: AI creates entire Cognizant Answer: These concerns are Cognizant Answer: Generative AI Cognizant Answer: As with other AI
new categories of risk that require ‘native’ to digital business. Processes can never be expected to provide questions, the key is to test and train
focused testing, governance and and tooling can help navigate and 100% predictability in output. This is Generative AI systems well before usage
assurance. LLMOps creates a safeguard here and every business why training, tuning and continuous and be selectively transparent with
framework for such responsible usage needs a unique plan monitoring must be integrated customers about use

11 The Generative AI Handbook

Applications
Consider the early plugins available
for ChatGPT, or bots on the Poe app,
and it’s clear that the use-cases of
Generative AI are about as vast and
varied as software itself – and those
are just chat interfaces.

12 The Generative AI Handbook

 Applications

7 proven use-cases of
Generative AI impact
Even at this early stage, the opportunities for Generative Al their own opportunity spaces, safeguard against disruption
across the enterprise are countless. With the right foundations, and begin building skills.
the only limitation of Generative AI solution-building may be a
company’s imagination. Fortunately, there’s no need to start this journey from scratch.
With the following seven example use-cases of Generative AI,
With so much opportunity and so many questions, it can be we’ll highlight just how varied the opportunity can be. Every part
hard to know where to start. As you’ll find in our discussion of of the value chain across every industry stands to be disrupted
Generative AI readiness later in this guide, what’s key is that in unique, differentiating ways as organizations bring their
organizations begin exploring this technology early to identify unique data, processes and POV to the discussion. Let’s dig in.

1 2 3 4 5 6 7
Market and Software Production Marketing Expert Employee Customer
Competitor Intel Development and Sales Advisors Engagement Experience

13 The Generative AI Handbook

1
Ingestion of Coverage adjustment
market data Identify new, relevant topics
React as persona
and add granularity to
Adopt customer and
existing topics being tracked
stakeholder personas
Applications and react to content

Market and Summarize and classify

Collate captured data

Competitive Intel
to identify relevance to
Gaining
defined topics of interest Ideate insights
Respond to insights to from data
generate new concepts
Turning data into human-readable, actionable and hypotheses
Prepare primary research
and contextualized guidance is a major strength
Define audiences and create
of Generative AI. Generative AI systems can
qualitative and quantitative Synthesize multiple sources
be used to industrialize data collection from
research questions Correlate, compare and
a range of sources, including curated market
research, real-time customer and competitive combine related content on
behavior, internet scraping and primary user a topic(s) to distill insights
Developing
research. Whether structured or unstructured, and validating Transcription
this data empowers systems to drive a range hypotheses Transcribe research
of automated analysis, summarization and data, extracting findings, Insight navigator
recommendations. confidence levels and aspects Provide a natural language
requiring follow-up research discussion interface to
interact with the corpus of
research data

Concept mock-up
Mock-up concepts Report writing
and stimulus for Generate narrative and Packaging and
research testing and explanation around raw distributing
user validation data and derived insights

14 The Generative AI Handbook

2
Generate Navigate current state
ideas Engage with AI to discuss User testing
enterprise structure, Use AI for scope testing
performance, code base, etc. sessions, to summarize
findings or suggest
Applications
improvements
Crowd sourced ideation

Software Collate employee

ideas, feedback and

Development
data to identify new
Requirements Develop
opportunities
elaboration concepts
Extrapolate concepts
Implementation planning to epics, stories and
Microsoft Visual Studio Code, the wildly Prioritize scope items, acceptance criteria
popular integrated development environment sequence to sprint plans and
(IDE), has long-supported GitHub’s Copilot assign to teams based on
product (by some estimates automating 40%- Concept mock-up
functions and skills
60% of code writing5) and now also integrates Create rapid product
ChatGPT directly into the developer interface. visualizations to bring to
But the utility of Generative AI during software life new ideas and support
development goes well beyond writing Quality assurance user testing
components. The entire software development Implement Assess risks and failure
process is set to see transformation as software modes, build test Release readiness
this technology impacts creativity, quality, scenarios, test data and Inspect code and project
productivity, compliance, utility and more. automations to validate actions to determine if go-no-go
criteria are met and the solution
is ready to release

Code completion
Create code based Change management
on collaborative Identify impacted groups Assurance and
developer input and and assist those groups in release
code base context addressing impacts

