When the lipstick no longer works….

Fixing our Legacy Debts to invest in our Future

HFS Summit, Cambridge University, September 26, 2024

PHIL FERSHT
CEO and Chief Analyst, HFS Research

© 2024 HFS Research. All Rights Reserved. 1

Five Seismic Human-made Disruptions

Disruption…
“Radical change
to an existing
industry or
market due to
technological
innovation"

Source: HFS Research, 2024

© 2024 HFS Research. All Rights Reserved. 2
2023 was about the WHAT 2024 is about the WHY 2025 will be about the HOW

Enterprise GenAI use cases

% use cases
AI Skills

Prediction 34%

ML
Culture Process
Personalization 28%

Productivity 24%
GenAI Data Tech

Others 14%

© 2024 HFS Research. All Rights Reserved. 3

 The Generative EnterpriseTM
is driving The Great Services Transition

Fixing 30+ years of legacy debts to capitalize

on AI-driven tech arbitrage: Generative EnterpriseTM Era
(AI-driven)
- Skills debt - Data debt
Generative AI
- Process debt - Technical debt ML
Process High degree of
mining mutual risk, trust
Value creation

IDP
Global Enterprise Era Tech and collaboration
augmentation RPA
(People-driven) DevOps
Lean & Six Sigma
Nearshoring Anywhere shoring
Offshoring • Additional 30-80% productivity in IT & process scope
Centralization & • New offerings to build, deploy, and manage AI-driven
standardization ecosystems
• Autonomous data-driven decision making and exception
processing
• Creative activities enable enterprise-wide end-to-end scope
1995 2000 2005 2010 2015 2020 2025
Time

© 2024 HFS Research. All Rights Reserved. 4

The “Scale of Technology Partnerships” becomes critical in the Generative Era
Services
Ecosystem
Orchestration

Applications
Consumer uses Enterprise stack Industry verticals Enterprise apps
General Law firms
Entertainment productivity
Creative
General and
administrative Health
Sales and
Productivity Defense
customer support
Agriculture
Marketing
and climate
Other
EPD, IT, security Construction

Infrastructure
Deploy and monitor Train and fine-tune models Open-source models & frameworks Full-stack large language models

Store and compute Hardware

Label and process data Data warehouses or lakehouses Cloud service providers

© 2024 HFS Research. All Rights Reserved. 5

 HFS Services and Ops Tech Vision 2030
Human Machine

Staff Technology- Platform-led AI-led Agentic Service-as-a-

augmentation enabled services services services Software
• Allows companies to quickly • Primarily driven by people but • Leverage built-in delivery • Augmenting human capabilities • Unlike traditional software-as-a-
fill skill gaps, scale teams up supported by proprietary platforms to enhance service with smart AI agents to optimize service (SaaS), this model
or down as needed, and solution accelerators, tools, delivery and efficiency. processes and decision-making. focuses on delivering services
maintain control over project and software. primarily through technology,
execution without the long- • Examples include Accenture • Examples of platforms include
minimizing human intervention,
term commitments associated • Most service providers use this SynOps, TCS Cognix, and Amazon Q, GitHub, Lyzr, Copilot,
Cognizant TriZetto, which Replit’s Ghostwriter, Google and maximizing efficiency.
with permanent hires. model to optimize processes
and deliver value efficiently, streamline operations and Gemini, Einstein Agent, Mindcorp. • Examples include startups like
such as Cognizant Neuro, provide consistent, scalable • Organizations like IBM and the Big rhino.ai, Now Platform, and
Infosys Topaz, TCS solutions. builder.ai
4 consulting firms are increasingly
WisdomNext & Wipro Lab45 adopting this model.
• Key Features:
• Flexibility: Easily adjust
• Key Features: • Key Features: • Key Features: • Key Features:
team size based on project • Human-Centric: Primarily • Integrated Platforms: Uses • AI-Augmented: Combines • Technology-driven: Primarily
needs. driven by skilled cohesive platforms for human expertise with AI agents. led by advanced software
• Expertise: Access professionals. service delivery. solutions.
• Cost-Effective: Achieves lower
specialized skills not • Tool-Supported: Utilizes a • Scalability: Easily scalable TCO through optimization. • Minimal Human Intervention:
available in-house. variety of technology tools and consistent across Reduces reliance on human
and accelerators. • Enhanced Capabilities: Expands
• Control: Maintain direct various operations. service potential with AI-driven resources.
oversight of projects and • Efficient: Enhances service • Efficiency: Enhances insights. • Efficient and Scalable:
processes. delivery through tech productivity and efficiency Provides efficient, scalable,
integration. • Typical commercial model:
• Typical commercial model: through platform support. Augmented FTE-based pricing or and consistent service
rate card • Typical commercial model: • Typical commercial model: outcome-driven performance delivery.
FTE-based pricing Transaction-based pricing pricing • Typical commercial model:
License / Subscription-based

