Integrating Generative AI Into Business Strategy: Dr. George Westerman

Key Concepts

• Westerman's Law: Technology changes quickly, but organizations change much more slowly.
• AI is not intelligent: It's a program that executes based on formulas and learned patterns, lacking context knowledge.
• Four Categories of AI: Rule-based systems (expert systems), econometrics (statistics), deep learning, and generative AI.
• Digital Transformation Opportunities: Customer experience, operations, business models, and employee experience.
• Risk Slope: The need to grow risk management capabilities alongside the capability to do more with AI.
• Transformation with a little 't': Smaller, systematic transformations that build towards larger transformations.

1. Introduction

• Dr. George Westerman discusses how to think about generative AI and integrate it into organizational strategy from a management perspective.
• He aims to demystify AI, helping leaders understand its implications for organizational design and transformation.
• The presentation covers: (1) What is AI? (2) GenAI in organizations, and (3) How companies are innovating with AI.

2. What is AI?

• AI definitions are constantly evolving. Terms like "deep learning" and "traditional AI" shift rapidly.
• Westerman's Law: Technology changes quickly, but organizations change much more slowly. The hard part is not adopting the technology, but changing the way you do business.
• Example: Matthew Evans (Airbus) states they invest in solving business problems, not just AI technology.
• Example: Fahim Siddiqui (Home Depot) emphasizes creating extraordinary user experiences, with technology as secondary.
• Technology provides zero value until it's used to change the business or products.
• Key Point: Artificial intelligence is not intelligent. It executes a formula without context knowledge. Aude Oliva suggests thinking of AI as "artificial idiots."

3. Digital Transformation and AI Opportunities

• Digital transformation research identifies four key areas for opportunities:
  • Emotionally engaging, targeted, personalized customer experience.
  • Adaptive and adjustable operations (Industry 4.0).
  • Innovative business models (e.g., turning products into services).
  • Employee experience (satisfied employees lead to satisfied customers).
• AI is the next stage of digital transformation, offering more powerful opportunities.
• Examples of GenAI applications:
  • Creating virtual presenters for corporate literature in multiple languages.
  • Coding assistance with Copilots (improves coding and documentation).
  • Cresta: A call center tool that provides real-time hints to improve sales performance (MIT randomized trial showed 14% improvement for senior people and 34% for junior people).
  • Personalized tutors for programming classes (e.g., Python for minority institutions).
  • Integration into products like SAP, Workday, and Adobe.
• Key Point: The best solutions combine generative AI, traditional AI, IT, and human processes.
• Example: Lemonade (insurance) automates 98% of policy writing and first claim notices, and 50% of claims using a combination of AI and traditional systems. Complex cases are handled by humans.
• Example: Sysco (food service delivery) applies AI across customer experience and back-office operations. Generative AI can help with call planning, warehouse routing, and suggesting alternative products.

4. Four Categories of AI (Westerman's Perspective)

• Rule-Based Systems (Expert Systems):
  • If/then statements.
  • Useful for simple problems like prescriptions and loan making.
  • Requires talking to an expert to program.
  • Provides precise and consistent answers but does not adapt.
• Econometrics (Statistics):
  • Uses structured (numeric) data.
  • Cheap to program and works well for identifying trends and regressions.
  • Can handle multiple dimensions (e.g., analyzing 100 million resumes).
  • Provides precise and consistent answers but requires numeric data.
• Deep Learning:
  • Uses neural networks to process inputs through weighted averages and make predictions.
  • Trained with labeled data (e.g., identifying cars in images).
  • Outputs are repeatable but not explainable.
  • Example: Recognizing handwritten digits (0-9) using a neural net.
    • Converts a 28x28 image into 784 pixels (one-dimensional set of numbers).
    • Uses random numbers and adjusts them through repeated reinforcement to improve accuracy.
    • Requires labeled data and can be biased if the data is not representative.
    • Example of Bias: Amazon's resume review system rejected women because it was trained on data primarily from male engineers.
• Generative AI:
  • Generates new content by predicting the next best word or pair of words.
  • Randomly generates outputs, leading to different answers each time.
  • Can be used for creative tasks but also produces "hallucinations" (incorrect information).
  • Example: A lawyer used ChatGPT to prepare court documents, but it cited non-existent cases.
  • Requires huge training data and energy.
• Key Point: Start with the problem and choose the right AI technique based on accuracy needs, explainability requirements, data availability, and confidentiality concerns.

5. Making AI Work in the Organization

• Challenge: Transformation, not technology, is the problem.
• Three Challenges:
  • Prioritization: What to do, what not to do, and what to do first.
  • Risk Management: What if we are wrong? What about privacy?
  • Capabilities: Ensuring safety, value, and continuous learning.
• Governance Process:
  • Top-down (centralized): Safe but slow.
  • Decentralized: Fast but risky and costly.
  • Example: Societe Generale used a centralized approach, collecting 700 use cases and prioritizing based on foundational capabilities.
  • Example: Sysco uses existing technology governance, prioritizing buying over building and using simpler AI techniques when possible.
• Culture:
  • Is the culture ready for AI?
  • Do employees have the humility to work with AI?
  • Are there ethical considerations?
  • How good is the company at experimenting and failing fast?
• Skills and Careers:
  • Daniel Rock estimates 46% of jobs will have 50% of their tasks replaced by AI.
  • AI should make jobs easier and reduce cognitive load, not just replace workers.
  • AI can be a tremendous teaching tool.
  • Example: Dentsu Creative systematically introduced AI, focusing on boring tasks first and involving employees in the process.
  • Global Opportunity Forum: Companies discussing career development and skills needed for the future.

6. Transformation with a Little 't'

• Companies are doing smaller transformations (transformation with a little 't') to prepare for larger transformations.
• Three Levels:
  • Level 1: Individual Productivity: Using AI for tasks like summarizing documents and updating spreadsheets. Low risk.
  • Level 2: Specialized Roles and Tasks: Transforming call centers, coding, and other specific areas. Human-in-the-loop.
  • Level 3: Direct Customer Impact: Personalizing customer interactions and automating first-tier customer service.
• Example: McKinsey has an LLM that searches across all their slide decks.
• Example: Online retailers are using AI to personalize conversational approaches.
• Key Point: Large process transformations will likely involve combinations of GenAI and other technologies.
• Risk Slope: Grow risk management capabilities alongside the capability to do more with AI.
• Challenge: Getting from proof of concept to large-scale implementation is hard.
• Example: A large bank found that "the more stuff you do, the more stuff you find to do" (more errors to solve).
• Example: H&M's approach is like "putting a tire on a car" - tighten each bolt a little bit at a time to avoid bending the rim.

7. Conclusion

• AI can seem intelligent, but be intelligent in how you use it.
• Just because it's not perfect doesn't mean it's bad.
• Start with the problem, not the technology.
• Get started now and work up the risk slope.
• Help your people be ready.
• Continuously improve.