2. Find the value

Where AI can help is not the same as where it will matter most.

Working with AI: Measuring the Applicability of Generative AI to Occupations — Microsoft Research

Scores occupations on how much of their work AI is already helping with, based on real conversations with Copilot. Useful for a first view of which role families to look at. Data is public at github.com/microsoft/working-with-ai.

Anthropic Economic Index and Labor market impacts of AI — Anthropic

Maps how AI is actually being used against O*NET's task lists, and tracks which occupations see the most real-world use. Good for separating what's theoretically possible from what's happening.

PI approach: weigh value, not just exposure

Research shows which roles are exposed to AI. It doesn't show where value is highest in your organization. For that, weigh each task:

Value = time per task × frequency × number of people × how well AI fits — minus the risk if it's wrong

The number of people is the multiplier most organizations overlook. Thirty minutes a week saved for 200 people beats three hours a week saved for two.