15 The Generative AI Handbook

3
PRD shaping Materials research
Shape product Assist in identifying and AI-powered scenario
requirements based on optimizing materials modeling
current and expected for a variety of Emulate real-world scenarios
Research and virtually at vast scale to
Applications Development
future demands requirements
optimize product design

Production Tradeoff-based
Niche user prototyping Design and
optimization Rapid virtual prototypes Prototyping
Product research, production and quality control Design critique AI Simulate specific based on adjustable goals
will see significant Generative AI impact in the Partner with user scenarios, and balance of trade-offs
coming years as organizations across industries humans to suggest variants
seek to unlock transformative new efficiency critically respond to that service new
Generative design
and product innovation ahead of competition. concepts and help groups
Create numerous variants
This zone is highly controlled and data-intensive, shape design
based on requirements,
making it a perfect early adoption area. The
accelerate creative work
IP established through smartly leveraging
Generative AI in this space will reshape industries Production quality
and establish new leaders. assurance
Production External systems
Systems to inspect integration
products, ID defects Generative AI can be
or inconsistencies integrated with ERPs and
accurately at scale CRMs for further insights

Go to
market
Cost/sustainability
optimization Supply chain optimization Continuous improvement
Analyze design-to- Predict supply chain feedback loop
production pipeline disruptions, optimize Analyze product
for cost savings sourcing and optimize performance data and
and sustainability logistics feedback to drive refinements

16 The Generative AI Handbook

4
Manage Software development
Channels Create software to support
User testing
interactive campaign
Scope testing sessions,
elements
summarize findings and
Applications suggest improvements

Marketing
Content production
Create digital assets: Lead nurturing
images, alt text, copy, Create personalized

and Sales translations, scripts, Nurture

comms, content
campaign plans and more Leads
and experiences for
engagement and
Sales prioritization nurturing leads
Generative AI improves planning, production
efficiency and effectiveness throughout the Identify the best activities
marketing and sales journey. As the technology for sales teams to focus
gains adoption, asset production cycles will on and craft personalised Lead capture
see a marked acceleration with a range of action plans Conversational data capture
potential new asset types and channel strategies at lead identification stage to
becoming available. Further, self-service provide rich lead information and
channels will become more personalized and Manage Sales presentations immediate customer response
impactful while sales staff will increase their Opportunities Generatively compose
productivity and knowledge to focus more time presentations that combine Negotiation and closure
on driving successful customer engagements. sales strategy, customer Assemble contracts and
context and standard offerings statements of work to
match proposals, MSAs,
standard terms, etc.

Sales effectiveness
Train sales staff on Customer RFP responses
latest propositions, Respond automatically Negotiate
messaging and using best practice and Close
tactics answers tailored to context

17 The Generative AI Handbook

5
Qualification
Understand user issue and Triage
Customer determine if expert consultation Understand the urgency
Requests is needed and appropriate and importance of the
Applications Service request to prioritize service

Expert Initial data capture Collect

Advisor
Address questions Information
in a conversational
Collection management experience to gather
Chase participants necessary data
to provide missing
Generative AI streamlines and accelerates the
information and correct
provisioning of expert advice to benefit end- Information extraction
errors
users and businesses alike. In many scenarios, Process unstructured data
Generative AI has the capacity to act in a self- sources to extract the
service model to provide expert guidance directly required information
to users. Where complexity is higher or in safety-
critical environments, Generative AI can facilitate Recommendation Downstream fulfilment
many stages of the process without acting in a Evaluate and Apply standard policies Interpret expert
fully autonomous way. With AI-driven pre- and Recommend and procedures to conclusions and
post-processing, experts can more effectively the input context to trigger appropriate
utilize their time and focus on the highest-value suggest outcomes follow-up activities Action
or most-critical scenarios.
Recommendations

Anomaly extraction Case discussion

Draw attention to Allow experts to Quality assurance
interesting or outlier interactively discuss the Review expert conclusions
input data features case with the necessary and highlight any causes
background knowledge to for concern (e.g., policy
take action misalignment, bias)

18 The Generative AI Handbook

6
Interview assistance
Suggest role-specific
Recruit Close
questions, transcribe and
Develop tailored offer
summarize conversations
strategy to close strategy
Applications using interview transcripts
Candidate screening

