Boards, CFOs and employees all want to know what AI is doing to work. Most organizations can't say — Gartner's 2026 research found that while 95% of organizations implemented AI in the past year, only one in five realized significant or transformational value.
Measure at four levels
- Adoption: share of employees using AI weekly, by function and role; confidence; training completion.
- Work: hours per task before and after; cycle times; quality and error rates; capacity freed.
- People: headcount and open roles in AI-affected roles; internal moves and redeployments; entry-level hiring; attrition; engagement and concerns.
- Business and fairness: cost per unit of work, revenue per employee, customer or employee satisfaction, and selection-rate gaps in AI-assisted decisions.
Rules for credible measurement
- Baseline first: record the "before" for each use and role.
- Use comparison groups where you can: a team that starts later is a natural control.
- Separate time saved from value created: freed hours only matter if they are used for something.
- Look for unintended effects: quality drops, burnout, fewer learning opportunities for juniors.
- Be honest about expectations vs. outcomes: surveys of what leaders expect (for example, how many leaders expect AI-related headcount cuts next year) are not results.
A people-and-AI scorecard
One page, quarterly, for the executive team:
- adoption and confidence trend (from pulse surveys);
- capacity freed and how it was used (new work, redeployment, slower hiring, reductions);
- reskilling: people trained, people moved;
- fairness checks done and findings;
- sentiment and concerns, with actions.
The workforce impact mapper and the pulse tracker in the AI policy kit feed this scorecard.