Gartner Business-Quarterly 3q2023

« «
Balance the Environmental Perils 4 Decisions to Make When

and Promises of Generative AI Creating a Generative AI Policy
AI Here
Enterprise leaders cannot ignore generative AI because Our article on balancing the environmental impact of GenAI
their rivals won’t. As the hype swirls — will it save untold encapsulates our practical perspective. Dream big: use these
hours of work? will it end life as we know it? — only one applications to suggest sustainable materials, markets and
thing is clear: It won’t go away. design. Minimize the negative: Consider so-called composite
and Now
AI, which uses knowledge graphs and causal networks for
Nearly half of more than 2,500 executives we polled said
optimal efficiency — tamping down the models’ voracious
they have been planning to spend more on AI of all kinds
appetite for energy. And buy clean power where you can.
since ChatGPT rolled out in late 2022. And separately,
dozens of executives have told us in conversations Company tactics come from ABB and HP Inc., while
that they want to move faster than they did on past important context emerges in our Q&A interviews with
Letter From the Editor AI initiatives and get something into production EU lawmaker Dragos Tudorache and decision-making
within a year. experts Daniel Kahneman and Olivier Sibony.
Their biggest question: What steps must we take over the GBQ helps you and your team align with others and
next 12 months so we don’t have to play catchup for years reach peak effectiveness, so your enterprise can achieve
to come — while protecting against pitfalls known and its goals, be bold and principled, and bring employees,
unknown? investors and the public along for the ride.
This issue of Gartner Business Quarterly will help you Our standing departments keep you up to speed — Cutting
act now. It’s packed with use cases along with guidance Edge is a look at provocative new data; Briefs offer short
for considering investments, enticing customers, and takes about smarter spending and planning, talent and
rethinking your talent strategy. Just as importantly, culture, growth and innovation, and data and technology.
you’ll find analysis of very real dangers ahead, some
We welcome your feedback. Please contact me at
that have hit the headlines and some that have not.
judy.pasternak@gartner.com.
— Judy Pasternak
Gartner Business Quarterly 3Q23 | 2

Credits
Research & Advisory Val Sribar, Executive VP Contributors
Mike Fang Kaelyn Lowmaster Sandy Shen
Executive Sponsor Scott Christofferson, Group VP Amisha Ajay
Peter Firstbrook Abbott Martin Matthew Shinkman
Global Chief of Research Chris Howard, Gene Alvarez
Ben Fisher Kate McLaren-Poole Jasleen Kaur Sindhu
Distinguished VP Analyst Vaughan Archer
Ron Friedmann Simon Mingay Gurbani Singh
Van Baker
Editor in Chief Judy Pasternak Tsuneo Fujiwara Kristin Moyer Shivendra Singh
Alexandra Bellis
Managing Editor Richard Eames Rupeesha Galhotra Anthony Mullen Lauren Smith
Kathon Betterton
Associate Editor Kelly Fiveash Dennis P. Gannon Grant Faulkner Nelson Marco Steecker
Erick Brethenoux
Contributing Editors Dian Zhang, Jonah Shepp Dan Gottlieb Kanwarpreet Singh Oberoi Stuart Strome
Frank Buytendijk
Dan Gutter Cian Ó Móráin Anushri Subramaniam
Project Manager Carolina Abalos Sharon Cantor Ceurvorst
Sonu Gupta Pedro Pacheco Noha Tohamy
Karen Carter
R&A Studio Irina Guseva Iga Pilewska Bettina Tratz-Ryan
Uma Challa
Creative Manager Steve Hennessey Ilona Hansen Helen Poitevin Kayla Velnoskey
Arun Chandrasekaran
Senior Designers Nicole Daniels, Divya Malkani Robert Hetu Amrita Puniani Sarah Watt
Jian Chen
Graphic Designer Tal Yaari Allysa Johnson Mark Raskino Lauren Wheatley
Laura Cohn
Associate Graphic Designer Akansha Garg Roma Kaur Gabriele Rigon Madison Wilson
Jeff Cribbs
Senior Editor Caroline Baker Rajesh Kandaswamy Eser Rizaoglu Shubhangi Yadav
James Crocker
Frances Karamouzis Jose Rosario
Meg Day
Lauren Kornutick Melissa Ruiz Hernandez
Michelle DeClue
Angela Kreiter Moutusi Sau
Stephen Emmott
Eliza Krigman Don Scheibenreif
Pieter den Hamer
Mike Lashinsky Tom Scholtz
Jeremy D’Hoinne
Leah Reidy Leachman Steve Shapiro
Ellen Eichhorn
Avivah Litan Tony Sheehan
Dan Essig
Table of Contents Gartner Business Quarterly 3Q23 | 3

Contents 31 Prepare for AI Regulation by
Addressing 4 Critical Areas 59 What Generative AI Means
for Your Talent Strategy
38 AI Systems ‘Need a Moral

Compass’ 64 Can AI Overcome Flaws
in Human Decision Making?
A Q&A With Dragos Tudorache, A Q&A With Daniel Kahneman and
Co-Architect of the EU’s AI Act Olivier Sibony
Feature Articles
«
10 Balance the Environmental

Perils and Promises 43 Generative AI Will Affect
Information Security 68 ABB Electrification Lights
Up ML Models for Financial
of Generative AI (and CISOs) Planning
Departments
«
17 The Whiteboard:
48 How Generative AI
Big Questions About How
to Invest in Generative AI
Can Help Meet Customer
Experience Expectations 05 The Cutting Edge 3Q23:
Cool New Data Points
26 4 Decisions to Make
When Creating a 53 Learn to Serve Your
AI-Powered Customers 72 Briefs:
Quick Takes on Fresh Research
Generative AI Policy Before They ‘Walk Away’
Gartner Business Quarterly 3Q23 | 4

The Cutting Edge 2023
Power Shifts Among Nations Prompt CEOs to Adapt Strategic Priorities

The actions they are taking depend on whether they believe global trade
is continuing much as it has or whether they see a reconfiguration.
Department
«
The Cutting Edge 3Q23

Response Strategies by CEOs’ Views of Globalization Scenarios
Percentage of Respondents (Multiple Responses Allowed)
Changing the Mix of Geographic 51% Reglobalization

(n = 250)
Sources We Rely on to Supply Us 37%
Cool New Data Points Continuing
50% p Globalization
Compiled by Laura Cohn and Dian Zhang Making Supply Chains More Resilient 35% (n = 52)
Changing the Mix of Geographic 48% p Statistically

Significant
Markets We Rely on Selling Into 37% Differences
44%
Making Supply Chains More Optimized
35%
Friend-Shoring, Reshoring, 32% p
Onshoring Key Operating Activities 12%
4%
No Changes
15% p
n varies
Q. What kind of changes, if any, are you making in response to continuing globalization/reglobalization/
slowbalization or deglobalization?
Source: 2023 Gartner CEO and Senior Business Executive Survey
Note: Respondents could choose from the following scenarios on globalization: globalization is continuing but changing,
as power shifts between major nations (reglobalization); globalization is slowing down (slowbalization); globalization is
continuing much the same as it has for decades; and globalization is reversing (deglobalization).
Table
Table of
of Contents
Contents Gartner Business Quarterly 3Q23 | 5
When Making a Strategic Pivot, Involving ERM Boosts the Chance of Success Despite Reports of ESG Backlash, Business Leaders Have Increased Commitments
Enterprise risk management teams can discuss relevant risk information, run scenario Almost three-quarters (72%) say they are building out their programs.
planning workshops, and map out and review the disruption response plans.
Impact of Consulting ERM on Strategic Pivot Outcomes Business Leaders Who Report Increased GC-Cited Examples of Increasing ESG
Percentage of Respondents Selecting ESG Commitments Commitments at their Organizations
Across the Past 12 Months
By Percentage Bringing on dedicated FTEs
Respondents who consult ERM in ESG-specific roles
are 19% more likely to achieve
intended pivot objectives. Increased ESG funding and/or
62% 28%
No establishing a dedicated ESG
budget
43%
Establishing ESG committees
and/or working groups
Issuing the organization’s

first ESG report
Investing in Board ESG education

Consulted ERM Did Not Consult ERM 72%
Yes Creating internal ESG intranet
n = 250 respondents
Q: What actions did you take that involved the risk and assurance functions: I consulted my organization’s ERM team. sites to communicate and
State your agreement with the following statements about the pivot, in retrospect: The pivot fully achieved its intended solicit feedback
objectives
Source: 2021 Gartner Resilience Survey n = 206
Note: Correlation is significant at the 0.01 level Source: Gartner 2023 ESG Goal-Setting Survey Source: Gartner

Nearly Half of Organizations Use Biometric Verification Even Top Performers Struggle With Feelings of Futility About Doing ‘Enough’
Companies are turning to biometric verification just as more jurisdictions plan Your top people need performance confirmation, not just conversation. Make sure you
to regulate it. To manage this risk, legal and privacy should work with information reward and recognize great achievements in real time to keep employees motivated.
security and the functions using biometrics to check how data will be stored and
protected, and how consent will be obtained.
Does Your Organization Use Biometric Verification? Prevalence of High Futility by Recent Performance Rating
Percentage of Employees Experiencing High Futility
Futility
48% • “I feel anxious about whether I am performing”
45% • “Reaching high performance feels hopeless”
5% for customers only • “Someone will always do more than me”
12% for customers and employees Does Not Meet Expectations 33%
Partially Meets Expectations 25%
28% for employees only Meets Expectations 16%
Exceeds Expectations 13%
No Yes Far Exceeds Expectations 21%

n = 58
Source: Poll from Privacy Strategy and Operations: Improving Privacy Metrics Webinar, 16 February 2023 n = 2,118 remote-capable employees who received a recent performance rating
Note: 7% Don’t Know Source: 2023 Gartner Achieving High Performance Employee Survey

Only Half of Boards Are Effective — But the GC Can Help Three Steps to Improve Board Operations
While boards face a host of challenges — whether it’s increased regulatory action, When general counsel take steps to make the board more effective, they play a big
heightened cybersecurity risk, or rising geopolitical tensions — there’s clearly room role in helping companies navigate an ever-shifting web of economic, social and
for improvement. geopolitical pressures.
Percent of Organizations With Effective Boards Maximum Impact of Each Factor on Board Effectiveness
Percent of Respondents Scoring an Average ≥6 on the 7-Point Board Effectiveness Indexa
53%
50% 50% 24%

Don’t Have Have Effective 17%
Effective Boards Boards
Director Quality and Information Access Meeting

Social Dynamics and Quality Coordination
n = 92 n = 92
Source: 2023 Gartner Corporate Governance and Board Management Benchmark Survey R2 = 55%
a
Board Effectiveness Index combines Executive Accountability and Board Oversight of Strategy and Risk. Source: 2023 Gartner Corporate Governance and Board Management Diagnostic Survey

Only 29% of strategists say their organizations change plans
Build a strategic plan

fast enough to respond to disruption.
In an increasingly volatile and uncertain world, strategy and
that works — even in

its execution can quickly go off course. Use the Gartner one-
page Strategic Planning template to:
volatile times! Set clear, strategic objectives for

your organization.
Combat seven mistakes

common to strategic planning.
Capture and communicate your plans

with an exclusive one-page template.
Download the Strategic Planning guide for your function.
Source: Gartner
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved. CM_GBS_2412262
Balance the Environmental Perils
and Promises of Generative AI
by Kristin Moyer, Sarah Watt and Pieter den Hamer
Generative AI consumes a lot of electricity and water. The

technology can accelerate positive sustainability and financial
outcomes, but executive leaders need to be selective about
use cases to make sure they are doing more good than harm.

The rapid adoption of ChatGPT and similar Recognize That Tech Has Reduce Generative AI’s Energy Footprint
general-purpose tools has elevated a Consumption Problem Executive leaders can make their organization’s generative
the negative environmental impacts of Generative AI relies on huge models trained with massive AI more environmentally friendly by leading efforts to:
generative AI from an insider topic to an amounts of data, making it thirsty for cooling water and • Make it as efficient as the human brain. One of the
immediate concern for executive leaders. hungry for electricity. In some cases, the technology can reasons the brain is so energy-efficient is that it organizes
Plenty of seemingly sustainable use cases consume vast quantities of both. Electricity-related GHG knowledge in network structures. Consider adopting
emissions will decline in the long term, as the generation so-called composite AI, which uses similar techniques
powered by this emerging technology capacity of renewable energy grows. However, more such as knowledge graphs, causal networks and other
will do more harm than good in terms powerful generative AI models will likely require increased symbolic representations to complement the current
of greenhouse gas (GHG) emissions computing capabilities. brute force, deep learning method.
and electricity and water consumption. The problem of tech-related electricity and water • Put it on an electricity and water diet. Monitor
But generative AI is not automatically consumption goes well beyond generative AI. Information energy consumption during machine learning.
or entirely bad news for sustainability. and communications technology (ICT) consumes less Improve awareness across the company about AI’s
than 1% of global electricity today, but it is forecast to use environmental footprint. Stop training AI as soon as
For example, Orbital Materials, a UK-based startup, improvements flatten out and no longer justify the costs
6.4% of total output by the end of the decade.2,3,4 If new
is seeking to apply it to accelerate the development of continuing. Keep data for model training local, but
power-hungry technologies scale quickly, ICT energy
of clean energy technologies.1 share improvements at a central level; this federated
consumption could be even higher.
machine learning reduces electricity consumption and
If used in the right way and with human oversight,
By 2030, 75% of CIOs will experience electricity bolsters data privacy. Reuse models that have already
generative AI can potentially help companies mitigate
constraints.5 For example, Singapore “implicitly imposed” been trained, contextualizing them if needed. Use more
sustainability risk, optimize costs and drive growth.
a moratorium on building new data centers in 2019 (but energy-efficient hardware and networking equipment.
Executive leaders should balance the perils and promises
lifted it in 2022).6,7 The Dutch government tightened rules Balance “follow the sun” data center workloads (which
of this technology by simultaneously:
for hyperscale data centers in 2022.8 As the needs of are better for clean energy production) with “unfollow
• Recognizing and reducing generative AI’s energy footprint technology and society compete more intensely with each the sun” measures (which are better for water efficiency).
• Identifying, evaluating and prioritizing its environmental other, CIOs do not want to be caught in the position of
sustainability use cases vying with local communities for finite resources.

• Run it in the right place and at the right time. When and where a generative AI workload Identify Potential Environmental Sustainability Use Cases
happens matters. The carbon intensity of the local energy supply varies by country,
generating authority, time of day, weather conditions, transfer agreements, fuel supply We picked out valuable examples of generative AI use cases that will accelerate
and other factors. Best practice is to use energy-aware job scheduling for generative AI, environmental sustainability by mitigating risks, optimizing costs or driving growth.
along with carbon tracking and forecasting services to reduce related emissions. We then developed descriptions of each that incorporated some suggestions from
ChatGPT. We found that the tool frequently overestimated its abilities but that we
• Aim to buy new clean power where you plan to consume it. Not all renewable electricity
underestimated its potential.
purchases are equal. Buying renewable energy certificates (RECs), which reduce or
offset GHG emissions, does not guarantee that the clean energy was actually consumed Mitigate Risks
or that any new renewable power generation capacity was built to match increasing
• Comply with regulations. Generative AI can identify relevant sustainability laws,
demand. Power purchase agreements (PPAs) can help but are not always available.
standards, directives and reporting requirements, including updates over time. It can
The Greenhouse Gas Protocol, a global standard-setting body, is considering requiring
interpret complex regulatory language and provide explanations in simpler terms.
companies to provide more detailed analysis of clean power by location, time of day
Generative AI can develop an action plan to achieve compliance, including insights
or both.9
on how to establish internal controls and auditing. It can develop training materials
Until then, executives should procure PPAs when possible, or source RECs from recently to educate employees on specific regulations.
commissioned projects (thus adding new renewable energy to the grid) from the region • Avoid stranded assets. Generative AI can identify trends, regulations and technology
where their company will consume the electricity. For example, IBM,10 Walmart11 and disruption that increase the risk of stranded assets.
Salesforce12 have policies to purchase renewable energy generated in the same grid
region as they consume it and/or to add new clean power to the system. • Develop organizational sustainability policies. Generative AI can help employees
identify sustainability policy best practices; develop a framework for such policies;
and draft statements, guidelines, procedures and protocols.
• Develop sustainability KPIs. Generative AI can help identify key performance
indicators to evaluate sustainability efforts and progress.