Employee Review applications,

assess fit and

Engagement
extract concerns
prior to interview Staffing
Onboard
Leverage enriched
findings to match
new hires to roles and
Companies that adopt Generative AI at a opportunities
Community connection
cultural level, going beyond asset production and
Suggest connection to
chat interactions to elevate all common touch-
communities of interest, New hire buddy
points for customers and employees alike, will see
sites, etc. Conversational onboarding
the biggest gains in the coming years. Employee
engagement is an exciting space for Generative assistant to guide employee
AI with the potential to impact recruiting, journeys for first month
onboarding, team-building, performance
management, support and more. The efficiency Enable
Support AI
gains here will empower innovation across the Assist with self-service Performance coach
business as Generative AI permeates the market. support for IT, Human Review activities, outcomes
Resources, etc. and feedback to suggest
specific learning and
development activities

Accelerate admin
Simplify standard Educational AI Develop
activities like timesheets, Provide interactive training
vacation requests, on skills development topics
expense reporting, and answer questions on
training, etc. company data

19 The Generative AI Handbook

7 Shopping and
Product Comparison
Virtual Demos / Try-Ons
Generative AI visualizes
product in various settings,
including worn, for better
Contextual UI
Adaptive user interfaces,
by context: behavioral,
Conversational UI
Users use natural
Applications product qualifying social, temporal,
language (voice or text)
emotional, personal

Customer
Intelligent to engage, filter, qualify,
Product Suggester assist, etc. during
Synthesize wide user buying experience

Experience
data to create a new
level of personalized Predictive Return Systems
shopping experience Based on prior buying Purchase,
behavior, post-purchase Returns
Whether they’re “just browsing” or already a Fraud Detection AI engagements, etc. predict
loyal customer, the way that people engage AI for analyzing customer and address concerns
with brands throughout the shopping and behavior to identify and Dynamic Personalized
post-purchase experience is set to dramatically respond to fraud or suggest Pricing
evolve with Generative AI. With answers new systems that might Potential for novel AI-
becoming more seamless and appetite for further prevent fraud driven pricing mechanisms
content noise decreasing, customers will expect based on various inputs to
personal, intuitive, adaptive touch-points that optimize sales
understand and serve their needs. It’s time to Churn Prevention AI
put more you into UX. ID behavioral patterns
Behavioral Campaign
of at-risk customers and
Loyalty Shaping
design engagements to
Generative AI learns from
prevent churn
loyal buyers to identify/
surface trends and formulate
full-funnel campaigns
Generative Loyalty
Programs Generative Remarketing
Novel loyalty programs Systems to dynamically
using Generative build cohort-based
Repeat
AI systems that and personalized
personalize incentives remarketing strategies

20 The Generative AI Handbook

Responsible AI
Unlike the software solutions of the
pre-Generative AI world, generative
solutions cannot be built, tested, and
released into an ecosystem without
continuous oversight. Continuous
governance is mandatory, and it’s cultural.

21 The Generative AI Handbook

Responsible AI

Deploying Generative AI in a responsible and effective way

With all of the compelling use-cases
for Generative AI and the immediate
accessibility of public tools in the market Guiding Principles for Responsible AI Development
today, it can be easy to get carried away
Tactical principles for developing Generative AI solutions in efficient, safe and value-oriented ways
in the AI hype. That same consumer
availability of basic AI tooling can
trivialize the complexity and downplay Be Robust, Be Safe Design for Transparency
the policy, process, partnership and skill AI systems should perform reliably and safely. By building and AI systems should be understandable. Build trust and drive
required to build tailored, production- deploying AI in accordance with best practices where we robustly understanding through silo-breaking collaboration and rich
grade solutions. test before deployment then monitor and improve operations communication across users and stakeholders, allowing them
regularly, we can reduce the risk of harm or unintended outcomes. to understand AI systems and system outputs within their own,
Enterprise AI and machine-learning personal context.
(ML) applications require automation Benefit People and Communities
engineering, data management, feature Build a more sustainable and inclusive world through AI Promote Inclusivity and Minimize Bias
engineering, resource management, QA, innovation. AI outcomes must incorporate human benefit and Bias exists in our data, models and our world; responsible AI
and testing teams, IT operations and environmental sustainability in order to deliver impact and value systems seek to ensure AI is fair, unbiased and representative end
more; all aligned with coherent, shared to shareholders, users, customers, employees and society at large to end and full-context. AI systems should treat people fairly and
accountability in order to effectively AI should be produced and reviewed by diverse teams.
deliver solutions. In the following pages, Protect Privacy and Respect Boundaries
we will explore how LLMOps expands our AI systems must be secure, compliant and respectful of people. Drive Accountability and Enable Participation
view of DevOps and how an updated view Affirmative consent and a human-centered, privacy-first People should be accountable and in control of AI systems. Clear
of Quality Engineering can safeguard AI approach ensures sensitive data is never used unethically. processes and incentives for engagement create a culture where
solutions with holistic automated testing. A variety of auditing systems and safeguards are key within every individual is empowered to protect people, minimize risk and
Generative AI systems. discover spaces of humane value.
It all starts with setting strong, enforceable
principles for Responsible AI development.

