Michelle is the HR manager at a 70-person professional services firm. She sent 15 colleagues on an AI course, and the feedback was glowing. A month later, the managing director asks: "So what's changed?"

Michelle has attendance records and satisfaction scores. Neither answers his question.

Measuring AI training ROI starts before the training begins. Here's a simple way to link learning to real work, and put a fair number on the result.

I. Pick one task to improve

Choose a recurring task the trainees actually do, such as writing proposals, recording meeting actions or drafting monthly report commentary. Agree with the manager what good output looks like.

A useful goal might be: "Cut the time to produce an approved proposal, while keeping the existing price and scope checks." Set a target after measuring today's process, not by borrowing another company's figure. Our corporate AI training checklist can help you link the training brief to that task.

II. Measure a baseline before training

Record several normal examples first: time spent preparing, drafting, checking and correcting, plus the volume of work and common errors. Agree exactly when the task starts and ends.

After training, compare similar tasks over a set period. Note anything else that changed, such as new templates, staffing or a busier month. If you can, compare with a team that hasn't been trained yet. A before-and-after comparison is useful, but it can't separate training from everything else.

III. Track AI adoption through finished work

Are people using the new workflow on real tasks, without the trainer? If they've stopped, find out why. Common reasons are missing inputs, no access to the tool or no time to review.

Michelle's measurement sheet has five columns:

  • Task and period: how many tasks were eligible, and how many were completed

  • Time: hands-on time per completed task

  • Quality: rework and errors, using the same review criteria

  • Adoption: the share of eligible tasks done with the new workflow

  • Outcome: for example, faster turnaround or a smaller backlog

Tell staff what you're measuring and why. Track the workflow, not each person's prompt count.

IV. Turn time saved into a fair number

Here's an illustrative example. A team completes 80 similar tasks a month. Each used to take 30 minutes. After training, including review, it takes 22.

  • Time released: 80 × 8 minutes, about 10.7 hours a month

  • At an assumed staff cost of S$40 an hour: about S$427 a month

  • Less S$150 a month for new software: about S$277 a month

That's capacity, not cash. It only becomes a saving if spending actually falls, for example through less overtime. Otherwise, explain how the time is used, such as clearing a backlog or serving more customers.

V. How to calculate AI training ROI

Use benefits and costs from the same period:

ROI (%) = (benefits – total costs) ÷ total costs × 100

Include training fees, staff time spent in training, implementation effort, software and ongoing support. Don't count review time twice if it's already in your task times.

Continuing the example: a one-off training cost of S$3,000 plus S$1,800 of software makes S$4,800 for the first year. Twelve months of benefits come to about S$5,120. That's an ROI of roughly 6.7%.

Modest, and that's fine. It also assumes benefits start on day one with no extra costs, so show a more cautious version too, and agree the assumptions with your finance team.

VI. Decide what happens next

Review after the first month, then again once habits settle. Keep what works, fix recurring problems and stop anything that adds effort. A disappointing result is still useful if it stops a costly company-wide rollout.

VII. Common questions

When should we measure AI training ROI? Take a baseline before training, check after about a month and review again a few months later.

What if revenue impact is hard to measure? Start with time and error rates on one task. They're easier to track, and easier to trust.

Want your team's AI learning linked to a clear business goal? Explore our AI for Leaders programme or book a discovery call.

Michelle is an illustrative example. All figures are hypothetical.