Optimize Costs Drive Growth
• Support decision making. Generative AI can analyze internal sustainability data and • Discover alternative resources and materials. Generative AI can provide suggestions
identify patterns, trends, areas for improvement, feasibility, risks and benchmarks that for sustainable substitutes for conventional inputs; insights on technological innovation
support decision making. It can provide insight into how organizational decisions will such as nanomaterials; and information on availability, performance and environmental
impact sustainability and forecast likely future performance. Enterprises can therefore impact.
plan and select optimal pathways to reach GHG emissions reduction goals. • Support sustainable product development. Generative AI can help with new,
• Improve corporate sustainability communications. Generative AI can generate press sustainable product ideation and insights on its viability; market research on trends;
releases, website content, articles, social media posts and campaigns, blogs, and suggestions on prototyping; sustainable design principles; information on patents,
external and internal communications on sustainability. It can suggest key messages trademarks or copyrights; competitive analysis; and pricing advice.
and best practices for conveying information. ChatGPT can develop frequently asked • Leverage generative design for buildings, parts and products. Generative AI can
questions and answers. suggest alternative design approaches, and parameter and constraint options, as well
• Engage supply chain partners. Generative AI can develop criteria for supplier as provide insight on design strengths and weaknesses.
assessment, create a code of conduct, suggest engagement strategies, develop • Accelerate sustainability innovation. Generative AI can support ideation, opportunity
training content, provide suggestions for incentives and corrective actions, and answer assessment, risk and compliance mitigation, and business case development.
questions.
• Train employees. Generative AI can develop employee training materials, create
sustainability awareness communications, provide insight on sustainable work practices,
explain sustainability concepts, answer employee questions and support learning by
recommending materials and courses.
• Embed generative AI in products to make them more sustainable. Generative AI can
suggest sustainable product uses to customers by automating reply generation, using
the right response tone and gamifying the customer experience through rewards.

Evaluate the Positive and Negative »Figure 1. Analysis of Generative AI Use-Case Examples
Impacts Higher Lower
Executive leaders weighing generative AI use cases Business Value Negative

for sustainability should analyze their: Financial Sustainability Environmental
Feasibility Impact
Value Value
• Business value — Financial and sustainability benefits,
Comply with regulations
Mitigate Risks
as well as feasibility (the level of difficulty and cost to
implement the use case) Avoid stranded assets
• Negative environmental impact — As measured by Develop organizational sustainability policies
GHG emissions and electricity and water consumption Develop sustainability KPIs
We evaluated sustainability use cases, employing general- Support decision making
Optimize Costs
purpose generative AI tools (such as ChatGPT) based Improve corporate sustainability communications
on a combination of primary and secondary research,
Engage supply chain partners
proprietary data and analyst experience (see Figure 1).
Train employees
Embed GenAI in products to make them more sustainable
Discover alternative resources and materials
Drive Growth
Support sustainable product development

Leverage generative design (buildings, parts, products)
Accelerate sustainability innovation
Source: Gartner

Prioritize Your Investments »Figure 2. Prioritization of Generative AI Use-Case Examples
Mitigate Risks Optimize Costs Drive Growth
Based on this assessment, executive leaders should categorize
Higher
generative AI use cases into three levels of priority (see Figure 2):
Reduce Risk and
• Invest now — Act immediately to advance use cases that will drive Do Not Invest Invest Now
Energy Use First
business value with a relatively low negative environmental impact.
• Reduce risk and energy use first — Lower intellectual property Discover Leverage Develop
and proprietary data risk and/or improve the energy efficiency alternative generative sustainability
Business Value and Feasibility

resources and design KPIs Develop
of generative AI before proceeding.
materials Avoid organizational
• Do not invest — Avoid use cases that could damage business Train Engage supply stranded sustainability
value or the environment. employees chain partners assets policies
Accelerate
Generative AI can help organizations accelerate positive sustainability Support sustainability
decision innovation Comply with
outcomes, provided they leverage only those use cases that create
making regulations
more value than they destroy. Support
sustainable
product
Embed GenAI
development
in products to
1
DeepMind Alum Wants to Use AI to Speed the Development of Green Materials, Bloomberg.
make them more
2
The Decadal Plan for Semiconductors, Semiconductor Research Corporation.
Improve corporate sustainable
3
World Energy Outlook 2022 — Stated Policies Scenario, International Energy Agency.
sustainability
4
Maverick Research: Net Zero Will Stall Tech Growth and Innovation.
communications
5
Gartner prediction based on Semiconductor Research Corporation’s Decadal Plan Update
and the International Energy Agency’s World Energy Outlook 2022 — Stated Policies Scenario.
Lower
6
Singapore Hits Pause on Building New Data Centres; Short-Term Rents Up, The Business Times.
7
Singapore Authorities Invite Applications for New Data Centers, DCD.
Higher Lower
8
Dutch Call a Halt to New Massive Data Centres, While Rules Are Worked Out, DutchNews.
Negative Environmental Impact (Electricity and Water Use)
9
Survey on Need for GHG Protocol Corporate Standards and Guidance Updates, Greenhouse Gas Protocol. Source: Gartner
10
Energy and Climate, IBM.
11
Climate Change, Walmart.
12
Inside Salesforce’s Clean Energy and Carbon Programs, Salesforce.

The Future of AI:
Reshaping Society
Can the autonomous decisions made
by AI for us, be compared to human agency?
With recent developments in generative AI, such as ChatGPT, AI is no longer

merely a technological or commercial instrument. The age of AI has well and
truly arrived, and it is about to transform society as a whole.
Download the research for more on:
What is up with AI: What is next with AI: What else with AI:
Introduction to some Future AI scenarios More insights on
fundamental issue through 2033 the impacts of AI
Download Research
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved. CM_GTS_2450512
The Whiteboard:
Big Questions About How
to Invest in Generative AI
by Van Baker, Erick Brethenoux and Arun Chandrasekaran
Contributions by Jeff Cribbs, Meg Day, Ellen Eichhorn, Dennis P. Gannon, Dan Gottlieb, Robert Hetu,
Grant Faulkner Nelson, Pedro Pacheco, Eser Rizaoglu, Moutusi Sau, Steve Shapiro, Tony Sheehan,
Jasleen Kaur Sindhu, Lauren Smith, Marco Steecker, Noha Tohamy and Shubhangi Yadav

Suddenly, executives How are other companies advancing with generative AI?
recognize how easily they can
take advantage of the most
disruptive set of technologies s
f ive o rg anization
and capabilities to hit the One in
e already
global market in decades. say they’r g it
il o t in g o r producin
p
are in the 1
But don’t get carried away and 70% e.
v e s t ig ation stag
in
by the hype surrounding
generative AI. Concentrate »Current Phase of Generative AI
on practicalities such as use Percentage of Respondents
cases, cost and vendors. 70% 19% are currently piloting

or producing generative AI.
Let’s head to the whiteboard
and sketch out how 15%
4% 10%
organizations can make
smart moves as they pilot Investigation Piloting Production None of the
and scale their initiatives. (Exploration) Mode Mode Above or Not
Mode Applicable
n = 2,544
Source: Beyond the Hype: Enterprise Impact of ChatGPT and Generative AI
Webinar Polls; 30 March and 21 April 2023

What new questions must we ask about possible investments?
Impact Process & Infrastructure Governance
he Stakeholders: Who is
n : H ow might t likeliest to be impacted?
Vis io by Protection:
s created
capabilitie t our strategy?
ac • How will we
GenAI imp ary
safeguard propriet
Processes: Which ones will we ?
and sensitive data
need to transform or create?
will
• What disclosures
Benefits: What new we need to make?
re:
opportunities and System architectu
GenAI to
capabilities will it present? How will we train
tructure?
leverage our infras
Prevention:
Objective: What are we Data: What data is available to • Who will be accountable
trying to gain by applying feed into GenAI? How much is for GenAI’s outputs, and to
it to this problem? there? Does it have sufficient what degree?
depth (i.e., metadata)? • What governance structures
will we need to detect and
Image generated using DALL-E, https://openai.com/dall-e-2.
respond when it acts in
Adoption: How will staff Skills: What skills are required to unexpected ways?
react to its use? iterate and collaborate with this tool?

What are some emerging functional use cases?2
KPIs and what if analysis:

Explain supply chain metrics Generative value
and their drivers, and messaging: Rapidly
evaluate options through Vendor engagement turn customer data into Generative business intelligence:
intuitive text-based queries and compliance: customized content Interpret and synthesize sales
Summarize contracts, data faster — add traditional AI
ongoing contract techniques (i.e., ML and optimization)
Complex model Supply compliance, vendor Digital sales Sales and to uncover hidden correlations and
explainability: Text- Chain onboarding/Q&As humans: Augment Customer Service gain prescriptive insights
based explanation sales bots and
of recommended avatars — they have
production, route Regulatory compliance: Explain trade, some human seller Sales learning gymnasium:
planning or customer healthcare, sustainability regulations in characteristics Use GenAI role-play bots to
fulfillment decisions simple Q&A text format streamline sales enablement and
L&D workloads, and quickly design
engaging learning experiences Instacart Inc. is
adding ChatGPT
chatbot technology
to its grocery-
delivery app.3

What are some emerging functional use cases?2 Users should have
Competitive appropriate security
Management
intelligence: and privacy policies
updates: Explain
Summarize in place for these
forecast and budget
public financial types of use cases
Company reporting: variances for FP&A
HR operations: disclosures (e.g.,
Create initial drafts to use in business
competitors’
• Augment and improve staff of management reviews; summarize
quarterly earnings
productivity in classifying, discussion and trends and insights
transcripts) Internal knowledge
summarizing and responding analysis, financial for the board and
and industry management:
to employee queries footnotes, etc., executives
publications to Ingest unstructured/
• Automated drafting of policies for quarterly/annual compile reports
filings nonintegrated
and document generation Contract and faster datasets and
document review: artifacts (e.g.,
Scan contracts for business cases,
Tailored job HR Finance errors and specific Strategy project plans) to
descriptions: terms; summarize organize and easily
Quickly create more and categorize query expertise
compelling and HR chatbots: Use these Coding assistance: documents for sorting, Assist in curating
inclusive text to better tools, with enhanced Translate code review, retrieval agendas and
attract diverse talent NLP capabilities from from older coding frameworks: e.g., Planning assistance:
with in-demand skills generative AI, to languages, like for annual planning Pre-fill templates
COBOL, into modern offsites or defining (e.g., auto-generate
• handle routine
programming Zoom Video industry ecosystems baseline Five Forces
employee requests
languages, like SQL, Communications is and SWOT analysis)
using human-like text
KnowledgeSQL, and using generative AI to to accelerate BU
• provide a better user Python create first drafts of leaders’ planning
experience tax documentation.4 efforts

What are some possible industry use cases?2
Financial Services Healthcare and Life Sciences Education
AI frontline co-pilot: Chat interface Conversational patient Scientific literature discovery: Student tutors: Conversational
helps client-facing employees get self — triage and checking LLMs help scientists identify UI to support personalized
important information faster symptoms: Chatbot makes relevant research, extract learning
suggestions and guides patients insights, aggregate findings and
regarding acute symptoms, generate new hypotheses Language training: AI reading
GPT-4 chronic condition management, and speaking companion
Morgan Stanley is training 5 Coding assistant for mainframe
ors. health and wellness activities,
to help its financial advis support: Helps software Faculty assistant: Accelerate
or behavioral health needs
developers generate, test, authoring of quizzes, tests,
Auto-composition of clinical debug code snippets in presentation materials,
Compliance and regulatory monitoring: messages: Automatic replies languages common to mainframe curricula, lesson plans,
Assist in verifying communications with based on content and tone technologies, like COBOL — often feedback, student referral letters
clients against internal codes and rules of patient message, accessible used in U.S. healthcare payers’
clinical data, and clinician’s claims processing systems Virtual student assistant:
Personalized customer support: tone and preferences Chat interface to integrated
Recommendations for contact center Consultative population student data
agents and relationship managers based on health analytics: Users ask
customer profile, needs and expectations Mass General plain language questions of a Student recruitment/
Brigham, a he
care system in alth report or dashboard in areas like enrollment/persistence:
the U.S., is test
Claims management: Individualized generative AI ing population health, costs and care Including nudging students
for patient port
suggestions/explanations on claims coverage messages and a l activities toward course completion
clinical notes.6
and applicant-friendly reasons for denials

What are some possible industry use cases?2
Retail Manufacturing Transportation
Supply chain optimization: Education and training: Direct an Customer interaction:

Improve predictions for employee with or without relevant Use of LLM chatbots
Tesco is using GenAI sourcing and procurement, technical knowledge to verify a factory-
es
and other technologi logistics, transportation, and floor machine in their chosen language(s) Maersk is using ChatGPT on its
to enhance customer collaboration with suppliers website to auto-generate FAQs
experience, predict Product innovation: Suggest alternative
and improve search accuracy.8
umer
demand, analyze cons Conversational chat interface: ingredients and packaging based on
vi or an d preven t fra ud.7 user sentiment and aggregated trends/
beha Interact with customers and
associates, which may include shopping patterns Vehicle damage estimation
facilitating a transaction -- enable for insurance claims: Help a
Enhanced search and upselling: Digital product interaction: Download smartphone camera recognize
human customers to converse
Improve customers’ abilities new behaviors/capabilities to digital damage more precisely even where
via their platform of choice
to find what they are looking for, products based on aggregated voice visibility and contrast are poor
and encourage more expensive Associate hiring, onboarding: feedback
purchases or add-ons Estimation of vehicle resale value:
Enhance recruiting and training
Product servicing: Help humans and Use GenAI on computer vision to
Social media customer through interactive individual AI agents in continuously diagnosing enable a smartphone camera to
sentiment: Quickly monitor experiences issues; order parts, complete assess value more accurately
customer and influencer social programmable maintenance or
media content, spot trends and schedule recommended servicing Assessment of mechanical
sentiments, predict outcomes needs. (Goal: reduce unplanned condition: Enable more precise
and inform future decisions downtime) evaluations

How much will generative AI cost?
Gartner Generative AI Framework The more you customize, the more it costs.
From almost nothing to many millions of dollars, depending on:

Enterprise Data Model Fine
Custom Model
PII/Enterprise and Model Tuning to is Created for
use case scale company requirements risk appetite Desired Non-Sensitive
IP Included Instructions Improve Use/
Content Type Text Unique Use Case
Needed Performance
Small and midsize enterprises … Larger organizations … LLM w/Privacy, Enterprise

LLM With Privacy, Policies, Data Hosted Custom
… could use the free versions of public, • … or enterprises that want to Enterprise No Controls LLM With
Privacy Policies and Data and Added Differentiated
openly hosted applications such as analyze or use their own data Control Level Needed Model Layers Model
OpenAI or pay low subscription fees more
• … or are looking to invest in LLM API with Modified LLM
LLM API Standard Privacy, With Transfer Custom Build
GenAI capabilities will increasingly be significant process or market LLM Using
Solution ChatGPT Accessed via Policy and Per Learning/Added
built into everyday software products disruption Enterprise
Needed Application Incident Data Layers to Adjust
• Effectively a “free,” or close to free, Frame Injection Model Output Accessible Data
will need higher levels of security,
tier with productivity benefits
IP and privacy protections …
• Vendors will pass costs on to Cloud Instance Licensed
customers as part of bundled … requiring a range of custom Cloud With LLM Customizable Custom Build
Required/ Open AI Proprietary
incremental price increases services. Recommended Instance With APIs, Prompt Model/
Hosted Engineering, Model, Data, ML
LLM APIs Proprietary
Technologies Application Platform
Custom Policies, Model, Data, ML
Do we build our own licensed, customizable and proprietary models with Indexed Database Platform
foundation model platforms that will require working with vendors and partners?
Cost $ $$$$$$$$$$$$
Negligible $$$$ $$$$$$$

