Data Science: The Hard Parts

Chapter 14. From Prediction to Decisions

According to a survey done by McKinsey, 50% of their respondent organizations had adopted artificial intelligence (AI) or machine learning (ML) in 2022, a sharp 2.5x increase relative to 2017, but still lower than the peak reached in 2019 (58%). If AI is the new electricity and data the new oil, why did adoption stall before the advent of large language models (LLMs) such as ChatGPT and Bard?¹

While the root causes are varied, the most proximate cause is that the majority of organizations have yet to find a positive return on investment (ROI). In “Expanding AI’s Impact With Organizational Learning”, Sam Ransbotham and his collaborators argue that only “10% of companies obtain significant financial benefit from artificial intelligence technologies.”

Where does this ROI come from? At its core, ML algorithms are predictive procedures, so it’s natural to expect that most value is created by improved decision-making capabilities. This chapter goes into some of the ways that predictions improve decisions. Along the way, I will present some practical methods that will help you move from prediction to improved decision making.

Dissecting Decision Making

Prediction algorithms attempt to circumvent uncertainty, and doing so is extremely important in improving our decision-making capabilities. For instance, I can try to predict tomorrow’s weather in my hometown for the pure pleasure of doing so. But the prediction itself facilitates and improves ...