22 The Generative AI Handbook

Responsible AI
The following two pages provide an introduction

LLMOps brings speed, support and safeguards

to LLMOps but remain too high-level to
sufficiently detail the orchestration of people,
tooling and processes required to operationalize
these practices. To explore more, engage with

to solution development
Cognizant’s Generative AI team.

Generative AI solutions can be surprisingly seamless. • Generative AI often leads to scenarios where privacy, These LLMOps KPIs can include:
The ability to understand users, act on their needs and regulatory compliance and data leakage are concerns –
provide human-like creative responses is what makes systems must be secure • Cycle Time: Duration from initiation to deployment of
Generative AI such a compelling solution today. Behind Generative AI-based solutions
the scenes, though, Generative AI solution development • Manual review of issues, anomalies and experience is untenable
adds layers of complexity to the work of digital teams and might remove any ROI or advantage of these systems – • Deploy Frequency: How often updates or new Generative
that go well beyond API keys and prompts. Orchestration automation is needed AI solutions are pushed to production
is critical.
Due to these conditions, new ways of working are required. • Review Time: Time taken to review and validate solution
Consider the following difficulties: Building on the popular concept of DevOps, LLMOps provides (model) outputs and performance
a path forward.
• Generative AI systems are difficult to test and • Automated Solution Ratio: The proportion of tasks handled
unpredictable by nature – they cannot be validated LLMOps defined: LLMOps aligns data engineering, agent autonomously by the system
and deployed in a ”set and forget” fashion development, software engineering and IT Operations to enable
continuous integration, continuous delivery and continuous model • Data and Model Drift: Discrepancies in solution (model)
• Generative AI development requires many skills and training with an emphasis on automation and monitoring at all steps. outputs or incoming data from the training scenario
teams must rely on each other – this takes coherent
organization and alignment programs As organizations seek to develop effective Generative AI- enabled
solutions for internal and external users, defining and enforcing their
• Generative AI systems must be at once predictable own LLMOps approach is imperative.
LLMOps and AIOps are often confused but they
but flexible – this takes training and continuous
describe entirely different disciplines: LLMOps
monitoring This often starts with defining the KPIs of Generative AI solutions
standardizes machine learning model deployment
(aligned to Responsible AI principles) and ensuring that processes,
while AIOps automates IT operations.
• Generative AI development requires too many skills governance and tooling are in place – made possible by LLMOps --
and processes to manage with manpower alone to monitor and influence those KPIs.

23 The Generative AI Handbook

Responsible AI

Transparency and automation at the heart of LLMOps

As organizations tip-toe into Generative AI, linear solution development Immature LLMOps
processes will be favorable for proof-of-concept development at speed. The Linear in nature, siloed Data Science, ML/AI, Engineering and Operations functions. Fine for POC development and lab
belief is that model training is something done early within a process and environments, lacks required automation, CI/CD, feedback loops and monitoring for enterprise Generative AI solutions.

that a trained model can be utilized endlessly. This can work for testing the Training Data
waters but is not a sustainable solution.
Software
Development
Mature LLMOps processes are iterative in nature with observability and Data Model Evaluation Trained LLM LLM Solution Deploy to
Application
Preparation Training of KPIs / Perf. Solution Registry Production
automation at their heart. As a continuous cycle, LLMOps allows data-
intake and learning to regularly impact the solution while automating as LLM R&D Operations
much as possible and keeping humans in the loop. This feedback loop is key
to responsible AI development. By ensuring that model behavior, application
performance, data protection and system changes are controlled through a Mature LLMOps
technology-driven workflow, organizations can operate more effectively. Cyclical in nature, integrates disparate workflow functions through automation and observability.
This system is heavily reliant on coherent workflows, culture and LLMOps tooling.