What does the vendor landscape look like?
1
2023 Gartner Beyond the Hype:
Enterprise Impact of ChatGPT
and Generative AI Webinar Polls. This
webinar was held on 30 March and
Generative AI Apps: Horizontal Apps Vertical Apps Vendors include: Jasper AI, Harvey, 21 April 2023 with 1,465 and 1,079
Horizontal or vertical Rytr, Mem, Notion, Glean, Algolia, respondents to the polling, respectively,
— available from both Marketing Sales Productivity Media BFSI Healthcare GitHub for a total of 2,544 responses. Results
of these polls should not be taken to
incumbent software represent all executives, as the survey
vendors and new SaaS Finance Customer Knowledge IT Life
Legal responses come from a population
start-ups Service Mgmt. Sciences that had expressed interest in AI by
attending a Gartner webinar on the
subject.
Generative AI 2
Sourced from news articles and
Engineering Tools: to Data Centric Model Centric Vendors include: Pinecone, conversations with executives.
operationalize and run Model Model Weaviate, Arize, Credo.AI, Snorkel
3
Instacart Joins ChatGPT Frenzy,
generative AI models Prompt Vector DB Fine- AI Trust, Risk Adding Chatbot To Grocery
Deployment API AI, Scale AI
safely and efficiently in Engineering Tuning & Security Shopping App, The Wall Street
Orchestration Mgmt (TRiSM) Journal.
production 4
Businesses Aim to Harness
Vendors include: OpenAI, Generative AI to Shake Up
Generative AI Models: Microsoft, Google, Cohere, Accounting, Finance, The Wall
Street Journal.
Versatile ML systems Model Hubs Anthropic, AI21 labs, AWS, IBM,
that can create human- Foundation Models Domain Models
5
How Morgan Stanley Is Training GPT
Alibaba, Tencent, Baidu, Hugging To Help Financial Advisors, Forbes.
like artifacts Face, EleutherAI, Stability AI 6
Mass General Brigham Creates
Video With Generative AI; Testing
It to Draft EHR Messages, Clinical
Infrastructure:
Vendors include: Nvidia, AMD, Intel, Notes, Becker’s Hospital Review.
Infrastructure Compute Network Storage AWS, Microsoft, Google Cloud 7
How This Company is Redefining
training and running Retail Experience with Generative
generative AI models Note: Vendor lists are non-exhaustive and AI, Analytics India Magazine.
do not imply any Gartner endorsement or 8
The AI chatbot ChatGPT is a game-
ranking of the providers mentioned. changer! maersk.com

4 Decisions to Make When
Creating a Generative AI Policy
by Laura Cohn, Stuart Strome, Anthony Mullen, Avivah Litan and Lauren Kornutick
Corporate leaders have been grappling in 2023 Based on our analysis of AI policies already instituted by companies and
city governments, the GC should direct organizations to consider the
with how to manage workplace use of generative
following questions before establishing a policy:
AI applications such as ChatGPT and Bard.
1. What is our risk tolerance for use of generative AI?
Though some of the most prominent voices in generative AI development
2. What restrictions should we put in place to mitigate risks — and how
have warned that these products need close oversight and regulation, that
will those differ for publicly available applications (such as OpenAI’s
task for now falls to the general counsel and C-suite colleagues who will
ChatGPT) versus generative AI models tailored to our business needs?
lead implementation or use of the new technologies.1,2
3. Who has the authority to make decisions on generative AI use?
Government requirements do loom on the horizon — the European Union,
4. What information do we have to share, and with whom?
China, the U.S., the U.K. and Canada are all working on rules and guidance
now. Having policies in place will also prepare your enterprise for whatever
measures authorities enact in the future.

What Is Our Risk Tolerance the brand might falter or even evaporate — a result that Compile a list of these more detailed use cases and
might constitute a must-avoid outcome. What if, however, organize them according to perceived risk — both the
for Use of Generative AI?
an LLM-powered chatbot provides potential ideas for uses likelihood and severity of the threat. If the organization
To get started on determining the company’s risk of your product, and some of those, while not dangerous, already has a team managing AI use cases, suggest the
tolerance, legal leaders should borrow a practice from are simply not optimal? That may fall into the “acceptable group extend its existing work to include policy and risk
enterprise risk management and guide a discussion with risk” category. considerations. Some potential risks to look out for include
senior management on “must-avoid outcomes.” Such unreliable outputs, outputs that violate copyright or IP
scenarios are defined as severe but avoidable events What Restrictions Should protections, and regulatory compliance. Tools from a
that will prevent a company from achieving its strategic trusted vendor (e.g., Microsoft, Google, Amazon) may pose
We Put in Place?
goals. The C-suite should use these sink-the-enterprise lower risk than those from publicly available applications.
possibilities as the basis for establishing guardrails. Now it’s time to dig a little deeper. Reach out to
For higher-risk situations (e.g., producing content
departments that may adopt generative AI models such as
First, you’ll need to know the potential applications of for customer consumption), consider applying more
marketing, data and analytics, sales, and customer service,
generative AI models within the business. Once you have comprehensive controls, such as requiring approval
to name a few. Ask them about timelines for adoption of
that list, consider the potential results and sort them into from a manager or AI committee. And in the highest-risk
generative AI models to help understand how urgent it
must-avoid outcomes and those that carry acceptable risk cases, you may want to think about outright prohibition.
is to address those use cases in your policy. And make
given the likely benefit of AI. For lower-risk use cases (e.g., coming up with ideas for
sure to ask whether they have witnessed, are aware of,
a fun activity during a staff off-site, or translating jargon
If your company makes consumer products, for instance, or have any planned uses in their part of the business for
or regulations written in a foreign language), consider
and a generative AI-powered application gives legally, generative AI applications that train on external data rather
applying basic safeguards such as requiring human review.
financially or morally unsound advice, consumer trust in than customized information from your own business.
Sample categories, examples and approval requirements
can be found in Figure 1.

Legal leaders should apply different restrictions »Figure 1. Guidance on Use Cases and Restrictions of Generative AI (Illustrative)
to publicly available applications than they do for
Category Examples Who to Notify
more business-specific applications from a trusted
third-party provider. Irrespective of provider, legal Brainstorming, translations, trusted software within
Approved e.g., manager, HR
leaders should consider some universal safeguards. the enterprise’s security perimeter.
For instance, since all generative AI models may Requires
Purchasing new software, using algorithms in HR or another department. e.g., legal, procurement
produce incorrect information or exhibit bias, be Approval
sure to include close monitoring for “hallucinations” Putting confidential company information, client data or personally Follow misconduct
— a common term of art meaning erroneous claims Prohibited
identifiable information into AI platforms that don’t protect it. reporting procedure
— and checking for skewed viewpoints.
Source: Gartner
An example of this in practice comes from the
City of Boston’s AI policy, which warns employees
who use AI applications to have a human being
»Figure 2. City of Boston’s Guidance on Human Review of Generative AI
scrutinize any output (see Figure 2).3
Restrictions on publicly available applications will

need to be more comprehensive. For example, they
often use information entered by users to improve
the model and that means sensitive material must
stay out of the mix.
Source: The City of Boston

When drafting policies, call out employee Consider defining low risk, acceptable Be clear if there are uses that do not need How Much Information
obligations in simple, unambiguous use cases directly in your policy (e.g., approval, specify what they are directly in
Should We Disclose?
language. Consider including a set producing a high-level summary of the policy, and provide examples — but
of “Dos and Don’ts” such as: secondary research on a non-sensitive also make sure employees know where to A critical tenet common of proposals
• Don’t input any personally identifiable topic) to provide more clarity. take their questions. Communicate which in a variety of jurisdictions, including
information. uses do need signoff to the workforce, the standard-setting EU, will be that
• Don’t input any sensitive information.

Who Has the Authority clearly document the role that can provide companies should be transparent about
approval, and list the relevant contact their use of AI. And consumers want to
to Make Decisions?
• Don’t input any company IP without information. know if you use generative AI applications
checking with the team charged with Work with functional, business and to craft corporate messages, whether the
Common points of contact for policies
reviewing use (e.g., the security team). senior leadership stakeholders to: information appears on a public website,
include:
• Do turn off history if using external • Align on who is responsible social channel or app.4
• Legal
tools that enable that choice. for mitigating a specific risk Therefore, it’s sensible to include a policy
• Information security
• Do closely monitor outputs, which • Review duties (e.g., whose responsibility of disclosing the use and monitoring of
could contain factual errors and it is to assess a novel use case or the • IT generative AI technologies to both internal
biased or inappropriate statements. output of an approved use case) • The company AI committee and external stakeholders. The GC should
(this might include the CEO, chief legal help companies consider:
Finally, don’t be overly restrictive. • Understand decision rights
Banning use of these applications (e.g., who should approve new use cases officer, chief information officer and • Whom to disclose to (customers,
outright, or applying hard controls (e.g., or exceptions), as well as any other other business leaders) employees, third parties or all three)
restricting access to websites), may result related activities, such as assessing risks, • What they need to disclose
in employees simply using them on their monitoring, imposing disciplinary actions
• What they need to be open about
personal devices. or fielding questions
beyond use (e.g., employee monitoring
for non-approved use of AI)

In practice, that means requiring employees to make »Figure 3. City of Boston’s Generative AI Policy’s Disclosure Rules
sure the output is recognizable as machine generated by
clearly labeling text. One option to consider: watermarks
in AI-generated images to the extent technically feasible.
Here, again, Boston’s guidance provides an example.

The city mandates agencies and departments not
only reveal the use of generative AI, but also name
the application used. Boston’s guidelines include an
explanation of why disclosure is important — as well as
suggested language to let users know how the content
was created (see Figure 3).3
A Final Tip: Build in Flexibility Source: The City of Boston
Organizations that have established guidelines for

employees typically make sure to note that they will
be updated from time to time as: 1
“The Godfather of AI” Warns of AI Possibly Outperforming Humans,” npr.com.
2
OpenAI CEO in “historic” move calls for regulation before Congress, axios.com.
• Generative AI develops 3
Boston Isn’t Afraid of Generative AI, wired.com.
4
2023 Gartner Consumer Community Survey. Gartner Consumer Community (n =320, 10 to 17
• Regulatory changes occur February 2023). While the Gartner Consumer Community (n = 500) resembles the U.S. general
population, the data cited is based on the responses of community members who chose to take
• Company risk appetite shifts each activity. These samples may not be representative of the general population and the data
should only be used for directional insights.

Prepare for AI Regulation by
Addressing 4 Critical Areas
by Laura Cohn
with contributions from Dan Essig, Mike Fang, Melissa Ruiz Hernandez, Lauren Kornutick, Anthony Mullen, Sandy Shen and Stuart Strome
Even before large language models (LLMs) Although lawyers specializing in privacy, cybersecurity EU Regulation May Influence Others
and AI told us the rules in the EU and Canada may not
took the world by storm, the number of The EU’s proposed AI law, first put forward in April 2021
take effect until at least 2025, considering what is under
enterprises deploying AI was on the rise.1 discussion now will prevent companies from running into and still under discussion, builds on the GDPR. Like that
But the public launch of ChatGPT and trouble with regulators later. China announced in July that regulation, lawyers and business leaders expect it to
other LLM applications catalyzed both AI interim regulations will take effect on 15 August, 2023.2 become a roadmap for other countries to follow.3,4,5,6
Elsewhere, Canada is weighing the Artificial Intelligence
investment and global regulatory efforts. Two of those proposals contain fines for noncompliance; and Data Act (AIDA), and China has interim rules for
The general counsel (GC) can prepare senior leaders maximum penalties in the EU’s proposed AI law exceed generative AI.7,8,9 Both the U.S. and the U.K., meanwhile,
and the board even as the rules come into focus. A close the attention-grabbing fines in the bloc’s General Data have issued guidance in the form of AI white papers
reading of proposed and new regulations and guidance Protection Regulation (GDPR). Even if authorities don’t (see Figure 1).10,11
— in the EU, Canada, China, the U.S. and the U.K. — reveals impose the highest possible fines, companies need
shared underlying principles. The GC can help organizations defensible AI oversight and risk management that enable
develop corporate AI strategy by prioritizing them. innovation and experimentation.

»Figure 1. Global AI Regulation and Guidance at a Glance
White House China’s Interim Measures
Canada’s Draft Blueprint for the Administration
Artificial Intelligence for AI Bill of of Generative Artificial
EU’s AI Act and Data Act Rights* U.K.’s AI White Paper* Intelligence Services
Release
April 2021 June 2022 October 2022 March 2023 July 2023
Date
“... providers placing Companies that “design Guidelines “We will not assign rules or Only applicable to services
on the market or or develop a high-impact apply to the risk levels to entire sectors accessible to the general
putting into service AI system” or “make a public and or technologies. Instead, public within China
Scope AI systems** in the high-impact AI system private sectors. we will regulate based on
Union.” available for use” or the outcomes AI is likely
“manage the operations to generate in particular * Nonbinding government guidelines
of an AI system.”*** applications.” ** The EU AI Act defines an AI system as “software that
is developed with one or more of the techniques and
approaches listed in Annex I and can, for a given set of
€30 million or CA$25 million or up to Not applicable Not applicable Not applicable human-defined objectives, generate outputs such as
Maximum up to 6% of total 5% of global revenue6 content, predictions, recommendations or decisions
influencing the environments they interact with.”6
Penalties worldwide annual *** Canada’s draft act states that “high-impact AI systems”
turnover5 will be defined by forthcoming regulations to account
for what’s laid out in the EU’s AI Act and technological
advances. Among the potential considerations:
Effective whether the system will impact human rights and
To be determined To be determined Not applicable Not applicable 15 August 2023
Date whether opting out is not possible.7
Source: Gartner

According to our analysis of the proposals, our interviews and our best- Be Transparent About AI Use
practice research, GC can get ahead by directing their organizations
to prioritize the following (see Figure 2):
Providers shall ensure that AI systems … are designed and developed
• Transparency in such a way that natural persons are informed that they are interacting
• Risk management with an AI system….6
• Governance, including accountability and human oversight EU’s proposed AI law
• Data privacy
A critical tenet common across jurisdictions is that companies should be open about using AI.
»Figure 2. Common Principles in Global AI Regulations and Guidance One way that lawyers we interviewed suggest organizations should prepare: Disclose
AI use in marketing content and the hiring process. At a bare minimum, legal leaders
can help by updating the privacy notices and terms and conditions on their company’s
website to reflect AI use. But it’s better to develop a separate section on the organization’s
Transparency Risk Management online “Trust Center.” Or post a point-in-time notice when collecting data that specifically
discusses the ways the organization uses AI, assures individuals that their privacy rights
won’t be negatively impacted and, just as important, makes clear that such use will deliver
value to the customer or individual.
Assess whether your company needs stand-alone AI guidelines. Organizations such as IBM
make it easy for clients and customers to understand how they are using AI by linking their
Governance,
Including Data Privacy stand-alone AI policy to the code of conduct, IT security policy and privacy policy.12
Accountability and
Human Oversight Let departments working on AI initiatives know they can help the business avoid privacy
risk by being transparent from the start. Otherwise, legal and privacy leaders risk finding
out about such projects only when they are finalized.
Source: Gartner

