How to measure ROI on enterprise AI training (and the metrics that mislead)

Most AI training engagements report two things to leadership: attendance and satisfaction scores. Both are vanity metrics. They tell you the training happened. They tell you nothing about whether it worked.

Three metrics that actually predict ROI

1. Behavior change 30 days after the cohort

Survey the cohort 30 days after the final session. Ask one question: "Name the most recent task at work where you used an AI tool you learned in the training." Track the percentage who can give a specific, concrete example.

Healthy: 60-80%. Below 40% means the training didn't stick.

2. Reused prompts in the wild

Hand each learner a personal prompt library at end of cohort. Count how many of those prompts they've executed in the 60 days after. If your training is real, this number compounds.

Healthy: 5-20 reuses per learner per month, with at least 3 distinct prompts.

3. Cross-departmental questions

Track how often cohort alumni get tagged in Slack/Teams with AI questions from non-cohort coworkers. This is the multiplier effect: each trained person teaches their team informally. The training scales beyond the cohort.

Healthy: alumni report 2-5 questions/week within 60 days.

Vanity metrics to ignore

• Attendance. Mandatory training has 95% attendance. Doesn't mean anyone learned.
• NPS or satisfaction. People rate training they enjoyed. Enjoyment and learning are weakly correlated.
• Capstone completion. Capstones get completed because they're graded. Doesn't predict real-world transfer.

How Apache-3 / LearnTrainAI measures

We track all three healthy metrics for every cohort. If 30-day behavior change drops below 50%, we offer a refresher session at no cost. We've never had to do it for a cohort that completed our 4-week format. We have had to do it for half-day bootcamps where the time pressure was too short for behavior change.

We tell prospective customers this upfront. It's a contractual commitment, not marketing.

See our pricing →