Expert Advice

“
Run Monitor Evaluate

We must build quality and control into Serve/Monitor Modal Outputs/Processes Automate or Alert Human Team

AI solutions to manage their continuous

evolution. Due to their broad capabilities Continuous Cycle Drive 360° Observability Maximize Automation Enable Human Actions Continuous Cycle

and emergent behaviour, management

”
Core Tenets of LLMOps Tooling
is needed across the entire lifecycle.
Deploy Quality Assure Refine
Andreas Golze
CI/CD Automated Review/Fix Improve Model/Monitor
Head of Quality Engineering
& Assurance Practice
Model Training (Initial & Continuing) - Automated, Monitored

24 The Generative AI Handbook

Responsible AI

Quality Assurance and testing systems in a Generative AI world

The evolved role of Quality Assurance’s (QA) teams and tooling within the delivery process will
be a critical focus area for organizations seeking to deploy LLMOps.
Deterministic Outputs Probabilistic

While QA automation has become an area of strength for many mature engineering
organizations, traditional approaches are insufficient for Generative AI. The scope of QA and
Consistent across data-sets Expected Behavior / Result Varies highly across contexts
test automation has changed, with new driving factors to consider for AI-based applications.

Low Concept / Data Drift High

Updated QA automation capabilities must cover:

Minimal Regression of Application High
• Systemic Data Assurance: Supports operations and decisions across process
Ex. Feature selection analysis, input data augmentation, data fitment analysis
Clear Test Coverage Unclear
• Systemic Model Assurance: Automation for monitoring model behaviors
Ex. Model fairness analyzer, model validator, model interpreter, adversarial robustness tools
A part Data The core

• Generative Output Assurance: Automation to monitor outputs of Generative AI

Ex. Naturalness, Toxicity, Coherence, Demo Polarity, Clarity, Truthfulness, Relevance, etc.
Not applicable Bias Needs focus

• Low data scientist QA efforts: Services and tooling to reduce specialist effort
Ex. Low-Code / No-Code solutions, customizable observability UI, etc.
Simple, in most cases Interpretability Complex, in most cases

• Deep insights: Automation of insights from AI solutions / workflows

Ex. Generative AI tooling to explain solutions, model outputs, live data, etc.
QA for Standard Applications QA for AI-Based Applications

25 The Generative AI Handbook

Readiness
Preparing an organization for the
Generative AI world doesn’t mean
moonshots and risk, it means
foundation laying and learning.
Setting up governance now will pay
dividends near and long term.

26 The Generative AI Handbook

Readiness

Practical guidance to reap the

benefits of Generative AI
Despite the hype around Generative AI, we’re still in the early days With so much hype and a sea of noise to cut through,
of the AI-driven business. It’s a certainty that AI will transform every many organizations are asking more tactical questions.
corner of our digital universe and yet we’re continuing to learn What must be navigated to move forward?
how. With new applications conceived daily and development of
next-gen Generative AI models underway, innovators are fast at To get practical about Generative AI, start with these
work reshaping the future of work. Adaptability in such a rapidly questions. In the following pages, we’ll double-click into
changing landscape is critical. each as we explore a path forward.

“To get practical about Generative AI, have we:”

1 2 3 4 5
Provisioned initial Designed a path Introduced AI to Modernized and Set new
access to to scale successful the development significantly expectations with
enterprise-grade POCs? process? automated core our suppliers?
Generative AI tools? business operations?