Risk Management Must Be “Continuous” Governance Must Include Human Oversight and Board Updates
The risk management system shall consist of a continuous iterative process Human oversight shall aim at preventing or minimizing the risks to health,
run throughout the entire life cycle of a high-risk AI system, requiring regular safety or fundamental rights that may emerge when a high-risk AI system
systematic updating.6,13 is used.6
EU’s proposed AI law EU’s proposed AI law
Like the proposed EU legislation, China also requires ongoing monitoring. China’s interim Keep humans in the loop of AI development; regulatory measures call for it. That means
measure states, “Before using generative AI products to provide services to the public, establishing controls for humans to view, explore and calibrate AI system behavior. Where
a security assessment must be submitted to the state cyberspace and information possible, these systems should be able to spell out why a particular result was achieved —
department [i.e., the Cyberspace Administration of China].” Then organizations must register sometimes called explainable AI — so users understand AI decisions. Legal leaders should
the algorithm on the official government website.2 mandate that third-party solution procurement and internal AI initiative development include
the use of human-in-the-loop tactics to provide explainability and decrease risk.
The emerging practice of an algorithmic impact assessment (AIA) can document decision
making, demonstrate due diligence, and reduce present and future regulatory risk and AI-mature organizations are much more likely to involve legal teams in the AI development
other liability. Creating an AIA must be a cross-functional endeavor. Besides legal, GC process, specifically in coming up with ideas for AI use cases.17
should involve information security, data management, data science, privacy, compliance
The GC could also establish a digital ethics advisory board of legal, operations, IT,
and the relevant business units to get a fuller picture of risk. Since legal leaders typically don’t
marketing and outside experts to help project teams manage ethical issues.18 The White
own the business process they recommend controls for, consulting the relevant business units
House blueprint notes that independent ethics committees can both review initiatives
is vital. The White House blueprint calls for organizations to make such assessments “public
in advance and monitor them to check whether “any use of sensitive data” infringes on
whenever possible.”10
consumer rights.10
Canada’s existing Directive on Automated Decision Making, which requires Canadian
government organizations to conduct AIAs, includes a tool that guides organizations
through the process — starting with an assessment of risk areas such as the reasons
for automation as well as the source and type of data used.14,15,16

Human oversight includes the board of directors. With 89% of boards Guard Against Data Privacy Risks
considering digital innovation an implicit part of their growth strategy,19
decisions to roll out LLMs within the organization should start with a well-
Instill “appropriate safeguards for the fundamental rights and freedoms
informed discussion with directors. Legal leaders should make sure board
members maintain a current understanding of the rapidly evolving LLM
of natural persons.”6,20,21
landscape. For a discussion guide, see Figure 3. EU’s proposed AI law
»Figure 3. Discussion Guide on the Opportunities and Risks of LLMs The advent of ChatGPT has also led national regulators within the EU to act. In May 2023,
France’s data protection authority, the Commission Nationale de l’Informatique et des
Libertés (CNIL), announced it will create a new department focusing on AI.22 The CNIL
Opportunities for Using LLMs
wants AI development to respect personal data rights such as individuals’ rights of access,
• Are we targeting the right LLM business opportunities? rectification and opposition.
• How do we identify appropriate LLM use cases for our business while Legal and compliance leaders should manage privacy risk by applying privacy-by-design
balancing risk with reward? principles to AI initiatives. For example, require privacy impact assessments early in the
• What are our competitors doing? project or assign privacy team members at the start to assess relevant risks. Better still,
work with business partners such as the project management office to build privacy impact
assessment requirements directly into project life cycles.
Risk Mitigation Plan
With public versions of LLM tools, organizations should alert the workforce that any
• Are there any risks we should not accept regarding LLMs?
information they enter may become part of the training dataset. That means sensitive or
• Are we taking an appropriate amount of risk? proprietary information used in prompts could find its way into responses for users outside
• Do we need any additional controls to mitigate risk? the business. Leaders must establish guidelines, inform staff of the risks involved and
provide direction on how to safely deploy such tools.
• How much oversight of LLMs do directors want?
Source: Gartner

Privacy liaisons, who act as a link between the privacy team and a business unit, With organizations under pressure to use AI innovatively, the GC and other legal executives
can help highlight the importance of privacy impact assessments and the risks to the must watch the regulatory landscape closely. Doing this can help establish appropriate
workforce. One tactic is to use liaisons to detect noncompliance or overly risky uses guardrails to manage potential risks — and make sure their companies don’t fall behind
of AI as a first line of defense. Recruit them from parts of the business with suspected competitors in the race to harness new technology.
exposure to AI risk, and provide extra training and guidance to help them identify and
assess potential problems.
Be Aware of These Differences

Despite the commonalities, jurisdictions have their own unique AI philosophies and
requirements. These two stand out:
• The U.K.’s guidance notes that “rushed attempts to regulate AI too early would risk stifling
innovation.” The document adds: “We recognize the need to build a stronger evidence
base before making decisions on statutory interventions.”11
• China’s interim measure states: “Content generated through the use of generative
AI shall reflect the Socialist Core Values, and may not contain: subversion of state
power; overturning of the socialist system; incitement of separatism; harm to national
unity; propagation of terrorism or extremism; propagation of ethnic hatred or ethnic
discrimination; violent, obscene, or sexual information; false information; as well as
content that may upset economic order or social order.”2

1
Gartner CIO and Technology Executive Agenda Surveys, 2019 through 2023. 12
IBM Trust Center, IBM. 19
2023 Gartner Board of Directors Survey on Business Strategy
Question: What are your enterprise’s plans in terms of the following digital 13
United States: Things We Learned at the 2023 IAPP Global Privacy Summit, in an Uncertain World. This survey was conducted to understand the
technologies and trends? In 2023, 32% of respondents said they plan to Mondaq. The EU’s AI Act defines a high-risk AI system as one that poses new approaches adopted by nonexecutive boards of directors (BoDs) to
deploy AI, up from 14% in 2019. n = 2,882 (2019), 1,063 (2020), 1,825 (2021), “significant risks to the health and safety or fundamental rights of persons.” drive growth in a rapidly changing business environment. The survey also
2,363 (2022), 2,186 (2023). This could include AI systems found in medical devices and autonomous sought to understand the BoDs’ focus on investments in digital acceleration;
2
Interim Measures for the Management of Generative Artificial Intelligence vehicles. sustainability; and diversity, equity and inclusion.
Services, Cyberspace Administration of China. Direct quotations came from 14
Algorithmic Impact Assessment Tool, Government of Canada. The survey was conducted online from June through July 2022 among 281
an earlier draft for comment translated into English by the Stanford Cyber respondents from North America, Latin America, Europe and Asia/Pacific.
15
Bill C-27: Canada’s First Artificial Intelligence Legislation Has Arrived, Norton
Policy Center of Stanford University and remain in the new interim measure. Respondents came from all industries, except governments, nonprofits,
Rose Fulbright.
3
Regulating AI in EU: Three Things That You Need to Know, and Three Reasons charities and NGOs, and from organizations with $50 million or more in
16
Directive on Automated Decision-Making, Government of Canada. annual revenue. Respondents were required to be a board director or a
Why You Have to Know Them!, Dentons. 17
2022 Gartner AI Use Case ROI Survey. This survey sought to understand member of a corporate BoD. If respondents served on multiple boards, they
4
Advancing Cooperative AI Governance at the 2023 G7 Summit, CSIS. where organizations have been most successful in deploying AI use cases and answered for the largest company, defined by its annual revenue, for which
5
EU Closes In on AI Act With Last-Minute ChatGPT-Related Adjustments, figure out the most efficient indicators that they have established to measure they are a board member.
Computerworld. those successes. The research was conducted online from 31 October Disclaimer: The results of this survey do not represent global findings or
6
Proposal for a Regulation of the European Parliament and of the Council through 19 December 2022 among 622 respondents from organizations in the market as a whole, but reflect the sentiments of the respondents and
Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence the U.S. (n = 304), France (n = 113), the U.K. (n = 106) and Germany (n = 99). companies surveyed.
Act), EUR-Lex. Quotas were established for company sizes and for industries to ensure a 20
EU Lawmakers Back Transparency and Safety Rules for Generative AI,
7
The Artificial Intelligence and Data Act (AIDA), Government of Canada. good representation across the sample. Organizations were required to have TechCrunch.
8
Bill C-27, Parliament of Canada. This includes the Consumer Privacy Protection developed AI to participate. Respondents were required to be in a manager 21
Data Protection, European Data Protection Supervisor. Privacy is considered
Act, the Personal Information and Data Protection Tribunal Act and the Artificial role or above and have a high level of involvement with the measuring stage
a fundamental right in the EU.
Intelligence and Data Act. and at least one stage of the life cycle from ideating to testing AI use cases.
Disclaimer: The results of this survey do not represent global findings or
22
Artificial Intelligence: The Action Plan of the CNIL, CNIL.
9
China Releases Rules for Generative AI Like ChatGPT After Alibaba,
the market as a whole, but reflect the sentiments of the respondents and
Baidu Launch Services, CNBC.
companies surveyed.
10
Blueprint for an AI Bill of Rights, The White House. 18
Holding AI Accountable: Three Considerations for the Tech and the Minds
11
A Pro-Innovation Approach to AI Regulation, gov.uk. Behind It, Fast Company.

AI Systems ‘Need Photo courtesy of the office
of MEP Dragos Tudorache
a Moral Compass’
Dragos Tudorache co-leads the European Parliament’s
work on the AI Act. A lawyer who began his career as a
judge in his native Romania, Tudorache was elected to
the European Parliament in 2019. Along with artificial
intelligence and new technologies, his legislative work
A Q&A With Dragos Tudorache, Co-Architect of the EU’s AI Act includes security and defense, transatlantic issues and
by Laura Cohn internal affairs. Prior to becoming an MEP, he headed
legal departments at the Organization for Security and
Co-operation in Europe and the United Nations Mission
in Kosovo.
Table of Contents
Q&A
In a video interview from Brussels, When it comes to AI regulation, a critical tenet across That’s how it was trained. That’s how it was instructed.
jurisdictions globally is transparency. In your view, That’s how it was tweaked to work. And then I can see
Dragos Tudorache discussed what
what does good transparency look like? What do whether there is anything wrong in the way it was done
business should home in on to prepare companies need to reveal? or not.
for AI regulation, lessons from the GDPR,
How your algorithm works, how you trained it, how you What transparency obligations do you see for
and his thoughts about warnings from developed it, how you instructed it to function, and to producers of large language model applications?
tech leaders that AI could lead to human actually reach its decisions or recommendations, or
extinction. This interview has been edited These systems need a moral compass. And that’s exactly
whatever it is that it generates as content. Is it something
what we’re saying with this text to developers of large
for length and clarity. that induces risk?
language models. They are great and fascinating, they
What do organizations need to do How can organizations explain their algorithms are beautiful products, no doubt about it — but they
now to prepare for the EU’s AI Act? — and where should they do that? need a moral compass. They need some rules and you
have a responsibility to proactively, in the design and
Well, if you are a company that uses AI in industrial The regulation will provide templates for how you can development, introduce safeguards that the content that
processes, in optimizing flows in a factory, in things comply with the technical data that you have to provide. is going to come out of your machine is not harmful and
that have nothing to do with humans, let’s say, then you There will be what is called an EU-wide AI registry, which is against the law. That’s one obligation and the second has
shouldn’t even bother asking your lawyers to read the text. going to be a public database where high-risk applications to do with copyrighted material.
of AI will have to publish all of this data that they have to
If you are developing AI in areas that have to do with us provide as part of their transparency obligations. What’s the obligation for companies there?
humans — whether it’s recruitment, whether it’s medical,
whether it’s AI that is optimizing decision making — then So if I want to understand how the algorithm works, and Large language models are also absorbing a lot of
it is likely that your AI is influencing environments that we the kind of data sets used, and what kind of instructions copyrighted material, whether it’s scientific articles,
believe need to be protected, and therefore you will need it received, then I open that, and look inside and say, ‘Aha.’ songs by ABBA or works of art.
to ask your lawyers to start looking at this text.

Q&A
Developers have an obligation to be transparent about the For the EU’s AI Act, how important are That’s something that’s one step too far, according to how
copyright material that they use. So that if I’m Drake and the biometric verification safeguards? we understand privacy here in Europe.
the Weeknd, I need to know that ChatGPT actually learned
This has been, and remains, the most ideological point in In terms of future enforcement of the AI Act, have
my songs. So in case it produces a song that very much
the text. Many misunderstand what we’re doing by saying you learned any lessons from the EU’s General Data
resembles mine, then I can go and knock on the door and
that we’re going to inhibit biometric technology. No, we’re Protection Regulation (GDPR)?
say, ‘Fella, you need to give me some dollars for that.’1
not. Biometric technology will remain. The one thing that
cannot be done — and that has less to do with companies We’re certainly trying to learn some lessons. First of all,
Let’s talk about human oversight. Do companies need to
developing it, but more to do with law enforcement in terms of avoiding the silo effect of different national
assign an AI point person, team or committee to review
agencies in Europe — is, you cannot put an algorithm to regulators without much coordination, we are introducing
any outputs from AI, for example?
run live, 24/7, in public spaces. We don’t want this a stronger mechanism of ensuring coherence in the
We have deliberately tried not to be overly prescriptive. in Europe. implementation of the law across the EU. That’s number
This is in order to leave room for companies to achieve one. Number two, we put some teeth in this law.
compliance in the best way they see fit. Why is this so critical?
Regulators will have the possibility to investigate
But can you give business leaders some examples of We don’t want to give a quasi-invitation to law algorithms, to request information, if it’s not clear from
how companies can provide effective human oversight? enforcement to run these systems all the time on the the data that companies have provided as part of the
grounds that maybe a criminal or missing child might pass transparency exercise. So regulators can go in, knock
I can’t tell companies exactly what to do. But you have to by on the street. on the door, and investigate the algorithm further. And if
have a human in the loop. Which means what? That I can’t they find infringements, basically the options can go from
let an algorithm be the sole deciding factor without having, How do you prevent abuse? And also, how do you protect
shutting down the AI system, to withdrawing it from the
at some point, a human who takes a look. So I need to privacy? Because basically what it means is that I will
market, to requesting changes, to very significant fines.
have someone, a person, who can have that final human know that I will be walking in the street and there’s always
touch to validate its content, its product. someone who will biometrically identify me.

Q&A
You were in Washington recently, talking to U.S. officials. What was your take on the message from OpenAI’s So for me, their plea confirms that we have done the right
What’s the outlook for international cooperation on Sam Altman and other tech leaders that AI could lead thing in writing up some rules.
AI regulation? to human extinction, and that mitigating the risks
“should be a global priority?”2
Unlike the GDPR, where major jurisdictions were not 1
An A.I. Hit of Fake ‘Drake’ and ‘The Weeknd’ Rattles the Music World,
ready to accept the same level of protection as we were It’s interesting that someone like Sam Altman tells you that nytimes.com
introducing in Europe, my feeling from talking to literally there is a risk of extinction. Compared to any other piece 2
A.I. Poses ‘Risk of Extinction,’ Industry Leaders Warn, nytimes.co
almost every democratic jurisdiction out there is that with of digital legislation that we have worked on until now,
AI, there is already a much better starting point. There is this time around I’ve heard businesses big and small say,
quite a lot of convergence in terms of understanding the ‘Listen, we think it’s time to have some rules in place.’ It is
challenges. good that they understand the responsibility they have.
Also, as tech leaders, it is good that they engage on this
Will all of them adopt an act like we do? Most likely not, and put pressure on lawmakers to start making decisions.
but my narrative has always been that this matters less.
As long as we agree on the big principles and on the
big political objectives and what we want to achieve, They know better than anyone else what’s behind
then you can accept diversity in terms of the type of this technology. If they sound alarm bells, that
legislation in place. means that our instincts in putting forward rules
were the right instincts.