Current Emerging
2000-2025 2025-2030
© 2024 HFS Research. All Rights Reserved. 6
Organizations are planning a phased implementation strategy to replace services with AI by 2030
Enterprises’ approaches to adopting AI to replace professional services

In 3-5 years Six out of ten major

enterprises plan to
replace people-run
services with
39%
37% software-run
services before 2030
24%

No major impact on our use of Anticipate replacing some Aim to replace a significant
professional services professional services with AI portion of our professional
solutions services with AI solutions

Sample: 1000 Major Global Enterprises

Source: HFS Pulse, 2024

© 2024 HFS Research. All Rights Reserved. 7

The Generative Era is not only about doing more with less… but also generating actionable business value

Enterprise GenAI use cases

% use cases

34%
28% 24%
14%

Prediction Personalization Productivity Others

• Facial skin analysis to

• Fast-food order • Accelerated
predict best treatments entry, app and compliance
• Soil analysis to predict menu checking
best crop use personalization
• Media
• CX in the
monitoring in
medical sector.
healthcare
products

Sample: 104 enterprise leaders actively exploring and deploying GenAI

Source: HFS Research, November 2023

© 2024 HFS Research. All Rights Reserved. 8

GenAI isn’t just ChatGPT… a plethora of LLMs are now available. Keeping pace is very challenging!

