Why AI time-savings math is usually wrong.

Short answer: the usual calculation multiplies hours spent on a task by a loaded hourly rate and calls the whole thing savings. Real automation recovers roughly 20 to 40 percent of the targeted task time. A business case built on the full number will miss, and it will miss in a way that is obvious about six months in.

Where the standard calculation breaks

Take a familiar shape. Someone spends ten hours a week producing a document by hand. Automate it, multiply ten hours by the loaded cost of that person, and the annual savings look excellent. Two things are wrong with that.

The first is that the task rarely goes to zero. It goes to review. Somebody still opens the output, checks the exceptions, and handles the cases that do not fit the pattern. That residue is usually a real fraction of the original, and on regulated or client-facing work it should be, because the alternative is unreviewed output going to a client.

The second is that recovered time is not automatically productive time. An hour returned in six-minute fragments across a week is not an hour of capacity. Zapier's own research put the gain at about 23 percent more productive time, and McKinsey has measured roughly 40 percent less context-switching. Both are real and neither is 100 percent.

What to use instead

Assume you recover 20 to 40 percent of the targeted task time in the first year, and build the case on the low end. If the project only works at full recovery, it does not work. Three adjustments make the number honest.

• Count the review time that remains, and count it at the rate of the person who will actually do the reviewing.
• Separate time that becomes capacity from time that becomes relief. Relief is worth having and it is not revenue. If the case depends on selling the recovered hours, say so explicitly and check that there is demand to sell them into.
• Put the error cost on the page. For most operational work the value is not only the hours. It is the missed deadline that did not happen, the follow-up that went out on day two, the notice that went out correctly. Those are frequently larger than the labour line and they get left out because they are harder to count.

Why we would tell you this

Because the alternative is worse for us. A project sold on an inflated number gets judged against that number, and no amount of real improvement survives being measured against a promise that was never achievable. We would rather scope against 20 to 40 percent, land it, and expand from something that visibly worked.

It also changes which projects are worth doing. Under honest math, the best candidates are not the ones with the most hours attached. They are the ones where the work repeats in a fixed shape, where a mistake carries a real cost, and where the delay itself is expensive. That is why lead response and deadline-driven paperwork tend to beat general administrative time, even when the administrative time looks bigger on a spreadsheet.

The free AI opportunity audit gives the top three opportunities in your business as ranges, for exactly this reason.