Find out how generative AI will
reshape your marketing function
One thing remains constant about generative AI tools, is that we all have questions
about them. For marketing leaders, the questions include how it will change the
function, what it means for how their organization operates, and what impact it will
have on their interactions with customers.
But how do marketing leaders determine what opportunities to pursue?
Join this complimentary marketing webinar, as we explore what changes generative

AI could create for your organization, your enterprise and your industry.
Find out how generative Explore the impact of Determine the next
AI tools could reshape generative AI. steps for CMOs.
marketing.
Watch now to find out what CMOs can and should do to take advantage of generative AI.
Generative AI Will Affect
Information Security (and CISOs)
by Jeremy D’Hoinne, Avivah Litan and Peter Firstbrook
The cyber security impacts of a flood of experiments with ChatGPT and

large language models (LLMs) will require chief information security officers
(CISOs) to move quickly — adapting to rapid change while helping business
units safely build and consume the new technology.
That means preparing in four areas:
Generative cybersecurity AI. Safe consumption of Development of new The evolving threat
Exploit opportunities to generative AI. ChatGPT was generative AI applications landscape. Malicious actors
better manage risks, optimize the first example; embedded for the enterprise. These will refine their tactics and
resources, defend against generative AI assistants in techniques have an leverage generative AI where
emerging attack methods or existing applications will be expanded attack surface it is a good fit.
reduce costs. the next. These applications and require adjustments
all have unique security to application security
requirements that are not practices.
fulfilled by legacy controls.

Generative Start with AI assistants for developing Recommendations:
secure applications and security
cybersecurity AI • Prepare and train security and risk management teams to deal with direct
operations chatbots. The former can be (privacy, IP, AI application security) and indirect impacts (from other teams)
included in code generation tools such of generative AI uses in the enterprise.
as GitHub Copilot, Tabnine and Amazon
CodeWhisperer. The latter have the • Run experiments with new features from existing security providers,
potential to improve the average security starting with targeted, narrow use cases in operations and application security.
analyst’s productivity and to lower the bar • Establish or extend existing metrics to benchmark generative cybersecurity AI
for skill requirements. against other techniques, and to measure expected productivity improvements.
Examples and first demos of generative • Determine your corporate position on providing feedback to the applications
cybersecurity AI products and features and improve their efficacy in the long run.
appeal to many security professionals. • Identify changes in data processing and dependencies in the supply chain for
However, our conversations with CISOs security tools and require your security providers to be transparent about data usage.
have highlighted questions and challenges
• Adapt documentation and traceability processes to augment internal knowledge
related to cost and quality. One example
and avoid feeding the tools with only your insights.
of a concern: building new workflows
and dependencies relying on immature • Monitor the release of fine-tuned or specialized models that align with the relevant
technology that would require more work security use case or for more advanced teams.
in a few years. • Remember that using generative AI base models “as is” might not be adequate
for advanced security use cases.

Safe consumption Consumption of generative AI applications Recommendations:
such as LLMs through business
of generative AI • Define a governance entity and workflow. Generative AI applications might require
experiments and unmanaged, ad hoc a specific approval workflow and continual usage monitoring where possible.
employee adoption creates new attack
surfaces and risks related to individual • Monitor and block. When ChatGPT became public, many organizations blocked access
privacy, sensitive data and organizational to OpenAI domains or applied some data leakage prevention. Now, they are shifting
intellectual property (IP). CISOs should to a mix of “disclaimer pages,” which include links to user policy and relevant contacts,
work with IT, legal and compliance and and more stringent blocking measures.
directly affected business lines to minimize • Quickly communicate a short acceptable use policy. Often, a one- or two-page policy
unsanctioned uses of generative AI. document can be used to share the internal contact for approval, highlight the risks
of the applications, forbid the use of client data, and request documentation and
traceability of generative AI outputs.
• Investigate prompt engineering and API integrations. Intercepting inputs and outputs
with a custom-made prompt might improve results but also enable more security
controls. Larger organizations might investigate prompt augmentation techniques,
such as retrieval augmented generation (RAG).
• Favor private hosting options when available, which affords extra security
and privacy controls.

Development of In addition to using third-party generative AI applications and services, Recommendations:
many organizations will build their own versions that will pose new risks,
new generative • Adapt your control strategy to hybrid development
such as attacks on ML models and data poisoning. That in turn creates models, such as front-end wrapping (e.g., prompt
AI applications new requirements for AI application security. CISOs should adjust security engineering), private hosting of third-party models and
for the enterprise controls and implement our AI trust, risk and security management custom GenAI applications with in-house model design
(AI TRiSM) framework — governance practices establishing security and fine-tuning.
policies that are specific to AI.
• Team up with the data science team to evaluate the data
Implementing controls for GenAI applications will depend heavily on what security options when training and fine-tuning models,
kind of AI architecture an organization is using. If an organization builds notably the potential impact on accuracy or additional
and trains an in-house model, security and development teams share costs. Weigh this consideration against requirements
responsibility for protecting against the entire AI attack surface. Using in modern privacy laws, which often allow individuals
synthetic data might improve security but also reduce the accuracy of the to request that organizations delete their data.
applications. These techniques are not available when hosting a pretrained
• Train your security champions as soon as training
out-of-the-box model.
on secure GenAI coding is available.
With third-party models, companies must still address adversarial direct • Apply AI TRiSM principles and update security best
and indirect “prompt injections,” or the insertion of hidden instructions practices to include AI applications requirements
or context in the application’s conversational or system prompt
• Add requirements for testing against adversarial prompts
interfaces, or in the generated outputs. There are also digital supply chain
and prompt injections.
management challenges related to the model itself. Then, even when
hosting an out-of-the-box third-party application or model, CISOs and their • Monitor model operation tool improvements.
teams will be responsible for infrastructure and data security, but will also
have to be involved in managing third-party risks and vulnerabilities.

As with any innovation (consider the rise of social media Recommendations:
and cryptocurrency as two recent examples), malicious • Remember that the right order for any security investment is people,
actors will seek creative ways to exploit immature security process and, only then, technology. Address threats that are tied closely
practices and limited awareness. Perhaps the biggest risk to human interpretation of generated content for which there are no
The evolving will be generative AI’s potential to rapidly create believable existing technical controls.
fake content.
threat landscape • Monitor industry statistics to measure the impact of generative AI on the
CISOs and their teams should anticipate the following: attacker landscape.
• Generative AI as a lure: Expect counterfeit GenAI apps, • Make sure you can measure drift in detection rate from existing controls.
browser plug-ins, applications and websites promising
• Elevate requirements for more adaptive behavioral and ML defenses in
“the best Gen AI” but delivering malware-infected
your existing security controls.
applications.
• Challenge all existing and prospective security infrastructure vendors to
• Digital supply chain tampering: Embedded third-party
outline how their product and research will evolve to address threats and
GenAI components (libraries, models) could be affected
tactics now possible due to generative AI. Beware of overstated claims.
by adversarial tactics, such as training data manipulation.
• Evaluate the business dependency of key digital supply chain software
• Adversarial prompting: This threat includes prompt
and develop playbooks for “zero-day vulnerability” attacks, where
injection.
malicious actors discover and exploit weaknesses before they are
Generative AI tools will enable cyberattackers to publicly known.
increase the quantity and quality of attacks at low cost. • Reduce the number of “blind spots” — assets, transactions and business
Attackers have already started to exploit the technology’s processes that you cannot monitor for anomalies.
capabilities to craft better malicious and fraudulent
• Address the potential increased risk of fraudulent content and influence
content at scale.
operations on corporate brands.
Uncertainty about how successful they will be will require • Add generative AI content to security awareness training and phishing
more flexible cybersecurity roadmaps. Security leaders simulations. Make business workflows more resistant to realistic phishing
need to reinforce their investments in resilience and in campaigns and voice and video impersonations.
reducing exposure to categories of threats, rather than
individual known attacks.

How Generative AI Can Help Meet
Customer Experience Expectations
by Gene Alvarez, Uma Challa and Leah Leachman
Customer retention and growth feature heavily in the plans »Figure 1. Primary Focus of Generative AI Initiatives
of executives interested in generative AI. Thirty-eight percent Customer Experience and Retention 38%
of leaders see improving customer experience and retention as
Revenue Growth 26%
the primary purpose of initiatives to deploy applications trained
on large language models, while 26% highlight revenue growth. Cost Optimization 17%
Only 17% cite cost optimization (see Figure 1).1 Business Continuity 7%
Generative AI could help enterprises achieve all three goals by addressing rapidly changing None of the Above or Not Applicable 12%
consumer expectations. CEOs report price sensitivity as the top shift in customer behavior (e.g., Vendor or Investor)
as inflation begins to bite, and 21% of them regard AI as the leading disruptive technology.2 n = 2,544
Source: 2023 Gartner Beyond the Hype: Enterprise Impact of ChatGPT and Generative AI Webinar Polls
Executive leaders should therefore pilot promising customer-oriented use cases of tools
Note: Results of these polls should not be taken to represent all executives as the survey responses come
such as ChatGPT and other AI techniques while guarding against the biggest risks. from a population that had expressed interest in AI by attending a Gartner webinar on the subject.

For example, snack company Mondelēz Smoother encounters with your integrated into self-service channels enterprise repositories and search
International created an AI avatar organization are the prize to pursue such as virtual assistants can interact engines for continuous improvement.
campaign to coincide with the Diwali — not just cost cutting. Replacing service in human-like fashion to provide detailed The technology can therefore increase
festival in India.3 Wendy’s is using Google people with bots that annoy or frustrate information on products and how to accuracy and significantly reduce the
Cloud’s generative AI technology to those seeking help will not boost retention. use them. time service agents spend compiling call
enhance its drive-thru ordering service.4 • Identify value enhancement notes, thus improving their experience
And U.K.-based Octopus Energy says AI Generative AI, however, has great potential and productivity.
opportunities for customers in real time
is doing the work of 250 people at the to improve customer experience — and
and automate or assist agents to give • Close the loop on customer issues.
company by replying to customer emails.5 customer relationships — even though
contextualized information and advice. Service organizations can generate
some of the use cases are complex and emails and texts to customers
require significant investments. Second, generative AI can reduce
Make Customer Experience instantly and efficiently, and enterprise
customer effort. Companies could
and Retention Your Core Aim First, the technology can enhance value integrate tools such as ChatGPT or other
communication platforms can
personalize follow-up messages.
for customers by: AI techniques to:
Customers expect effortless access to
• Assist customer support agents by
content, quick resolution of issues and • Educating them on how to use products • Better understand customer intent and
monitoring live interactions to suggest
seamless product experiences. Churn and services better sentiment (through virtual assistants)
the next best sentence or action. A
rates can be high if: • Advising them on new uses and help resolve issues efficiently — or
study by Stanford University and MIT
route the customer to the best channel
• Information provided by the company • Validating their purchase decisions found that agents who used AI tools that
to solve them.
is inconsistent or even contradictory. created conversational scripts increased
• Anticipating their needs • Summarize customer interactions.
• Resolving problems takes a lot of their productivity, measured as issues
• Helping them achieve a goal Generative AI models can do after-call resolved per hour, by 14% on average.6
customer time.
work such as producing call notes,
• The value of the product is not intuitive For example, consider using generative AI to: providing updated summaries during
or doesn’t meet expectations. • Onboard new customers or educate the interaction on both voice and text
existing ones. Generative AI solutions channels, and placing information into

Customer-Centric Content Executive leaders should look for points »Figure 2. CEO’s Top 10 Strategic Business Priorities for 2023 and 2024
of overlap where generative AI can
Is Vital for Revenue Growth Change From 2022
enhance both content production and
Growth is CEOs’ No. 1 business priority personalization. Don’t leave this mission
Growth 49% q 5%
(see Figure 2).2 Economic uncertainty on autopilot. Guide it with clear use cases,
means efficiency is imperative, and a renewed emphasis on customer research 34% q 2%
Technology-Related
generative AI’s ability to automate the and human oversight to avoid biases
rapid production of more personalized and inappropriate uses. Get this step Workforce 32% p 2%
creative content makes it a “must have” right, and:
for marketing, sales and digital commerce • Marketing will feel empowered to shift Corporate Action 31% p 6%
initiatives. More than half of marketing focus from more tactical tasks, such
teams (64%) are already leveraging AI Financial 19% q 6%
as design production, to customer
heavily to create, execute and optimize research, strategy and content
campaigns.7 Cost Management 14% p 69%
curation. More resonant and engaging
Although generative AI promises increased customer experiences will result. Regard Products and Services 13% q 18%
productivity, simply doing more does generative AI as a complement to human
expertise, not a substitute. Environmental
not guarantee growth. While the volume 12% p 25%
Sustainability
of content, ads and notifications that • The speed and range of content
Customer 8% q 45%
customers receive daily is proliferating, produced will enable your organization
the value they attach to the content is to reach customers worldwide more Efficiency and
7% p 46%
strikingly low. Only 5% of consumers recall efficiently and allow your brands to Productivity
experiencing valuable digital interactions manage complex customer journeys
with a brand when considering a more effectively. n = 422, all respondents
purchase.8 Q. Tell us about your organization’s top five strategic business priorities for the next two years (2023 and 2024).
Source: 2023 Gartner CEO and Senior Business Executive Survey

• Optimizing product choices and • Summarization. ChatGPT and other • Better webpages and merchandise Remember this emerging technology is not
potentially influencing customers’ generative AI techniques can give offerings, boosting sales volume. risk-free. Consider the following factors to
decisions — by nudging them toward customers concise instructions to solve • Virtual influencers and brand avatars to plan and implement mitigating steps:
different options — can accelerate a problem — freeing time for customer- mitigate friction in the customer journey, • Nonsense answers (which nonetheless
adoption of new products or services. facing employees to handle more alleviating some of the burden on service sound authoritative) are a pervasive
requests. teams. problem.
But be cautious with recent advances
in generative AI. These tools create • Content creation. Generative AI can • In a publicly shared model, data privacy
derivatives of what already exists, which help salespeople create personalized Experiment With Pilots is an issue.
RFP responses with industry terms and
include hidden biases and the possible and Mitigate Risks • As with every machine learning
violation of ethical, copyright and processes aligned to the products or
services being sold. The technology technique, models may provide biased
regulatory standards. Executive leaders should:
will base these drafts on all the RFPs responses, depending on the training
• Try small generative AI pilots aimed data.
the organization has previously replied
Explore How Generative specifically at improving customer
to and will use best practices. This tactic • Copyright violations are a further risk.
AI Can Boost Operational retention, revenue growth or operational
can accelerate responses to customer • Governments may impose or interpret
efficiency.
Efficiency requests and reduce overall deal regulations that could add risk to current
cycle times. • Locate and carefully curate data sources
In 2023, CEOs rank talent shortages implementations in the future.
that will train the models for these
and inflation as higher concerns than Marketing can use generative AI’s initiatives.
a recession or rising interest rates.2 capabilities to efficiently create:
• Question whether generative AI would
Organizations are therefore seeking
• More personalized and in-context best serve your use case or whether
productivity gains to offset labor
customer messaging, increasing the other more traditional techniques would
constraints and increasing cost pressures.
number of customers that purchase better address both customer and
Generative AI innovations can help raise
the product or service. commercial objectives.
efficiency, including through:

1
2023 Gartner Beyond the Hype: Enterprise Impact of ChatGPT 3
2022 One Show - Creative Use of Data, Shah Rukh Khan-My-Ad, their interactions with brands. The research was conducted online from
and Generative AI Webinar Polls. The One Club for Creativity. November through December 2022 among 1,076 consumers from North
This webinar was held on 30 March and 21 April 2023 with 1,465 and 1,079 4
Wendy’s Is Working With Google to Test an AI-powered Drive-thru, America (n = 717), Western Europe (n = 136) and Asia (n = 223) and among
respondents to the polling, respectively, for a total of 2,544 responses. Results Restaurant Business. 509 B2B buyers from North America (n = 327), Western Europe (n = 78) and
of these polls should not be taken to represent all executives, as the survey 5
AI Is Doing the Work of 250 People at an Energy Company and Satisfying Asia (n = 104). Consumer respondents were required to have considered a
responses come from a population that had expressed interest in AI by Customers Better Than Trained Workers, CEO Says, Business Insider India.
attending a Gartner webinar on the subject. purchase within the last month, used a brand’s digital information sources
6
Stanford and MIT Study: AI Boosted Worker Productivity by 14% and recalled a specific digital interaction with a focal brand. B2B buyer
2
2023 Gartner CEO and Senior Business Executive Survey. — Those Who Use It ‘Will Replace Those Who Don’t’, CNBC.
This survey was conducted to examine CEO and senior business executive respondents were required to be involved in a significant purchase on behalf
7
2023 Gartner Multichannel Marketing Survey. This survey was conducted
views on current business issues, as well as some areas of technology agenda of their organization in the last six months, used a supplier’s digital information
to determine best practices for maximizing multichannel marketing
impact. The survey was fielded from July 2022 through December 2022, with sources and recalled a specific digital interaction with a focal supplier or
investments in response to evolving customer journeys in a fluid marketing
questions about the period from 2022 through 2024. One-quarter of the department (n = 205).
environment. It was conducted online from November through December
survey sample was collected in July and August 2022, and three-quarters
2022. In total, 397 respondents were surveyed in their native languages Forty-seven percent of B2B respondents came from organizations with $2
was collected from October through December 2022. In total, 422 actively
across North America (n = 201), Western Europe (n = 161) and the Nordics billion USD in annual revenue. B2B respondents came from a wide variety of
employed CEOs and other senior executive business leaders qualified and
(n = 35). Qualifying organizations reported enterprisewide annual revenue industries, including technology products (n = 78), financial services (n = 76),
participated. The research was collected via 382 online surveys and 40
for FY 2021 of at least $100 million, with 83% of the respondents coming manufacturing and natural resources (n = 71), retail (n = 69), IT and business
telephone interviews. The sample mix by role was CEOs (n = 277); CFOs
from organizations with $1 billion or more in annual revenue.
(n = 95); COOs or other C-level executives (n = 19); and chairs, presidents services (n = 41), construction (n = 30), healthcare (n = 24), wholesale (n =
or board directors (n = 31). The sample mix by location was North America The respondents came from a variety of industries: financial services (n = 61),
19), transportation (n = 17), consumer products (n = 16), real estate (n = 11),
(n = 169), Europe (n = 105), Asia/Pacific (n = 102), Latin America (n = 29), the manufacturing (n = 58), consumer products (n = 46), retail (n = 45), travel
and hospitality (n = 45), healthcare (n = 44), pharmaceuticals (n = 35), media government (n = 10), media (n = 10), pharmaceuticals (n = 10), travel and
Middle East (n = 11) and South Africa (n = 6). The sample mix by size was $10
(n = 33), and IT and business services (n = 30). All respondents were senior hospitality (n = 9), education provider (n = 7), agriculture (n = 3) and other (n = 8).
million in revenue to less than $50 million (n = 3), $50 million to less than $250
million (n = 51), $250 million to less than $1 billion (n = 102), $1 billion to less leaders who manage multiple marketing channels or were responsible for the Disclaimer: The results of this survey do not represent global findings or
than $10 billion (n = 190) and $10 billion or more (n = 76). execution of their organizations’ multichannel marketing strategies. Eighty-five the market as a whole, but reflect the sentiments of the respondents and
percent of respondents were aligned to the marketing function, 9% to brand companies surveyed.
Disclaimer: Results of this survey do not represent global findings or the
management and 7% to sales or other business units.
market as a whole, but reflect the sentiments of the respondents and
companies surveyed. Disclaimer: The results of this survey do not represent global findings or
the market as a whole, but reflect the sentiments of the respondents and
companies surveyed.
8
2022 Gartner Digital Customer Engagement Survey. This survey was
conducted to explore consumers’ and B2B buyers’ experiences in interacting
with brands using digital channels and their perceptions of the value of

Learn to Serve Your AI-Powered
Customers Before They ‘Walk Away’
by Mark Raskino and Don Scheibenreif
AI will accelerate the replacement of human customers by machine customers,

which will act very differently. By adopting smart tactics now, executive leaders
can make sure their organization benefits from this emerging megatrend.

The popular obsession with AI’s effect Let Go of the Idea That All • DoNotPay’s ChatGPT bot renegotiated a customer’s
broadband provider bill with Comcast.7
on employment is understandable, Customers Are Human
but it deflects attention from a greater • Greg Brockman, co-founder of OpenAI, demonstrated
Machines are increasingly becoming customers in a recent TED Talk how ChatGPT could help him plan
business challenge. AI will accelerate on our behalf, and AI is speeding up that trend. a dinner party menu, send the menu to DALL-E to create
the replacement of human customers by • Walmart’s AI-powered negotiation software with a an invitation, make a shopping list and send that list to
machine customers. This megatrend will text-based interface (i.e., a chatbot) has so far closed Instacart.8
disrupt markets, often by stealth. agreements with 68% of suppliers approached, with • Lazada, an e-commerce company in Singapore, created
each side gaining something it values.2 LazzieChat, a ChatGPT-powered bot that provides users
While AI bot replacements for employees might do as
• Ticketmaster admitted that a “staggering number with a more personalized shopping experience.9
you say, AI-enabled machine customers — nonhuman
of bot attacks” disrupted its site selling tickets for
economic actors that obtain goods and services in AI is accelerating the trend of machines becoming
a recent Taylor Swift concert.3
exchange for payment — will do as they see fit. If you serve your customers. That is not a contrived or exaggerated
them well, they can grow your addressable market. If you • LexCheck, a contract review platform, uses AI to perspective: Machines are gradually taking over all the
ignore their needs, you may not even see them walk away. help study legal, procurement and sales proposals.4 tasks of being a customer. They can determine needs,
• Nike adjusted its terms of sale to curb the effect of they can shop and compare, and they can negotiate,
The new generation of deep neural network AI systems will
“sneaker bots” that resellers sometimes use to automate agree and pay. The argument that the “real customer”
change the role of the customer — in many cases, for the
the buying of scarce inventory.5 is always the human or organization behind the bot
better. While some shopping for handbags, rare whiskeys
is irrelevant to your ability to sell. Once the machine
or a new car might count as a leisure activity, a great deal In 2023, large language model (LLM) generative AI has assumes the role of purchaser, it is the machine you
of customer work is a laborious chore. For example, few added an extra twist. must convince to buy.
buyers enjoy shopping for toilet paper, life insurance or a
• Expedia announced in April 2023 that it’s adding ChatGPT
large technology purchase for their organization. Likewise,
features to its mobile app to provide an interactive
a recent study of a four-day workweek experiment in the
experience for basic travel details. The AI service cannot
U.K. found employees devoted their additional time off
plan an entire trip yet, but that is in the works.6
largely to “life admin” such as chore shopping.1 AI can
and will help chip away at that work too.

The Automotive Industry Meanwhile, Elon Musk has repeatedly asserted that to reduce internal operating costs, in a prior technology
Tesla cars will earn money in the future by operating wave. Be ready for this monumental transition as
May Show the Road Ahead
as autonomous taxis.15 it unfolds throughout this decade. For example,
Cars will be one of the most important breakthrough organizations that produce physical products or software
These developments point to a world in which driverless
categories of machines that will become customers have some vital decisions to make (see Figure 1).
cars roam around earning money and spending it
for everything from tires and recharging, to passenger
directly from their own payment wallets. But machine The spread of machine customers might shrink some
snacks and cleaning services. The demise of autonomous
customers do not have to be physical things like cars or markets, but for many or most companies, it will be an
vehicles has been exaggerated: They are already
home appliances. Software agents also will be important opportunity for total market growth. That’s because
operating on the streets of San Francisco, Phoenix and
players, as embedded generative AI massively boosts the humans tend to be inefficient customers who frequently
Los Angeles,10 and in Chongqing, Wuhan and Beijing.11,12
abilities of intelligent assistants such as Amazon’s Alexa. dither, forget and fail to complete specific actions.
What’s at stake is so profound that big car companies Once we all have access to powerful shopping robots in Did you ever miss the moment to buy that event ticket,
are radically reorganizing themselves to accelerate our work and home lives, what won’t we be prepared to or open your toolbox and find you still had not got around
transformation and insure against disruption. In 2022, have them buy for us? to replacing that worn out drill bit?
Ford split itself into two divisions: Ford Blue to maintain
Little wonder, perhaps, that HP Inc. now has over
the heritage combustion engine business, and Ford See the Opportunity 11 million customers for its HP Instant Ink service,16 where
Model e to aggressively develop electric autonomous
for Market Growth people delegate to their printer the task of remembering
vehicles.13 Also that year, General Motors CEO Mary
to buy ink. That’s a lot of people who get to print pages
Barra declared that the company’s Cruise subsidiary In many B2C and B2B settings, human consumers and
when they otherwise might not have — expanding the
could generate $50 billion of annual revenue by 2030 business professionals will start delegating machines
total market. In 2022, HP Inc. extended that service into
from autonomous vehicle services and technology.14 to do their customer work for them. Often, that work
an adjacent revenue opportunity: Your printer can now
is the “self-service” effort that companies foisted on
order paper for you too.17
their customers in the name of convenience, but really

»Figure 1. Machine Customer Decision Path
Does your company Is that product connected to Are there services or consumables that Upgrade the product and test market it,
produce a physical the internet? a machine could buy from you? offering auto-order or autoreplenishment.
product or software? E.g., a connected toothbrush orders
replacement brush heads.
Could a future version be Test market a

connected to the internet? connected version.
Develop the product’s capability to buy

While your customer buys or uses what you sell, is any software (e.g., Are there services or consumables that a directly from suppliers.
GenAI chatbot) or digital device (e.g., IoT power tool) involved? machine could buy from others? E.g., manufacturer of a robot lawnmower
builds a platform for ordering fertilizer,
weedkiller and grass seed.
Wait one year. Wait one quarter. Is that kind of software or device starting to act as a customer for what you provide?
Reach out to the producer of that software or

Today’s generative AI can appear human-like. Are you Watch out for any suspiciously “inhuman” customer device to agree on a partnership strategy that
certain all your customers are humans? interactions. Track down and analyze that bot! will provide win-win growth benefit for both.
Source: Gartner

Here’s What You Should Do • Start hiring and training sales and service staff Would You Prefer a Four-Day Working Week? University of Cambridge.
1
2
How Walmart Automated Supplier Negotiations, Harvard Business Review.
to work with AI agents. They will need to understand 3
Taylor Swift | The Eras Tour Onsale Explained, Ticketmaster.
Executive leaders can position their organizations to and possibly crack the algorithms that drive a 4
LexCheck, LexCheck.
benefit from the AI-fueled machine customer trend by machine customer’s purchase behavior or after-sale 5
Nike Moves to Curb Sneaker-Buying Bots and Resale Market With
Penalties, CNBC.
taking the following steps: service demands. These employees will need a basic 6
Expedia App Integrates ChatGPT, Forbes.
• Make all your product and service information easily understanding of how the technology works; some 7
Watch DoNotPay’s AI Chatbot Renegotiate a Comcast Bill to Be $120
positions may require a data science background. Lower, PCMag.
accessible to machine customers. They may be 8
The Inside Story of ChatGPT’s Astonishing Potential, TED.
searching on 100 different variables, and you’ll need to • Alert and train customer-facing staff to spot machine 9
Lazada Unveils New eCommerce AI Chatbot LazzieChat,
customers. Machines (generative AI LLM-powered Marketing-Interactive.
provide data for all of those, depending on where they 10
Waymo Is Starting Driverless Taxi Tests in Los Angeles, Engadget.
are in the purchase process. Provide and encourage API bots) posing as humans may already be trying to 11
Baidu Launches China’s First Driverless Taxi Services in Chongqing
access, and make sure CAPTCHA and other bot-thwarting negotiate, book and buy from you through text-chat- and Wuhan in Landmark Moment for Autonomous Motoring,
South China Morning Post.
tools are not shutting out legitimate machine customer based customer service lines, and perhaps even via 12
Baidu, Pony.ai Win Permits to Offer Driverless Robotaxi Services in Beijing,
revenue. telephone calls using high-quality voice synthesis. Reuters.
13
Announcing Ford Blue™ and Ford Model e™, Ford.
• Add machine customers to your core digital commerce 14
GM’s Cruise Robotaxi Unit Drives Deeper Into the Red, Reuters.
15
Watch Elon Musk’s Full Interview With CNBC’s David Faber on Twitter,
and sales strategy. Strive to be exceptional at digital
Tesla and A.I. Advances, CNBC.
commerce. It will be the first place machine customers go 16
D. Scheibenreif and M. Raskino, “When Machines Become Customers,”
when they want to buy from you. Consider the impact of Gartner, 2023.
17
Never Run Out of Paper Again! HP Inc.
these buyers on the way you sell and provide information.
• Develop a strong commercial partnership between
sales, marketing, supply chain, IT and analytics. This
team should create one to three scenarios that explore
what happens when AI-enabled machine customers start
buying from you. The supply chain must be agile enough
to respond to unexpected demand patterns.

What Is
What Generative AI Means for Business
Generative AI?
Business leaders face 3 new sets of expectations —
What Is Generative AI?
and generative AI can help
AI techniques that learn from data
Generative AI
Investors will expect new sources of growth
1 about existing artifacts and use this
to generate new artifacts.
and better margins. Foundation

Large machine learning models that
are trained on a broad set of unlabeled
2
Models data and are fine-tuned to a wide range
of applications.
Customers will leverage GenAI in their daily
Large Language AI that is trained on vast amounts
lives — and expect you to do the same. 3
Models (LLM)
of text to interpret and generate
human-like textual output.
Employees will leave organizations 4 ChatGPT

An OpenAI service that incorporates
a conversational chatbot with LLM to
where humans are doing work that GenAI

create content.
could handle.
Explore Gartner resources to get

started and move forward with GenAI
Source: Gartner
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved. 2421958
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved. CM_GBS_2463403 in your enterprise.
What Generative AI Means
for Your Talent Strategy
by Helen Poitevin and Pieter den Hamer
The accelerated investment in generative AI has unsurprisingly AI applications have been affecting workers for years, but this time feels different.
During the late 2010s, few executive leaders would claim publicly that layoffs were in
led to concerns about how this technology will affect jobs,
any way due to investments in AI or automation. They would usually say they identified
including those once thought impervious to automation. Some impacted staff, retrained them and shifted them to other roles. Today, however, as
roles will cease to exist, while many others will change radically, use of generative AI tools such as ChatGPT spreads rapidly, leaders are more willing
encompassing new tasks and requiring new skills. Faced with to explicitly call out AI as one of the reasons positions will disappear — whether the
a potentially historic disruption, executive leaders should shape technology is truly the culprit or not.1,2
their talent strategies using a new framework for understanding

generative AI’s impact on the workforce.

Assess Impact Within Your Business Context • Technology drivers relate to how your organization implements and uses technology.
They affect how you organize work within and across teams.
Generative AI will affect each organization differently. Executive leaders must consider
– Talent impact: Organizations adopting generative AI will design roles and workflows
their enterprise’s unique situation as they anticipate the effects of this technology (and
differently depending on what they are using it for. Executive leaders must also
AI more broadly) on the workforce. Most available models for measuring automation risk
account for the combinations of skills available within the organization and the wider
from AI analyze only existing roles and estimate the potential for AI to replace current
labor market.
employees. While these models have merit, they omit two important factors in any
executive leader’s talent decisions: Generative AI will directly alter tasks such as content creation, question answering
• Demand drivers show whether your organization should scale up products and services, and discovery, translation, document summarization and software coding. But how
scale them down or hold them steady. it will transform individual jobs is more challenging to predict. For example, knowing
generative AI will disrupt copywriting and customer support doesn’t tell you what it
–
Talent impact: Future demand for specific capabilities within your organization,
means for the people providing those services in your organization. Your business
industry or broader market will determine where you invest in talent and
context will determine whether you use the technology to make these employees more
technology. Demand also affects your ability to meet your talent needs: Sought-
productive, phase out jobs that can be done by machines or reconfigure these roles
after workers are harder to attract, and low-demand jobs become less desirable
to provide new types of services.
and more challenging to fill.
The following steps will help executive leaders identify the impact of AI on their
organization’s workforce and adapt their business, talent and technology strategies
accordingly.