Most Popular / Widely Used Highly Recognized Notable for Technical Innovation Emerging or Specialized Usage Additional Models with Specific
or Niche Applications Contributions
1.GPT-3 and GPT-4 (OpenAI) - 11.PaLM (Google) - Known for its 31.ALBERT (Google) - A version of
Advanced AI capable of understanding performance in both language and 21.Wu Dao 2.0 (Beijing Academy of BERT optimized for lower memory 41.Retro (Google) - Incorporates
and generating human-like text, widely multimodal tasks using Pathways, a Artificial Intelligence) - Multimodal consumption and increased speed. retrieval capabilities into the model,
used in various applications. scalable architecture. model capable of understanding both 32.DialoGPT (Microsoft) - Tailored for enhancing information access.
2.BERT (Google) - Revolutionized 12.ERNIE series (Baidu) - Focuses on text and images. generating dialogues, simulating 42.Music Transformer (Google) -
understanding of context in language, enhancing model understanding by 22.MoE (Google) - Uses a "Mixture of conversational exchanges. Generates music with long-term
essential for improving search engines. integrating knowledge graph data. Experts" to scale efficiently to trillions of 33.Codex (OpenAI) - Geared towards coherence.
3.T5 (Google) - Converts all text-based 13.Megatron-Turing NLG (NVIDIA and parameters. understanding and generating 43.Luminous Base (Luminous AI) -
language tasks into a unified text-to-text Microsoft) - One of the largest 23.Pangu (Huawei) - A large model programming code, powering tools like General-purpose model aimed at a wide
format, facilitating a wide range of NLP language models aimed at natural focused on Chinese language GitHub Copilot. range of applications.
tasks. language understanding. processing. 34.ERNIE 3.0 Titan (Baidu) - An 44.Hive-Cote (Community) - Combines
4.RoBERTa (Facebook AI) - An 14.LLaMA (Meta/Facebook) - 24.Anthropic AI's Claude - Designed iteration that further integrates multiple machine learning models for
optimized version of BERT, achieving Recognized for providing high-quality with a focus on safety and reliability in AI knowledge graphs for improved improved predictive accuracy.
state-of-the-art results on various NLP performance with fewer parameters, systems. semantic understanding. 45.DeepSpeed (Microsoft) - A library
benchmarks. focusing on efficiency. 25.BigBird (Google) - Handles long 35.Whisper (OpenAI) - Specialized for designed to accelerate the training of
5.ChatGPT (OpenAI) - Designed for 15.Gopher (DeepMind) - Known for its sequences of data, making it suitable for speech-to-text tasks, featuring robust large-scale models.
conversational AI, providing responses large-scale and broad coverage of tasks like document summarization. performance across languages. 46.AdaGram (OpenAI) - Adjusts word
that are contextually relevant. diverse language understanding tasks. 26.Switch Transformers (Google) - 36.Dragon (Baidu) - Focuses on high- meanings based on their use context.
6.Transformer-XL (Google/CMU) - 16.CLIP (OpenAI) - Bridges the gap Introduces a technique to train very large performance across various NLP tasks, 47.Sparse Transformer (OpenAI) -
Introduced a novel way to handle long- between visual and textual content, models efficiently by using sparser heavily used in Chinese-language Implements sparsity to scale to larger
range dependencies in text. enabling models to understand images models. applications. contexts efficiently.
7.XLNet (Google/CMU) - Combines the via text descriptions. 27.Performer (Google) - Provides an 37.Lucy (OpenAI) - Emphasizes safety 48.MT-NLG (Microsoft) - Known for its
best of BERT and autoregressive 17.DALL-E (OpenAI) - Capable of efficient way to scale attention and ethical considerations in AI capabilities in natural language
models, handling permutation-based generating novel images from textual mechanisms in Transformers. deployment. generation, particularly in generating
training. descriptions, demonstrating creativity in 28.Reformer (Google) - Known for 38.M6 (Tencent) - A multimodal model coherent long texts.
8.DeBERTa (Microsoft) - Enhances the AI. processing long sequences using less designed for diverse applications, 49.BART (Facebook AI) - Blends the
BERT and RoBERTa models with 18.Chinchilla (DeepMind) - Optimized memory, making it efficient for large including text and image understanding. benefits of pre-trained autoregressive
disentangled attention mechanism. for training efficiency by using more datasets. 39.Z-code (Meta) - Focuses on models and autoencoders, enhancing
9.OPT (Meta/Facebook) - Open- data but fewer parameters. 29.ByT5 (Google) - Treats every input programming and technical tasks, both generation and comprehension
sourced alternative to GPT models, 19.Jurassic-1 (AI21 Labs) - Designed as bytes, simplifying the processing of assisting developers. tasks.
designed for scalable language to handle a wide variety of language multilingual text. 40.Alexa Teacher Model (Amazon) - 50.Leviathan (Facebook AI) - Designed
understanding. tasks, known for its versatility. 30.CTRL (Salesforce) - A conditional Enhances Alexa's interactions, for complex reasoning and knowledge-
10.LaMDA (Google) - Specializes in 20.BlenderBot (Facebook AI) - language model that can control style, improving its conversational capabilities. intensive tasks, pushing the limits of
generating more sensible and specific Designed for building more engaging content, and task-specific behavior. what AI can understand.
responses in dialogues. and natural long-term conversations.

© 2024 HFS Research. All Rights Reserved. Not Exhaustive 9

© 2024 HFS Research. All Rights Reserved. 10
 Enterprise GenAI use-cases show potential
Competitive Advantage Faster revenue growth and increased market share