27 The Generative AI Handbook

Readiness

Provisioning initial access to enterprise-grade Generative AI tools

In the wake of ChatGPT’s emergence, it’s safe to say that considerations of Privacy Policy adherence, Terms of Service,
every enterprise was abuzz with cautious excitement about regulatory considerations and more bans are surely on Steps to take
the potential of this new technology. the horizon.
• Update policies and governance
At the individual team-member level, workers around the But still, the advantages… A blocker for many orgs today, set clear policies and
world began testing Generative AI for their own use-cases. governance for a generative AI world
A recent survey from The Conference Board found that 56 As new products go, any amount of friction (cost, risk, etc.)
percent of workers are using generative AI on the job, with can have a chilling effect on adoption. But Generative AI isn’t • Educate employees actively, communicate often
nearly 1 in 10 employing the technology on a daily basis. Yet simply a new product; it’s a transformative technology that Clear communication to associates around access,
just 26 percent of respondents said their organization has can change the world in striking, progressive ways.   policy leadership, support and appropriate usage
a policy related to the use of generative AI, with another 23
percent reporting such a policy was under development.6 Early adopters will have the advantage in this new world. • Set governance over approved services
Beyond the obvious cultural and process execution benefits of Clarify approved GenAI solutions and establish
Organizations have been relatively quick to respond Generative AI, we expect a patent boom in the coming years ownership for access provisioning and support
to these risks. Amazon, notably, found7 indication of as organizations invent novel uses of Generative AI-based
proprietary data in public model usage and responded with tools within their business. • Identify and supervise adoption
a ban. Maintain observability over approved AI services,
Preparing the business for Generative AI means getting monitor who is using them and for what purposes
This isn’t Luddite behavior, it’s just good practice. At serious about near-term, safe-guarded adoption with well-
this early stage, it’s unclear exactly how customer data, integrated monitors and control of usage. Gain advantage • Learn and capture best-practices
proprietary business data and other protected data is either while minimizing risk and learn as you go. Source insights and best practices within an
being exposed to the operators of public LLMs or used organizational context, communicate learning
to train the models themselves. Couple this with the simpler

28 The Generative AI Handbook

Readiness

Design a path to scale successful POCs

We’re quite early in Generative AI’s Diffusion of Innovation Quietly building a boom
bell-curve. Early adopters are establishing and quantifying To ready themselves for the road ahead, it is imperative Outcomes in focus
basic use-cases – gaining earned media as a result – that organizations go beyond provisioning access to public
and most would-be digital leaders are watching with tooling and begin developing their own inside use-cases to • Build a business case for expanded focus
curiosity.   drive a business case, spark thinking and lay a foundation Give enterprise context to Generative AI use-
for future development. This can be done in phased, cases by showing viability and capability
controlled and protective ways. against revenue-driving focus areas, expand on
success.
We suggest two complimentary approaches:
We are • Gain edge, protection from disruption
here 1. Establish and run pilot projects Leverage Generative AI early for disruptive use-
Pilot projects build a ‘light house’ for future innovation cases that drive business value, safeguard the
and expansion. By establishing specific initial goals business and accelerate future projects.
for a cross-functional pilot project team to pursue,
Early Early Late Left
Innovators adopters majority majority behind
organizations can create disruptive proofs of concept • Develop early internal SMEs and insight
and establish an internal POV. Begin establishing subject matter expertise
within core internal teams by capturing
2. Enable and accelerate with AI innovation “Labs” and sharing knowledge during practical
In an August 2023 report by Bain and Company , only 6% of
8
We have supported multiple organizations on application.
surveyed health systems executives have a clear Generative establishing their own innovation lab environments where
AI strategy in place, yet 75% believe that the technology governance, collaboration and technology enablement • Build internal and external reputation
can reshape their industries. The reasons are the same are high. These environments become particularly Establish industry point of view on Generative
that we’ve already discussed: uncertainty, risk, lack of inside powerful when formed in collaboration with hyperscalers AI, attract business and talent attention and set
knowledge and indecision. This lull in early adoption is where who might provide innovative organizations with access a cultural foundation for an AI enterprise.
the advantage lies. to advanced models, education and specialized tooling.

29 The Generative AI Handbook

Readiness

Identify opportunities to add AI to the development process

As covered in our section on LLMOps, Generative AI a job for designers and instead be handled by Generative User feedback
development implies systemic changes to the way that AI tools. Fed with design principles, systems and reference
software is delivered and supported within organizations. designs, these prototype design tools will produce unbiased
Zooming in, the same goes for software product prototypes best fitting the market data available. The job Behavioral data
engineering processes. of designers will be to identify the most promising solutions
and refine them.
Market response
Here are just some ways that Generative AI might impact
product development. Risk Mitigation
A core responsibility in product management is to manage
Idea Generation and mitigate risk. With its predictive analytics capabilities,
The ability of Generative AI applications to work with AI tooling can help in identifying potential risks and
trained models while evolving those models (and the roadblocks early on in the prototyping phase. Quality, Idea Product
application’s outputs) with the consumption of real-time market readiness and future success can all be gauged by Generation Design
data can unlock compelling use-cases for product idea- having algorithms analyse historic data, user preferences
generation. Rather than relying on surveys and user and even real-time market trends.
reviews for qualitative data, Generative AI agents might
deliver new concepts frequently based on real-time Resource Optimization
analytics. Product managers can then link these ideas to Sustainability is the challenge of this generation of Resource Risk
business goals and set a path forward. business. Generative AI can support sustainability efforts Optimization Mitigation
by optimising resources and material mix for minimized
Product Design waste and environmental friendliness. It can take regulatory
As multi-modal models (capable of intaking and outputting processes into account, report on data and even affect
images, text, audio, etc.) mature and see enterprise subsequent production processes for both software and
adoption, ”clickable prototype” design will become less physical goods. GenAI-enhanced product development