Step 1: Forecast Demand • Decreasing/Low — The service or product targets a You could:
niche market or is being phased out. Ongoing support • Stay within their boundaries — Many emerging use
First, examine how you expect demand for a given product is required, but there are no plans for investing further cases for generative AI are far from transformative. For
or service to change, based on forecast business conditions to try scaling up to a bigger offering. This includes example, a recruiter could use embedded generative AI
and company strategy. Apply this exercise at the enterprise, instances where a company continues to serve existing functionality in a talent sourcing tool to automatically
business unit, department or team level, and for both customers for a product or service but no longer offers apply search criteria based on natural language input,
internal- and external-facing services. This helps clarify it to new ones. without having to manually apply filters. This change
how much a given team will need to deliver in the future. – Examples: Legacy products no longer sold but with simply makes an existing work pattern easier and faster.
multidecade contracts that need to be honored; • Go beyond their boundaries — Generative AI, alongside
Demand for a product or service could be:
niche products such as parametric insurance other AI techniques, can change how people access
• Increasing/High — An offering’s success is driving a
products and services. For example, AI can help create
marked rise in demand, with high growth targets. This Step 2: Envision AI Technology webpages, videos, apps or other content quickly, without
demand may also be for a particular capability — such as
Applications requiring any technical knowledge. This capability allows
AI, software or other digital technologies — that needs to
for a fully personalized customer experience.
scale to an increasing number of domains. After forecasting demand, formulate your future vision
• Create new boundaries — Autonomous business,
– Examples: Medical diagnostics, software engineering, for how you will apply generative AI and AI more broadly.
in which AI manages or carries out the majority of
This helps identify the degree of impact on a given team
AI development, prompt engineers operations, has already started to emerge. Machine
or role. Will people in certain jobs be replaced? Will they
• Stable/Core — Volume and future evolution are customers, augmented managers and autonomous
need new skills? How critical will it be for a worker’s
predictable, possibly with some variability linked to operations are central features of these new business
success and competitive positioning that they use
seasonality or similar factors. There are no plans for models. We anticipate the labor ratio to substantially
generative AI effectively?
either high growth or phasing out the service. change, with enterprises needing fewer people to
Consider what your planned applications of AI mean generate the same amount of revenue.
– Examples: Payroll services, customer service for existing professions and industries.
supporting stable markets, internal IT services,
insurance claims processing

Consider a Matrix »Figure 1. Talent Impact Matrix for Generative AI and AI More Broadly
of AI Futures Low AI Impact High AI Impact
Combining these demand and technology How AI Is Applied in Industries and Professions
factors creates a range of situations Within the Boundaries: Pushing Boundaries: Breaking Boundaries:
(see Figure 1). Within each organization, Shifting roles through automation Newly configured roles Game-changing roles.
and even within each team, multiple and augmentation of existing work through transformation and Autonomous business with
cases are likely to apply. For each patterns and tasks. augmentation; new work significantly lower labor-to-
situation, specific investments in talent patterns. revenue ratio.
development and workforce planning will Increasing/High: Scale Up: New Impact Level: Symbiosis:
Business Context and Demand Volumes

help executives lead their teams through Scale up and grow Productivity goes up with AI and High performance expectations Machine economy with few
an AI-driven wave of transformation. to meet demand. humans accomplishing more work in many newly configured roles, employees compared to
with less effort. Still hiring to meet transformed through AI and impact, reach and revenue.
We expect generative AI to have targets. human partnership. Top design teams required.
two broad effects on the workforce, Stable/Core: Shift and Not Replace: Run Smarter: Synergy:
each with significant implications for Scale through efficiency Work shifts. Many are not replaced, Increased performance New pockets may emerge
executive leaders. and productivity gains. and most complex work remains. expectations per worker in terms for machine customers and
Roles must be reconfigured, of reach or impact. Only those autonomous operations.
often toward assisted multiskilled who can work effectively with AI Innovative redesigner teams
generalist roles. remain competitive. required.
Decreasing/Low: Last Ones Standing: Niche Innovations: Experiments:
Scale down or maintain Hard to attract talent and likely Hard to attract and find talent in Rare, but micro niche
low niche-style activity. losing talent. Automation is an hyperspecialized markets. May investments. Few employees.
opportunity to maintain service open up some new markets that
despite talent shortage. can now be served profitably.
Source: Gartner
a
Demand can be for internal or external services, at the organization, business unit or department level. It drives activity volumes and portfolio prioritization.

Generative AI Within Generative AI Beyond BT to Cut 55,000 Jobs With Up to a Fifth Replaced by AI, BBC News.
1
2
IBM CEO Among the First Major Executives to Say They’ll Replace Jobs
the Current Boundaries the Current Boundaries With AI, Axios.
Using generative AI in this way will result in the shifting Second, using generative AI to push the boundaries of
of roles over time. Fewer people will be needed to professions and industries, or even break them, will set
complete the same amount of work. Mass layoffs driven off a race for performance. Expectations will be higher in
solely by generative AI adoption are unlikely, especially newly configured roles, and organizations won’t be able
considering that labor markets remain historically to compete without using AI. The question will be not
tight. However, people who leave jobs affected by the which tasks go to AI and which to humans, but how people
technology are less likely to be replaced. Employees can use AI creatively to reach new heights. New, highly
hired into these roles will be expected to accomplish specialized jobs will emerge where generative AI and
higher work volumes more quickly. It will also become related technologies are used creatively and strategically
increasingly challenging to find talent willing to take on to transform what teams do and what their clients expect.
jobs that automation will likely displace. This requires a rich blend of business and technological
acumen that few possess.
Executive leaders implementing generative AI in this
context should anticipate headcount reductions over For example, organizations will need executive-level
time. They will need to redesign jobs displaced or business architects who wield both types of expertise in an
disrupted by AI into smaller numbers of multiskilled entrepreneurial way. When it’s too hard to find the right fit
generalist roles that encompass a wider range of for these roles, build on experience with cross-disciplinary
capabilities and offer a more compelling employee fusion teams that design and deliver digital products and
value proposition. services. Whether this specialized role is filled by a person
with this rare mix of talents or by a fusion team, it will be
critical for enterprises that make the leap to create new
boundaries in a machine economy.

Photo courtesy
Wikimedia Commons
Can AI Overcome
Daniel Kahneman is professor of psychology and public
affairs emeritus at the Princeton School of Public and
International Affairs, the Eugene Higgins professor of
Flaws in Human
psychology emeritus at Princeton University and a fellow
of the Center for Rationality at the Hebrew University in
Jerusalem. He is the author of Thinking, Fast and Slow.
Decision Making?
A Q&A With Daniel Kahneman and Olivier Sibony
by Steve Shapiro
Photo courtesy
Wikimedia Commons
Olivier Sibony is a professor, author and expert in strategic

thinking and decision processes. He is professor of strategy
at HEC Paris and an associate fellow of Saïd Business School
in Oxford University.
Table of Contents
Q&A
When presented with the same facts, What’s the biggest thing that people feel more natural, meaning more human to the human
get wrong about “noisy” decisions? user, it intentionally introduces an element of randomness
people will disagree and even draw
to the answers so that it will not feel like a machine.
different conclusions themselves from Kahneman: People are not aware of the amount of
But at a minimum, it is clearly noisy, so if you were hoping
one day to the next. noise. And they’re certainly not aware of the damage
that those models would be a substitute for human
that it does. One of the motivations for the book was
judgment, that’s going to be a problem.
This variability in decision making — what Daniel a conversation with somebody who runs a hedge fund and
Kahneman, Olivier Sibony and their co-author Cass he thought that errors cancel out. With that misperception What about algorithms in general? Can they solve
Sunstein call “noise” in their book Noise: A Flaw in Human you’re not going to take noise very seriously. When the problem of noise if they don’t act like humans?
Judgment — can create costly problems for organizations. they have disagreements, it appears to be a one off,
Disagreements don’t cancel out, but add up, they argue. but disagreement is actually the rule. People ignore Kahneman: Moving to algorithms will improve the quality
the problem. and the accuracy of judgments. Algorithms tend to beat
We should expect such inconsistency in humans — humans, or at least tie with them.
after all, we are not machines. But with machines now Has the emergence of new generative AI tools such
making decisions, can we overcome this flaw in human The main reason for human inferiority in that regard is
as ChatGPT changed how you think about noise?
psychology? And what does that mean for organizations? noise. Bias in human judgment tends to be drowned out
Sibony: One of the main reasons to use models and by the noise. When you have a system that is not noisy,
In separate conversations, Kahneman and Sibony told us algorithms instead of humans to make judgments is that biases stand out very clearly so there is something utterly
why decision making often goes wrong and whether AI they may not be unbiased every time, but at least they are nonsensical in all this talk about bias in AI. My guess is that
can provide answers. The interviews have been edited for noise free — if you ask the same question twice, you will it is going to be a rare case where the AI is more biased
length and clarity. get the same answer. Now, as we’ve all experienced with than the humans that it replaces.
ChatGPT or Bard or whatever your favorite large language
I think some of the resistance to algorithms that we see
model is, that’s not entirely true.
comes from a deep misunderstanding of what’s going on.
If you ask ChatGPT why it’s not giving you the same answer Humans are noisy and that’s why you sometimes don’t see
every time, its answer, tellingly, is that in order to look and how biased they are.

Q&A
Sibony: Using algorithms in the most generic sense What about situations where you have If you can move to the mindset where your own judgment
to make a judgment is going to be less noisy than using competing AI models or algorithms? is one of the inputs, another human is another input,
human judgment. That point obviously still stands. one AI model is another input and another AI model is yet
Kahneman: It’s going to probably produce less noise
ChatGPT is the exception, not the rule, and any simple another input, then you’ve recognized noise and you’re
than humans would and that’s how you’ve got to look
algorithm is noise free. But since we have a preference for doing something about it. You’re not giving precedence
at it. You’ve got to look at the alternative. And of course,
sophisticated and user-friendly algorithms over basic ones to a single point of view, which happens to be yours.
now people find it shocking when different models
that apply simple rules, it’s going to be more and more
disagree. But if they knew how much people disagree, So far it’s slightly theoretical because I haven’t seen many
of a problem.
they would be less shocked. situations where people are using more than one AI model
Should we use algorithms to try to solve the same problem and wondering which device
Of course there will be noise with multiple algorithms;
to eliminate noise completely? to follow. But more and more this is going to be the case.
that’s bound to happen. But there is something else
Kahneman: I think the moment you start becoming that happens when there are competing algorithms. The more inputs we have access to from various
aware of noise you know that you cannot eliminate it. Because typically algorithms are based on a lot of data, models, the more we recognize that under uncertainty,
And you don’t want to reduce it to zero because you want it’s also possible to reconcile them and reduce the noise a multiplicity of points of view is actually a good thing.
individuals to exercise their judgment. You do not want in a way that is really not possible when it’s people
to completely turn them into machines. generating noise.
There are costs to noise, but there are also costs to Sibony: If you’ve got several models, you will have several
reducing noise and you want to reduce those costs as points of view, just like when you have several humans.
much as possible. You want people to feel that they’re There is an advantage though here, which is that when you
expressing themselves. People need to see AI and have one point of view and it’s yours, that judgment seems
algorithms as noise reduction tools that help them rather to you a lot more correct than the judgment of another
than as bureaucratic constraints on the way that they human. When you’re looking at three different algorithms,
operate. You can’t get to zero noise if you’re using human you will tend to assume that each of them is to be taken
judgment, but zero is not the best possible outcome. with a grain of salt, and that’s a better attitude.

“Gartner Conferences offer a unique opportunity to network with your peers, New!
learn new and best-in-class practices and be exposed to new solutions all
in one event. Well worth the time to step away from the business for a few Most popular
days to expand your expertise.” 2023 Gartner Conference Attendee
Reintroduced!
Gartner CFO & Finance

Executive Conference
18 – 19 September 2023
11 – 13 September 2023
London, U.K.
Gold Coast, Australia
16 – 19 October 2023
Orlando, FL
6 – 9 November 2023
Barcelona, Spain
Gartner Supply Chain Gartner ReimagineHR Gartner Security & Risk

Planning Summit Conference Management Summit
30 – 31 October 2023 11 – 12 September 2023 26 – 28 September 2023
London, U.K. London, U.K. London, U.K.
29 – 30 November 2023 23 – 25 October 2023
Over 60,000 attended Phoenix, AZ Orlando, FL

4 – 5 December 2023
a Gartner conference Sydney, Australia
last year — Don’t miss

out this year Why attend Gartner
conferences
© 2023 Gartner, Inc. and/or its affiliates.

All rights reserved. EVTM_NC_2398966
ABB Electrification Lights Up
ML Models for Financial Planning
by Madison Wilson and Mike Lashinsky
Machine learning (ML) can make Alessandro Marchesano, head of FP&A at Switzerland- This ML pilot helped them gather unbiased data results and
based ABB Electrification (ABB EL), a business unit of ABB, gain executive leadership buy-in to build on the project’s
financial planning more efficient and recognized that making headway with ML in financial success. As a result, they plan to scale the use of ML to
accurate, serving as the basis for planning required his team to take the time to learn how the other financial planning activities, such as on-demand
generating meaningful forecasts that technology works and understand what it does best. The scenario analysis and planning assumption updates.
help executive leaders prepare for future other vital step was to identify the specific role humans will The team is also well-positioned to expand ML to other
play in building, training and governing ML models. areas of planning and, eventually, other areas of ABB.
disruptions. Yet, when FP&A leaders rush
to replace traditional forecasting with ABB EL tested a human-machine learning loop method that
empowered the FP&A team to:
this technology, their underdeveloped “Our forecasting transformation triggered deep
1. Integrate complex external drivers into ML models. discussions on key business drivers, is as accurate
models — trained on poor historical data
— can lead to untested algorithms that 2. Test, iterate and refine ML-based drivers. as our traditional bottom-up process, and is much
3. Refine the algorithms regularly to maintain performance. faster,” said Marchesano.
frustrate progress.

Integrate Complex External As part of this strategy, ABB EL »Figure 1. Weighting of Internal and External Drivers as ML Model Inputs
strengthened the ML algorithms with Illustrative
Drivers Into ML Models
more complex external drivers that
First, ABB EL created driver-based a business cannot directly control
Internal Drivers External Drivers
ML models, which are advantageous or influence, such as GDP or consumer
because they: sentiment. • Pricing • GDP
• Headcount • Labor cost index
• Explain the story behind a forecast. These external drivers change frequently, • Consumer sentiment
• Find material relationships between which trains the ML model to self-correct
business drivers and financial outcomes. and enables a quick, optimized business
response to shifting market conditions.
• Model short-term effects of external
Put simply, when one driver changes,
market changes or disruptions.
multiple algorithms can automatically
• Adapt plans to market volatility and revise their forecasts with limited human
better map business complexity. Driver-Based
oversight. This technique also provides ML Model
• Highlight the data with the largest the company with early-warning signals
influence on predictions. to model multiple future possibilities
(see Figure 1). Source: Adapted From ABB

Test, Iterate and Refine »Figure 2. Validation Process for Human-Sourced Drivers
Illustrative
Your ML-Based Drivers
Organizations tend to only refine drivers if there is a Source Validate Create
material macroeconomic change. But when executive
leaders overlook driver validation, they lose the
opportunity to better understand what spurs financial
performance.
Having recognized that the first set of drivers chosen for OUTPUT:
INPUT: Source Use both AI/ML and Drivers validated
each algorithm would always need fine-tuning, ABB EL’s Determine if drivers Data scientists
drivers for each traditional statistical by ML
FP&A team quickly set about substantiating the ML results are statistically use validated
ML model from models to test
to make sure they reflected financial performance. To relevant to drivers to
business and and verify business- Drivers invalidated
performance create ML
achieve this goal, FP&A created a three-step process to finance experts sourced drivers by ML algorithms
test the relevance of the drivers and measure their effect
on outcomes (see Figure 2).
The team:
1. Sourced external drivers for each complex business
area from business and finance experts.
Source: Adapted From ABB
2. Used ML and statistical models to validate those
business-sourced findings — and, if the FP&A team
hit a dud, business analysts went back to the first step
and continued to iterate until they uncovered the most
effective drivers.
3. Created algorithms with data science teams using
the validated drivers from the second step.

