Mary Fung
essayJune 28, 2026

Measuring AI team progress

The easy AI metrics count activity. The useful metrics show whether the work, quality, cost, or customer experience changed.

The easiest AI metrics are usually the weakest.

Number of employees trained. Number of tools used. Number of prompts written. Number of pilots launched. Number of demos shown. Number of ideas generated.

These numbers may help with administration. They do not prove the team is better.

A team can train everyone and still change nothing. It can launch pilots and still retire no work. It can use five tools and still make worse decisions. It can show impressive demos and still have no adoption.

Better metrics sit closer to the work.

Speed matters when the cycle time actually changes. Did a recurring workflow move faster from request to usable output?

Quality matters when the output is more consistent, more accurate, easier to review, or less dependent on one person.

Cost matters when something gets smaller, simpler, automated, redeployed, or removed. If AI only adds another layer, the cost story is probably incomplete.

Customer experience matters when the person receiving the work gets a clearer answer, faster response, better interface, or more useful next step.

Employee leverage matters when the same person can do higher-quality work, not just more work. More output is not automatically more leverage.

Decision quality matters when leaders can see better context, clearer tradeoffs, and stronger evidence before choosing.

Workflow adoption matters when people keep using the new path after the experiment is no longer being watched.

Revenue impact matters, but it is often downstream. Before a team can claim revenue impact, it should be able to show the workflow, adoption, and behavior change that plausibly created it.

The metric to avoid is "AI usage" as a proxy for value.

Usage can mean value. It can also mean curiosity, pressure, confusion, or people using a tool because leadership asked them to. It is a starting signal, not an outcome.

The best measurement question is plain:

What got better, what got cheaper, what got faster, what became easier to trust, and what stopped?

If the team cannot answer those questions, it is not measuring progress. It is counting motion.

← back to the field