30 The Generative AI Handbook

Readiness

Modernize and significantly automate core business operations

There is no doubt that Generative AI will revolutionize how In our opening section of this document covering the 2. Assess and expand on existing process automation
we work. Process automation has long been a popular use- future of Generative AI, we touched on a shift from a Apply Generative AI solutions to existing process
case in our digital world and AI is going to open entire new war for talent (commonly discussed in the 2010s and automations to further streamline for efficiency
opportunity spaces here. The debate around automation will pandemic era) towards a war for innovation as all
continue to be more focused on how regulators will impose businesses use Generative AI to gain efficiency. AI- 3. ID opportunities to pre-process data
limitations on the technology instead of how much potential enabled automation of core business processes will drive Utilize Generative AI to reduce or replace human data
the technology affords us. this shift. Savings must be reallocated to accelerating processing, gaining speed and accuracy
business value.
Like media streaming, ride-sharing and other tech-driven 4. Deploy generative assurance
economic catalysts that regulators have tried to stay in front 1. Target core functions for AI-driven augmentation Apply Generative AI to business processes to continuously
of, resistance is a bit futile. Technology will be squeezed for as Explore areas for generative AI to support knowledge and completely audit quality, regulatory compliance
much juice as it can offer in a free market. work and accelerate creativity and more

By 2032, few jobs could go untouched by generative AI

Some impact: Greatly impacted: Little impact:

90% Exposure scores of at least 5% 52% Exposure scores of at least 25% 10% Exposure scores of 5% or lower

Most jobs will see some change from gen AI, and over half could be greatly impacted.

Source: Oxford Economics and Cognizant9.

31 The Generative AI Handbook

Readiness

Set new expectations with your suppliers

It’s every company’s job to evolve Every one of your suppliers will be at a different stage of this
journey. As they navigate use-cases, seek to answer questions From supplier to partner
The early big press-makers of Generative AI have been about risks and control and otherwise dive into Generative AI,
the expected parties. Hyperscalers have introduced join them. As the innovation potential of Generative AI becomes
new or evolved platforms for building AI solutions within clear to more organizations, the opportunity to create
their ecosystems. Myriad ultra-specialized startups have Why engage now? wholly new experiences, services and processes
announced compelling new solutions to old problems by partnering with suppliers on a joint journey will
(ex: Hyfe’s10 cough sound monitoring for illness diagnosis). • You stand to gain from their improvements become compelling for many businesses.
And service providers, like us, are launching new Suppliers are critical to your bottom line. Ask how they plan
accelerators and labs for Generative AI development. to improve SLAs, decrease total cost of ownership, operate Going well beyond the cost savings of a joint
faster and otherwise drive more business value for you and investment, with enriched data, access to more skills
But Generative AI is coming for every product, in every other customers. and beyond, these partnerships might benefit both
market. That goes well beyond the early adopters and it parties in dramatic ways when executed well. Consider
includes you, your competition and your suppliers. • You will learn from one another the role of each key supplier within your service or
As all companies are learning, work with suppliers to product delivery and move the discussion beyond
As you seek to leverage Generative AI to unlock understand their own findings, partnerships and interest what they can do with AI for you.
new efficiency, differentiate experiences, maximize areas. Formalize shared learning between strategic teams
quality, find cost-savings and evolve the business model, and capture insight.
don’t discount the role your suppliers will play in
these improvements. • Open the door to more data
As you engage with your suppliers, consider internal
Whether a service provider, a manufacture or raw goods solution opportunities and how supplier data might
provider, a logistics service or any other company that improve model training and solution delivery. Bring
plays a role in your operations, there is an advantage to predictability, new product ideas, streamlined operations
engaging early in a dialogue about Generative AI. and more to life.