Refine the Algorithms Regularly • Do we need to refine any underlying trend models? »Figure 3. Learning Loop for Refining ML Algorithms
Many algorithms are based on seasonality or trend
to Maintain Performance
models. Answering this question tests if those models
After creating first drafts for each algorithm, the team are not performing as well as they could.
members didn’t stop there. They knew the first version of • Should we consolidate or break up any business areas
an ML algorithm was unlikely to be the best possible one. to make them easier to model? Sometimes, a business
Establish Identify a few
They found some initial assumptions were incorrect, area may be too large or too small for a single algorithm monthly review ML algorithms
drivers could be more specific or ML models didn’t to best represent it. This question reminds the team that cycle with data that are low
it might be necessary to adjust the size of the business scientists and accuracy and
accurately reflect the market environment. To combat
area in their models to make sure the algorithm performs FP&A team high materiality
this issue, they assessed ML performance on a rolling
monthly basis, hand-picking algorithms that missed the as expected.
mark on expected accuracy yet were critical to company Weeding out underperforming algorithms helped the team
performance. learn how to spot languishing ML models and correct how Evaluate the
they are governed, which establishes the human-machine refinement to Refine chosen
The team now has a framework to repeat this process until create continuous algorithms
they identify and improve all weak algorithms. learning loop. The models then iteratively improve over
learning loop
FP&A guides this refinement with four questions: time, making ABB EL’s financial planning speedier and
more objective (see Figure 3).
• Did we choose the most impactful inputs for the
algorithm? This question reminds them that it might
be necessary to revisit the drivers put into each
algorithm to make sure they are still relevant.
Source: Adapted From ABB
• Did we write the best algorithm to model this business
area? This question prompts data scientists to check for
a better way to create this algorithm and make it a more
accurate model.

Department
«
Briefs: Quick Takes on Fresh Research
Smarter Spending & Planning

Are Alternatives to Layoffs Better for Shareholder Returns? A Funnel to Send Cost Cuts to Growth Projects
Beyond hefty severance payouts, reducing the workforce can carry hidden costs. When CFOs promote cost-cutting measures in response to pressures on profitability, they
The short staffing that follows layoffs may lead to the need for expensive contractors. can end up starving, rather than protecting, their organization’s in-progress investments.
The remaining employees may demand faster promotions. And a vicious cycle may kick
But it doesn’t have to be this way. One answer lies in “cost savings winbacks,” a mechanism
off: overstretched people leading to lower morale, increasing turnover. When it’s time
that strikes the right balance between these competing priorities.
to get back to hiring, candidates may only accept offers at higher pay.
CFOs have a choice of two types:
Small wonder, then, that only 11% of organizations maintain layoff savings for three
consecutive years. C-suite executives must collaborate to define the potential financial 1. Direct return — Give a portion of the resource saved to the budget owners
and non-financial risks of layoffs. to use for projects that will drive long-term revenue increase.
Consider these alternatives: 2. Central — Send the money saved to an enterprisewide pool for reinvestments
designed to grow revenue based on organizational priorities.
1. Voluntary reduction in hours — Many employees may willingly work fewer hours
at commensurately lower pay. For instance, the finance function at a U.S. medical equipment company offered business
leaders concrete benefits based on central winbacks. They could submit proposals for
2. Internal redeployment — Even where headcount must decrease, some may transfer
what they would do with the money if they could reclaim half their savings.
to other parts of the business where their skills are in demand.
3. Sabbatical — An unpaid break can give employees the opportunity to pursue The result? Different business units competed to unearth more cost reductions
a professional interest that still contributes back to the company. — increasing the amount available to fuel growth.
4. Executive compensation cuts — Lower base pay temporarily while preserving They can use these funds as a cushion against budget overruns, without delaying other
long-term incentives. projects. Plus, it will bake the concept of uncovering resources into a “business as usual”
practice instead of a one-time action.
None of these options will make everyone happy, but they can keep a company
in position to recover faster when better days arrive.
— Vaughan Archer — Roma Kaur

Briefs
Talent & Culture

A Four-Day Workweek Is More Feasible Than You Think Three Cues to Help Employees Do Their Best and Keep It Up
The four-day workweek is a valuable tool for attracting and retaining talent, preventing Sixty-five percent of employees whose roles could be remote experience at least one
burnout and boosting flexibility. For most candidates, this benefit — even more than a of the following:
pay bump — is the top perquisite that would draw them to a position. Media buzz about 1. Devoting too much time to tasks that are necessary but not considered in their formal
this trend is not just hype: 21% of organizations have already started experimenting with workload (for instance, checking schedules to plan a meeting).
a shorter workweek.
2. Growing exhausted because they don’t rest.
What holds HR leaders back from putting the concept into practice? They worry
3. Feeling that more effort is futile because performance expectations seem unattainable.
that leaders will resist and that operational adjustments will prove too challenging.
Performance cues from the organization can signal to employees if they are working
Finding the model that works best for your organization can mitigate those
efficiently, setting a sustainable pace, and making progress.
perceived barriers.
These require both HR policy changes and management execution:
HR leaders have more options than they think. Ask:
1. Giving context so employees can work flexibly but waste less time on mundane
1. How many hours should employees work?
decisions — One HR team taught managers to facilitate discussing all stakeholders’
Choose a reduced (e.g., 32 hours in four days) or a condensed schedule
needs before setting team norms for how they will work together.
(e.g., 40 hours in four days), depending on the nature of employees’ work.
2. Building wellness into planning and execution of work — HR executives should consider
2. When should you offer the additional day off?
making the team’s well-being part of leaders’ performance goals so they see it as
Opt for a universal (e.g., Fridays off for all employees) or distributed (different days
essential to success.
off for different employees) weekday off, depending on business continuity needs.
3. Recognizing great work in real time — One total rewards group within an HR function
3. How widely should the program apply?
created a system for managers to pre-define high-performance actions and reward
Employees can decide whether to participate or you can mandate it for everyone.
teams when they deliver.
— Shivendra Singh and Amrita Puniani — Kayla Velnoskey, Iga Pilewska and Kate McLaren-Poole

Briefs
Growth & Innovation

Marketing Must Do Less to Get More and Drive Develop the Five Markers of Innovation Potential
Profitable Growth Technical expertise is common in the R&D workforce but is not enough to constantly generate
Chief marketing officers are investing in more technology, channels, data and new value to the enterprise. What R&D leaders need is a team with high innovation potential
personalization — with demonstrably diminishing returns. Few can afford to be stuck that can move beyond solving technical problems — identifying and developing novel
in this “cycle of more,” especially when companies face budgetary pressures and solutions for customer challenges and building stakeholder support for big bets.
are focusing on margins.
R&D executives looking for talent to nurture should check for these five behavioral markers
The key is discipline: creating fewer, but more meaningful, customer experiences. of innovators:
In both business-to-business and consumer purchase decisions, people who have just 1. Result seeking — They leave no stone unturned in achieving goals.
one interaction that changes understanding of their own needs and makes them feel
2. Customer empathy — They understand customer needs and expectations,
more confident moving in a new direction, are twice as likely to engage in commercially
and relate them to relevant technologies.
productive behaviors — such as paying a premium or referring other customers to the
brand. 3. Idea integration — They pinpoint issues and extract key information from large datasets.
4. Influencing — They network proactively.
This shift requires a compatible set of leadership techniques:
5. Risk taking — They take calculated risks, and identify opportunities and threats
1. Clarity — Decide what initiatives you will, and will not, support over the life of your
from market information and competitors.
strategy. Give equal thought to those that don’t make the cut, and why.
2. Connections — Success is almost always a collaborative endeavor. Start with a shared Managers should set performance objectives that develop these capabilities
understanding of where your function intersects with others to get meaningful work done. in addition to the usual proficiency oriented goals.
3. Courage — Defend your choice to do less, pushing back on urgent but unimportant Some examples: Internalize customer perspective while developing ideas and solutions.
requests. All major investments should be able to justify themselves based on Lead a topic-specific brainstorming session to generate unique insights. Work with
measurable contribution to future goals, not those in the past. business partners to agree on transformational targets and update them quarterly about
whether project teams are taking enough risk in their work.
— Sharon Cantor Ceurvorst — Amisha Ajay

Briefs
Data & Technology

Changes in Digital Business Strategy Make Cybersecurity A Shipbuilder’s Hackathon to Steer Toward Digital Technology
More Collaborative When organizations introduce a new technology, employees often lack the hands-on
Business owner accountability for cyber risk is fundamental to effective digital transformation. experience to leverage its full capabilities. General Dynamics Electric Boat found an
engaging and effective solution: an all-day, offsite hackathon. Participants learned
Sixty-seven percent of CEOs and senior executives want more technology work done how to make the most of a new digital visualization tool by creating deliverables for
directly within business departments. And 57% of enterprises have made business leaders real-world use.
formally accountable for the cybersecurity risks associated with their resources. That’s
sensible; cyber risk decisions cannot be made in isolation — organizations must balance High adoption was just one outcome. One team’s work resulted in a dashboard now in use.
such decisions with reputational, competitive and legal impacts. Several teams set up new visualization projects. Employees also began experimenting with
new technologies beyond the one they’d practiced working with.
As a result, those who drive digital change initiatives must join forces with the chief
information security officer (CISO) to: Three elements led to success:
1. Support effective cyber risk assessment with risk acceptance, escalation, exemption 1. Using real company datasets for hackathon activities quickly connected visualization
procedures and a representative steering committee that includes business unit leaders to participants’ daily responsibilities. It also made the outputs immediately useful for
to enable shared decision making. their job.
2. Make security policy more flexible by allowing choice among technology options that 2. Training high-potential talent as peer guides provided support where facilitators
achieve the same objectives. Additionally, rationalize the number of security policies, couldn’t. The fact that these colleagues had only recently received their own
while co-creating new policies and guidance with the business. introduction to the tool reduced skepticism in the ranks.
3. Go beyond traditional security awareness training with tools and playbooks that build 3. Inviting top executives to evaluate the outputs offered employees exposure to senior
business technologists’ independent cyber judgment. leadership, and motivated presenters to impress when they discussed their work.
Management benefited, too, by witnessing the technology’s impact and potential
4. Clearly communicate to the enterprise that the primary role of the CISO is enabling
firsthand, acquiring more ideas for innovation, and gaining an understanding of the
digital innovation — rather than enforcement of security controls.
business case for additional investment.
— Tom Scholtz — Jose Rosario

Access to Gartner Content Caveat
Some content may not be available as part of your current Gartner subscription.
© 2023 Gartner, Inc. and/or its affiliates. All rights reserved. Gartner is a registered trademark of Gartner, Inc. and its affiliates. This publication may not be reproduced or distributed
in any form without Gartner’s prior written permission. It consists of the opinions of Gartner’s research organization, which should not be construed as statements of fact. While the
information contained in this publication has been obtained from sources believed to be reliable, Gartner disclaims all warranties as to the accuracy, completeness or adequacy of
such information. Although Gartner research may address legal and financial issues, Gartner does not provide legal or investment advice and its research should not be construed or
used as such. Your access and use of this publication are governed by Gartner’s Usage Policy. Gartner prides itself on its reputation for independence and objectivity. Its research is
produced independently by its research organization without input or influence from any third party. For further information, see “Guiding Principles on Independence and Objectivity.”
Any third-party link herein is provided for your convenience and is not an endorsement by Gartner. We have no control over third-party content and are not responsible for these
websites, their content or their availability. By clicking on any third-party link herein, you acknowledge that you have read and understand this disclaimer.
795276
Actionable, Journal Journal
objective insight Q1 2023 Gartner Business Quarterly:

Help Your Talent Keep Transformation
Q2 2023 Gartner Business Quarterly:
Getting From Pilot to Innovation at Scale
On Track Helping C-suite leaders get more experiments into the
Position your organization for Achieve digital ambitions to thrive through upheaval business for the biggest impact and the best outcomes.
success. Explore these recent

Download Now Download Now
editions of the Gartner Business
Quarterly:
Webinar Webinar
Reduce Information Overload to Boost Your Don’t Own Innovation; Enable It Across
Business Strategy and Retain Employees the Business
Identify and mitigate information overload in Establish competitive advantage by enabling your
your organization. organization to innovate autonomously.
Already a client?
Watch Now Watch Now
Get access to even more resources
in your client portal. Log In

Gartner Business-Quarterly 3q2023

Uploaded by

Copyright:

Available Formats

Gartner Business-Quarterly 3q2023

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gartner Business-Quarterly 3q2023

Uploaded by

Copyright:

Available Formats

« «

Balance the Environmental Perils 4 Decisions to Make When

Gartner Business Quarterly 3Q23 | 2

Table of Contents Gartner Business Quarterly 3Q23 | 3

38 AI Systems ‘Need a Moral

10 Balance the Environmental

Gartner Business Quarterly 3Q23 | 4

Power Shifts Among Nations Prompt CEOs to Adapt Strategic Priorities

The Cutting Edge 3Q23

Changing the Mix of Geographic 51% Reglobalization

Changing the Mix of Geographic 48% p Statistically

Issuing the organization’s

Investing in Board ESG education

Table of Contents Gartner Business Quarterly 3Q23 | 6

Partially Meets Expectations 25%

28% for employees only Meets Expectations 16%

Exceeds Expectations 13%

No Yes Far Exceeds Expectations 21%

Table of Contents Gartner Business Quarterly 3Q23 | 7

50% 50% 24%

Director Quality and Information Access Meeting

Table of Contents Gartner Business Quarterly 3Q23 | 8

Build a strategic plan

that works — even in

volatile times! Set clear, strategic objectives for

Combat seven mistakes

Capture and communicate your plans

Download the Strategic Planning guide for your function.

Generative AI consumes a lot of electricity and water. The

Table of Contents Gartner Business Quarterly 3Q23 | 10

Table of Contents Gartner Business Quarterly 3Q23 | 11

Table of Contents Gartner Business Quarterly 3Q23 | 12

Table of Contents Gartner Business Quarterly 3Q23 | 13

Executive leaders weighing generative AI use cases Business Value Negative

Support sustainable product development

Table of Contents Gartner Business Quarterly 3Q23 | 14

Business Value and Feasibility

Table of Contents Gartner Business Quarterly 3Q23 | 15

With recent developments in generative AI, such as ChatGPT, AI is no longer

Download the research for more on:

Table of Contents Gartner Business Quarterly 3Q23 | 17

cases, cost and vendors. 70% 19% are currently piloting

Table of Contents Gartner Business Quarterly 3Q23 | 18

Impact Process & Infrastructure Governance

Table of Contents Gartner Business Quarterly 3Q23 | 19

KPIs and what if analysis:

Table of Contents Gartner Business Quarterly 3Q23 | 20

Table of Contents Gartner Business Quarterly 3Q23 | 21

Financial Services Healthcare and Life Sciences Education

Table of Contents Gartner Business Quarterly 3Q23 | 22

Retail Manufacturing Transportation

Supply chain optimization: Education and training: Direct an Customer interaction:

Table of Contents Gartner Business Quarterly 3Q23 | 23

From almost nothing to many millions of dollars, depending on:

Small and midsize enterprises … Larger organizations … LLM w/Privacy, Enterprise

Table of Contents Gartner Business Quarterly 3Q23 | 24

Table of Contents Gartner Business Quarterly 3Q23 | 25

Table of Contents Gartner Business Quarterly 3Q23 | 26

Table of Contents Gartner Business Quarterly 3Q23 | 27