Creating Transformative Impact with AI

Introduction

• Importance of questions and interaction during the presentation.
• Focus on AI in the context of delivery services.

Delivery Service Overview

• Main value proposition: Quick delivery of favorite restaurants to the door.
• Three key user groups: customers (eaters), couriers (riders), and restaurants.
• Optimization problem: matching orders, riders, and restaurant locations in real-time with uncertain variables.
• Business goal: Find best approximations for the optimization problem.

Initial Approach

• Inspired by Y Combinator's advice: "Do things that don't scale."
• Early solution: manual processes managed by McKinsey analysts using Google Sheets.
• Growth from manual systems to serious automation starting in May 2016.

AI Implementation

• Transition from manual to automated systems using intelligent automation and machine-learned models.
• Significant business impact by AI-driven full automation:
  • Reduced manual tasks, enhancing efficiency.
  • Faster and more reliable service for all stakeholders.
  • Key metrics improvements:
    • Late orders down 40%.
    • Order duration decreased by 12%.
    • Rider to restaurant time decreased by 13%.
    • Late pickup reduced by 48%.

Key Principles for AI Impact

Incremental ISM

• Breaking down complex systems into themes (travel-time model, rider delay model, food prep model, etc.) and iterating.
• Importance of live experiments at each step to ensure continuous improvement and responsiveness to market changes.

Bias to Simplicity

• Focus on simple, robust models (e.g., generalized linear models) backed by strong domain knowledge and empirical validation.
• Using simplest models within the acceptable error margin.
• Relying on practical implementation over theoretically perfect but complex solutions.

Rigorous Experimentation

• Necessity of live experimentation to validate models in real business environments.
• Complex experimental designs to ensure valid, causative data analysis (e.g., randomized block design).

Challenges and Solutions

• Handling competitive market landscape by focusing on short-term value delivery instead of long protracted builds.
• Using a mix of empirical data and theoretical knowledge to navigate the unpredictable nature of real-time optimization problems.
• Importance of having a capable team that is constantly iterating and validating through practical experimentation.

Technical Aspects

Machine Learning

• Heavy reliance on feature engineering and classic models like linear regression for scalability and performance.
• Importance of domain knowledge and robust sampling methods.
• Empirical validation favored over complex closed-form statistics.

Optimization

• Use of simple deterministic hill climbing with domain-specific insights over fancy solvers.
• Radical approximations of objective functions due to the impracticality of exact expectations.
• Emphasis on neighborhood operators for efficient optimization.

Conclusion

• The transformative impact of AI achieved through fundamental principles of incrementalism, simplicity, and rigorous experimentation.
• Importance of practical, hands-on approaches to solving real-world business problems.