32 The Generative AI Handbook

To learn more about Cognizant’s
Generative AI solutions:
https://www.cognizant.com/us/en/
services/ai/generative-ai

Dive deeper into this topic and learn

about the kinds of Generative AI work
Cognizant is doing with our clients:
https://www.cognizant.com/us/en/
insights/tech-to-watch/generative-ai

33 The Generative AI Handbook

Contributors
A special thanks to team members across the organization who have contributed to the thinking found here.

Adam Kincaid Claudio Gut Jason Vigus Matthew Mcnaghten Shantanu Sengupta

Andreas Golze Clayton Griffith Jatil Damania Matthew Smith Shareen Harvey

Andreea Roberts Dana Kovaleski John McVay Michael Francis Valocchi Shridhara Bhat

Anthony Lui Daniel Fink Juan Barrera Mike Turner Sid Stuart

Arturo Miquel Veyrat Daniel Teo Julian Krischker Narayanan TK Simon Baugher

Arunadevi Kesavan David Colon Justin Shepp Naveen Sharma Tara Whitehead Stotland

Balazs Vertes David Fearne Karthik Padmanabhan Oana Trif Tia Eady

Bobak Hodjat David Sauer Kathiravan Sadasivam Olivier Francon Todd Chapman

Bogdan Orasan Deviprasad Kuppuswamy Kirtikumar Shriyan Pramod Bijani William Mahony

Brian Delarber Girish Pai Kristi Blosser Rachit Gupta Yogesh Karve

Byan Mahin Huw Tindall Kritikumar Shriyan Rutvikkumar Mrug

Caroline Ahlquist Indranil Sen Mahadevan Krishnamoorthy Sankar Melethat

Christian Moos Ines Casares Margaret LeBlanc Saravanan Mohan

34 The Generative AI Handbook

References
1. https://www.cnn.com/2022/07/23/business/google-ai-engineer-fired-sentient/index.html 6. https://www.conference-board.org/press/us-workers-and-generative-ai
cognizant-technology-to-invest-1-billion-in-gen-ai-over-three-years-123080300490_1.html
7. https://gizmodo.com/amazon-chatgpt-ai-software-job-coding-1850034383
2. https://www.crunchbase.com/hub/artificial-intelligence-startups
8. https://www.bain.com/about/media-center/press-releases/2023/majority-of-health-system-
3. https://thehill.com/policy/technology/3954570-google-ceo-says-ai-will-impact-every- executives-believe-generative-ai-will-reshape-the-industry-yet-only-6-have-a-strategy-in-
product-of-every-company-calls-for-regs/ place/

4. https://tvpworld.com/71221233/ai-threatens-nearly-30-of-jobs-within-oecd-report 9. https://www.cognizant.com/us/en/gen-ai-economic-model-oxford-economics

5. https://devclass.com/2023/02/16/github-claims-new-smarter-copilot-will-block-insecure- 10. https://www.hyfe.ai/

code-writes-40-60-of-developer-output/

35 The Generative AI Handbook

About Cognizant
Cognizant (Nasdaq-100: CTSH) is one of the world’s leading professional services companies, transforming clients’ business, operating and technology models for the digital
era. Our unique industry-based, consultative approach helps clients envision, build and run more innovative and efficient businesses. Headquartered in the US, Cognizant is
ranked 185 on the Fortune 500 and is consistently listed among the most admired companies in the world. Learn how Cognizant helps clients lead with digital at cognizant.
com or follow us @Cognizant.

World Headquarters European Headquarters India Operations Headquarters APAC Headquarters

300 Frank W. Burr Blvd. 280 Bishopsgate 5/535, Okkiam Thoraipakkam, 1 Fusionopolis Link, Level 5
Suite 36, 6th Floor London Old Mahabalipuram Road, NEXUS@One-North,
Teaneck, NJ 07666 USA EC2M 4RB Chennai 600 096 North Tower
Phone: +1 201 801 0233 England Tel: 1-800-208-6999 Singapore 138542
Fax: +1 201 801 0243 Tel: +44 (01) 020 7297 7600 Fax: +91 (01) 44 4209 6060 Phone: +65 6812 4000
Toll Free: +1 888 937 3277

© Copyright 2024, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by any means, electronic,
mechanical, photocopying, recording, or otherwise, without the express written permission of Cognizant. The information contained herein is subject to change without notice. All
other trademarks mentioned here in are the property of their respective owners.