Will that AI initiative actually pay back?
Your numbers, an honest range — and the weakest assumption named, the way a board would find it anyway.
How the AI payback period is calculated
Payback is how many months it takes for an initiative to earn back what it cost to build or buy. The calculator works it out in three steps.
- Yearly value. People affected × hours a week on the work × 52 × loaded hourly cost × share of time saved × share of people using it.
- Net yearly value. Yearly value minus the yearly run cost: licences, hosting and upkeep.
- Payback. One-time cost ÷ net yearly value × 12 months. If the net value is zero or less, there is no payback.
A worked example
Take the example figures in the calculator: a $120,000 assistant costing $24,000 a year to run, used for work that takes 6 hours a week of 40 people's time at $50 an hour. It saves 20% of that time and 60% of the team use it.
That is worth $74,880 a year, or $50,880 after the run cost, so the base case pays back in 28.3 months. Move time saved and adoption 40% either way and the window is 17.8–68.8 months. The same project, honestly described, could take under two years or well over five.
Which number to challenge first
Every input multiplies the result, so a mistake in any of them moves the answer by the same proportion. What differs is how likely each one is to be optimistic. The calculator names the one to question first:
- a time saving of 30% or more, because sustained gains that large are the exception;
- adoption of 75% or more, because use of internal tools tends to settle below the rollout plan;
- a run cost of zero, because very few tools cost nothing once live;
- otherwise, the hours a week, because people usually overestimate time spent on a task.
Measuring a sample of the real work before the decision is the quickest way to firm any of them up.
Questions about AI payback
How do you calculate the payback period of an AI project?
- Work out the yearly value: people affected × hours a week each spends on the work × 52 weeks × the loaded cost of an hour × the share of that time the tool saves × the share of people who actually use it. Subtract the yearly run cost (licences, hosting, upkeep). Divide the one-time cost by that net yearly value and multiply by 12 to get months.
What payback period is realistic for an AI work assistant?
- There is no honest single figure, because the answer is driven almost entirely by two of your own numbers: how much time the assistant really saves, and how many people keep using it. Test your case two ways: does the value still clear the run cost in the pessimistic case, and does it pay back well before you expect to replace the tool? If either answer is no, the case is not ready.
Why does the calculator give a range instead of one number?
- Time saved and adoption are guesses about how people will behave, and they are the numbers organisations most often get wrong. The calculator moves their combined effect 40% either way and shows the payback for the optimistic and pessimistic cases, so you see how much the answer depends on them.
What does it mean when the calculator says there is no payback?
- At those assumptions the yearly value never covers the yearly run cost, so the one-time cost is never recovered. That is a legitimate answer: it is cheaper to learn it from a calculator than from a project that has already been funded.
Want the projection checked against your own data? Our readiness assessment applies the same method to a real initiative. Or try the other free tools.