Buyer guide

    Measuring the return on AI training

    Start from the work, not the classroom. The three things worth measuring are how long specific tasks take, whether people are still working that way weeks later, and whether the quality of the output held up. Attendance figures and happy sheets answer none of those.

    This page sets out a method you can run yourself, the formula for turning minutes into capacity, an evaluation template to copy, and the caveats that keep the number honest.

    Five steps

    1

    Pick two or three tasks, not the whole job

    Choose tasks that are repeated weekly, take a measurable amount of time, and are done by most of the people being trained — a weekly report, a category of customer email, a first-draft document. Anything vaguer than that cannot be measured honestly.

    2

    Record the baseline before training

    For each task, capture three numbers: how long it takes now (minutes, self-timed over a couple of weeks is fine), how many times it happens per person per week, and how often it comes back for rework. Baseline collected after training is not a baseline.

    3

    Define what good looks like, and what safe looks like

    Write down the quality bar for each task — accuracy, tone, completeness, whatever a reviewer would reject it for. Alongside it, write the safe-use checks: what must be verified, what data must never be pasted in, and where a person has to sign off.

    4

    Set checkpoints

    Suggested rhythm: a light check at two weeks (is anyone using it?), a fuller measure at 30 days, and a repeat of the baseline measurement at 60 to 90 days. These are suggestions for planning, not contractual promises about results.

    5

    Measure sustained use and checking time

    Two things quietly decide whether the number is real. Are people still using the approach after two months, or did it fade? And how long does checking the output take — because a task is not faster if verification eats the saving.

    The formula

    Capacity hours = (baseline minutes − post-training minutes) × task volume ÷ 60

    Post-training minutes must include the time spent checking the output. Task volume is the number of times that task is done across the group in the period you are measuring.

    A hypothetical worked example

    The numbers below are illustrative only. They are not a Day Seven client result and we are not presenting them as a typical outcome.

    10 people, each doing the task 5 times a week.

    After training, and after checking the output, each run takes 12 minutes less.

    10 × 5 × 12 ÷ 60 = 10 hours a week of released capacity.

    If you assume an internal cost of £30 an hour, that is £300 a week of capacity value — an assumed rate for illustration, not a Day Seven price or a quoted saving.

    That £300 is not automatically £300 in the bank. It becomes cash only if a real cost is removed — overtime, contractor spend, an outsourced task. Otherwise it is time that has to be deliberately spent on something worth more than the work it replaced.

    Capacity released is not the same as cash saved

    Keep the two figures on separate lines of the business case. Capacity released is hours that stopped being spent on a task. Cash saved is money that has genuinely left the cost base. Report them separately so the business case says exactly what it means.

    On the cost side, include participant time in the session, the training fees, any AI tool licences attributable to this group, and internal support time. Count each item once. If a licence is already justified elsewhere, do not charge it here as well.

    Evaluation template

    Copy it into a document or spreadsheet before the training happens — the baseline fields are the ones you cannot go back for. No email address required.

    No sign-up needed — copy and use it
    AI TRAINING EVALUATION — [Team name]
    
    TASK 1: ..................................................
      Baseline minutes per run (before training):
      Runs per person per week:
      Rework rate before:
      Quality bar (what a reviewer would reject):
      Safe-use checks required:
    
      Day 30  — minutes per run (including checking):
      Day 30  — % of team still using the approach:
      Day 60/90 — minutes per run (including checking):
      Day 60/90 — rework rate:
    
    TASK 2: ..................................................
      (repeat the same fields)
    
    TASK 3: ..................................................
      (repeat the same fields)
    
    CAPACITY CALCULATION
      Capacity hours per week
        = (baseline minutes - post-training minutes) x weekly task volume / 60
    
      Task 1:            hours/week
      Task 2:            hours/week
      Task 3:            hours/week
      TOTAL:             hours/week
    
    COSTS TO SET AGAINST IT (count each item once)
      Participant time in the session (people x hours x internal hourly cost):
      Training fees:
      AI tool licences attributable to this group:
      Internal support/admin time:
    
    WHAT WE DID WITH THE CAPACITY
      Reduced backlog / increased volume / redeployed to:
      Any cash cost actually removed (be strict):
    
    OTHER THINGS THAT CHANGED IN THIS PERIOD
      New tools, headcount changes, seasonality, process changes, a new manager:

    Caveats that keep the number honest

    • Other things change at the same time — a new tool version, a quiet month, a process change, a team reshuffle. Note them, and do not credit the training with all of it.
    • Self-reported time is approximate. It is still useful if collected the same way before and after, by the same people.
    • Small groups produce noisy numbers. Two or three tasks measured well beats twenty measured loosely.
    • Faster is not better if quality drops. If the rework rate rises, the time saving is not a saving.
    • Attendance, satisfaction scores and completion rates measure the event, not the work. Keep them if you like, but do not present them as return.

    Related: AI training costs, choosing a provider, enterprise AI training and AI upskilling for staff.

    Frequently asked

    Because they tell you the session happened and people enjoyed it. Neither shows whether a task got faster or whether people are still doing it that way two months later. Start from task time, adoption and quality instead.

    Capacity hours = (baseline minutes − post-training minutes) × task volume ÷ 60. Use post-training minutes that include the time spent checking the output, otherwise the number flatters itself.

    No. Hours released are only cash if a cost is actually removed. Otherwise they are capacity that must be deliberately used for something.

    A light adoption check at two weeks, a fuller measure at 30 days, and a repeat of the baseline at 60 to 90 days. Those are planning suggestions, not guarantees about what will have changed by then.

    Participant time in the session, training fees, AI tool licences attributable to the group, and internal support time. Count each item once — if licences were already bought for another purpose, do not load them here as well.

    Agree the measures before the training

    Tell us the tasks you want to get faster. We will recommend the format, scope and delivery options, with pricing after scoping.