• Accelerated Medical Diagnostics: AI in diagnostics streamlines medical

assessments, offering a competitive edge with faster and accurate diagnoses. • Utilizing GenAI for Market Analysis: Analyze market trends, consumer
behavior, and competitive pricing to identify opportunities.
• Proactive Equipment Maintenance: AI analysis enables proactive maintenance,
minimizing downtime and maintaining operational efficiency. • Continuous Monitoring of Competitor Activities: Keep track of pricing,
promotions, and product launches to adjust strategies promptly.
• Product Customization: Gen AI allows for product customization, meeting
diverse market demands and setting the company apart with unique offerings. • Optimizing Prices in Real-time: Adjust pricing strategies dynamically to
maximize revenue and capture market share.
• Personalization in Offerings: Offering personalized products and services using
generative AI gives a competitive edge by meeting individual customer • Streamlining Clinical Trials: Accelerate research timelines for faster
preferences. revenue growth.
• Predictive Trends Analysis: Utilizing AI for predictive trend analysis allows • Efficient Lead Qualification: Streamline the process of identifying and
companies to stay ahead of market shifts and anticipate consumer needs. qualifying leads.
• Streamlined Supply Chain Operations: AI optimization of supply chain • Enhanced Customer Experience: Implement GenAI in customer service for
operations reduces costs and improves efficiency, providing a competitive personalized and efficient support.
advantage in operational excellence. • Data-driven Pricing Strategies: Analyze customer purchase histories and
• Expediting R&D Processes: GenAI expedites research and development preferences for optimizing profitability.
processes, enabling quicker product development and innovation. • Predicting Demand Accurately: Reduce inventory costs and ensure timely
• Personalized Healthcare Solutions: Providing treatments based on individual delivery by forecasting demand.
genetic profiles offers a competitive edge through personalized healthcare • Precision Marketing Strategies: Enhance market share through
solutions. personalized engagement.
• Market Trends Monitoring: Employing GenAI to continuously monitor and • Amplifying Production Output: Use Generative AI to optimize production
analyze market trends enables rapid adaptation and staying ahead of efficiency and meet market demands.
competitors.
• Rapid Prototyping and Testing: AI-driven rapid prototyping and testing
accelerate innovation and reduce time-to-market for new products.

© 2024 HFS Research. All Rights Reserved. 11

Fueling the Hype…

© 2024 HFS Research. All Rights Reserved. 12

Reality-check: GenAI is a smoldering platform, not a burning one

Only 5% of enterprises have committed significant technology spend on GenAI and successfully deployed GenAI
solutions across multiple parts of their business

45%

27%
22%

5%

Deniers Wait and watch Fast Followers: Pioneers

No GenAI plans or spending No significant commitment Already committing Already deployed GenAI
to spend on GenAI and significant tech spend to solutions across multiple
mainly in planning stages GenAI or deploying GenAI parts of the business and
solutions across multiple committing significant
parts of their business technology spend to GenAI
Sample: 550 Enterprise Leaders
Source: HFS Research, 2024

© 2024 HFS Research. All Rights Reserved. 13

Can Enterprise Middle Management Withstand the C-Suite’s Impatience?
C-Suites are bullish on GenAI’s
benefits, but they are impatiently
witnessing their organizations
“Wait and Watch”

Enterprise Middle Management is strangled by: The current response is:

1) Everyone jockeying to show some implementation in 1) Middle management is funding GenAI

GenAI without governance, limiting ability to focus from leftover change in their budgets.
investment on what matters.
2) GenAI subjected to “death by a
2) Traditional funding being tapped out by high fixed costs thousand pilots”
of legacy IT app development, maintenance, and
infrastructure. 3) Few meaningful large scale impacts
3) Slow delivery of meaningful GenAI functionality by many 4) Enterprises can’t get past data privacy
legacy technology software providers, so the business rules/controls
is cobbling together custom implementations
4) Internal skill gaps – its all so new.

© 2024 HFS Research. All Rights Reserved. 14

HFS is All-in on GenAI:
Research in the Context of YOU!
Impact of adopting GenAI as a game changer:

• First to market – launched in June 2023

• HFS has uploaded over 1,300 pieces of unique

research

• Since January 2024 we have had over 110,000

questions asked and answered by our GenAI tool

• No more document-based searching, our insights

now come in the context of your unique needs.
“Humata and HFS have created a game changing
experience that delivers contextual answers via a
secure, private genAI and eclipsing traditional
search.”
– Cyrus Khajvandi, CEO Humata.ai

© 2024 HFS Research. All Rights Reserved. https://www.hfsresearch.com/research/trygenai/ 15

HFS disrupts the analyst industry with our GenAI-based research platform

Analyst value survey

Organizations (Tech buy-side)
2023 Revenues

79% HFS Research

65%
59%
45% 44% Forrester
33% 29% 25%

Gartner

Gartner HFS Forrester IDC Everest ISG Frost & Omdia

Research Research Group Sullivan (Ovum,
Informa)

Sample: Demand-side organizations that responded to the Analyst Attitude Survey conducted by the Analyst Observatory at the University of Edinburgh.

