How to measure the ROI of AI training
Record four numbers before training and the same four after: how many people use AI weekly, how long three or four named tasks take, a short knowledge test score, and how many assistants or automations the team built itself. Hours saved on the named tasks, times an hourly cost, against the price of training, is your return.
What should you record before the first session?
- Adoption: people using an AI tool at work at least weekly.
- Task time: minutes per occurrence for three or four named, repeated tasks — a proposal, a meeting follow-up, a weekly report.
- Knowledge: a short entry test on prompting, tools and safe use.
- Self-built tools: assistants and automations people built themselves (usually zero).
When should you measure again?
Use the same tasks and a mirror of the entry test. Measure at the end of the programme and again a month later — the second number tells you whether it stuck.
How do you turn it into one figure?
For each named task: (minutes before − minutes after) × occurrences a month × people doing it. Add the tasks up, convert to hours, multiply by a loaded hourly cost. Compare that monthly saving to the price of the programme. Use conservative numbers; leadership trusts an under-claim that is checked more than an exciting one that isn’t.
What does it look like in practice?
At ImLink, adoption went from one person to fourteen in four weeks and the team built nine agents of its own; tasks that took a day moved to minutes. See the case study. Our own client projects outside training show similar patterns on single processes: post-call admin down 80–90%, proposal preparation down 70–90%.
Which mistakes should you avoid?
- Measuring satisfaction instead of time. A happy cohort is not a return.
- Choosing tasks after the training. Pick them before, or the numbers will flatter.
- Counting licences as adoption. A seat is not a habit.
Questions
How soon can we see a return?
On named, repeated tasks, often within the programme itself. Count only what is still saved a month later.
What if people don’t track their time?
Time three or four occurrences of each task by hand before and after. A rough baseline is far better than none.
How we write these guides: figures about our own work come from before-and-after measurements on client projects (names withheld where under NDA); market figures come from the sources listed, with the date we checked them. Prices exclude VAT.
