Answers

How to measure the ROI of AI training

Measure AI training against work the business already tracks. Before training, record a baseline for one task, such as cycle time or rework rate. Measure the same task afterward and check where any saved time went. The return is the change in that measure plus the value of redirected hours, less the training cost and people's time in it.

Take the free AI Profit Readiness Assessment

Why completion rates don't answer the question

Most training reports count who finished the course and how they rated it. Those numbers record attendance. A leader asking about return wants to know whether the work changed, and a completion rate cannot say.

Training delivered and skill gained are far apart. In BCG's 2026 survey, 72% of workers said the skills expected of them have shifted, while only 36% felt they had received adequate upskilling (BCG, June 2026).

A five-step method

1. Pick one task and one measure. Choose work that happens every week and a measure the business already reports, such as cycle time or rework rate. A number nobody recognizes gets argued about instead of acted on.
2. Record a baseline before training. A few weeks of the measure, taken before anyone is trained, is enough. Without it, nothing afterward proves anything.
3. Train inside the work. BCG found that people who receive five or more hours of training, especially in person and with coaching, are significantly more likely to become regular users (BCG, June 2025). Use the task you chose in step one as the training material.
4. Measure the same task afterward. Check at 30, 60 and 90 days. Early numbers often dip while people learn, so judge the trend.
5. Follow the hours. If the task now takes less time, find out what that time went to. Saved time that nobody redirected has been absorbed, and it is not a return yet.

The arithmetic

Training ROI = (value of the change in the measure + value of hours redirected to work that earns − training cost − cost of time spent in training) ÷ (training cost + cost of time spent in training).

Count saved hours as value only when you can name the work they went into. Counting every saved hour at full salary is the most common way AI training ROI gets overstated, and finance teams spot it quickly.

What to count, and what to leave out

Count	Leave out, or report as context
The change in the task measure against its baseline	Course completion rates
Hours freed and redirected to named work	Satisfaction scores for the training
Rework or errors on the same task	Login counts and seat utilization
The full cost, including people's time in training	Projected savings no one has seen yet

Who should own the measurement

The team that ran the training should not be the one grading it. The leader who owns the task owns the measure, because they already report it and will still be reporting it after the program ends. Ask finance to agree, before training starts, how a redirected hour will be valued. Settling that in advance stops the result being argued about afterward, and it keeps the number credible when it goes to the CEO or the board.

Keep the first measurement small. One team and one task, measured properly, gives you a result you can defend and a method the next team can reuse.

When the number comes back flat

A flat result tells you which link to look at. The AI Profitability Gap™ is the distance between what a company expected AI to deliver and what it is delivering, and training only touches one of its three conditions, Skill. If people are more skilled and the measure still has not moved, the AI work may not be pointed at that measure (Direction), or the saved time may be going back into the same workload (Reinvestment). Look at both before buying a second round of training.

Frequently asked questions

What is a good ROI for AI training?

There is no reliable public benchmark for AI training ROI, so set the bar against your own baseline. A training investment is paying back when a measure the business already reports has moved and the freed time has gone into work that earns.

How long after AI training should I measure ROI?

Measure at 30, 60 and 90 days after training, on the same task you baselined beforehand. Early results often dip while people learn, so judge the trend across all three points.

Can I measure AI training ROI with completion rates or surveys?

No. Completion rates and satisfaction surveys show that training happened. They cannot show whether the work changed, which is what return depends on.

How many hours of AI training do employees need?

BCG's 2025 AI at Work survey found that people who receive five or more hours of training, especially in person and with coaching, are significantly more likely to become regular users. Hours spent on real work count for more than hours spent on demos.

What if usage went up after training and the measure didn't move?

Then training raised usage and something after usage is blocking the return. Check whether the AI work is pointed at that measure, and whether anyone decided what the saved time is now for.

Next step

Before you buy more training, see whether Skill is your biggest shortfall. Eight questions, about two minutes.