© 2024 HFS Research. All Rights Reserved. 16

 GPT-4o with Eyes

© 2024 HFS Research. All Rights Reserved. 17

 Can GPT-4o pour fuel onto the GenAI smoldering platform?

1. Multimodal makes everything much more human. Text, vision and speech into the same neural network…Old
GPT was like texting a friend, GPT-4o is like calling a friend
2. Real-time human-2-machine conversation is now possible. converse naturally without first converting words to text,
with real energy, emotion, and expressiveness
3. Enhanced multilingual support and capabilities

4. It really does have human eyes now

5. It’s being incorporated into Apple’s iPhone and Google’s Android operating systems.

6. Summaries are concise and relevant. finally the end of legacy Google search and bad call transcripts?

7. Visual interpretation and data tables are much more usable and accurate, ready to support business needs. It
accurately converts image data into a clean table format without misinterpretations
8. Image generation capabilities are just so much sharper.
9. The cost of accessing its APIs is 50% cheaper. such as Chat Completions API, Assistants API, and Batch API

10. Coding is vastly improved. solve many coding projects, such as multiple thousand lines of code in under 10 minutes,
which previously took prompt engineering processes many hours

© 2024 HFS Research. All Rights Reserved. 18

Winners in the Generative Era will require strong business acumen

What percentage of your enterprise’s technology-related spending is controlled by IT?

The new growth mantra

for IT services:

As traditional IT services
Tech Spend start to flatten, we need to
Outside of align our value proposition
Tech Spend more closely with the
IT control, Under IT increasing tech-spending
46% control, 54% outside IT control

Sample: 551 Global 2000 enterprise executives

Source: HFS Pulse, 2023-24

© 2024 HFS Research. All Rights Reserved. 19

We need to pay our debts before we can become a Generative Enterprise

Which are the most significant challenges in implementing GenAI in your organization?
61% 60%

39% 38%
34%

19%
10%
5%

Access to data Regulatory Complexity of Skill gaps Building the Costs of Narrowing Setting and
and compliance & integration among required tech implementation down the following
maintaining security risks employees stack number of governance
data quality potential use policies
cases

1. Data debt 2. Process debt 3. Skills debt 4. Tech debt

Sample: 550 Enterprise Leaders

Source: HFS Research, 2024

© 2024 HFS Research. All Rights Reserved. 20

AI isn’t replacing jobs… but may get replaced by someone who understands AI

© 2024 HFS Research. All Rights Reserved. 21

 The ability to combine business and technical skills will be critical to succeed in the AI-led era

Business skills needed to drive successful GenAI initiatives Technical skills needed to drive successful GenAI initiatives
N =550 N =550

The ability to combine business skills

with technical knowledge 44% Data science and machine learning
expertise 53%
Data analysis, interpretation, and
Understanding of ethical AI practices 41% 41%
visualization
The ability to combine technical
Critical thinking and analytical problem- knowledge with business skills 35%
solving 39%
Software development and
engineering skills 29%
Effective project management and
execution 39% Cloud computing and deployment
experience 29%

Adaptability and agility 36% Data engineering and big data

technologies 25%

Strategic planning and business Data Security and Privacy

36% Awareness 23%
acumen
DevOps and model maintenance
skills 21%
Industry-specific knowledge (e.g.,
healthcare, finance, etc.) 30%
Natural Language Processing (NLP)
knowledge 20%
Communication and collaboration skills 28% Prompt Engineering 16%

Sample: 550 enterprise leaders

Q: What are the top three business skills you believe most crucial to drive the success of your GenAI initiatives? Source: HFS Research, 2024

© 2024 HFS Research. All Rights Reserved. 22

Transformer-based models produce a colossal
carbon footprint

© 2024 HFS Research. All Rights Reserved. 23

© 2024 HFS Research. All Rights Reserved. 24
So where do we go from here?
AI is a smouldering platform, not a burning one, but the key now is to scale
what we have, and to improve what we have before we add more.

Our industry has obsessed with shiny new tech and not real business change.
This has to change

We must fix our past debts to invest in the future

It takes both enterprises and their partners to learn and work together

AI won’t replace people, but people who don’t work with AI are at risk

We are barely more than 18 months in, so let’s all take stock, there is time to
adjust, learn and adapt
© 2024 HFS Research. All Rights Reserved. 25