Owners get told AI hands back hours. My read on a real week is different. AI saves minutes, not hours, four here, eleven there, spread too thin to hand to anyone. Part of it goes to checking the work the tool just did. The rest fills back up with new tasks before Friday, and if nobody named a home for the freed hours, the week spends them for you. I think the gains are real. They’re just a different shape from the one you were sold.
Canadian data points the same way. Statistics Canada found that 44.4% of businesses using AI changed their training or staffing practices because of it. The tool changed the work. Handing back hours is a separate question.
You did nothing wrong in the buying. The stopwatch story is the story every seller tells, because a saved hour photographs well and a saved eleven minutes does not. But a firm does not run on the demo week. It runs on ordinary weeks, and in an ordinary week the honest unit of AI savings is the minute.
What does a saved hour look like up close?
It looks like small change. Four minutes off an email. Eleven off a summary. A first draft that starts halfway down the page instead of at the top, partly because your clients are prompting the first draft themselves and the work arrives half-shaped.
None of it arrives as a block you can assign. It arrives as gaps between jobs, and gaps that small get absorbed by the day before anyone can spend them on purpose. Where the minutes do stack into real capacity, I’d point them at work a small firm used to turn down.
Add the pieces across a month and they are not nothing. On paper they might stack into a few real hours. But each piece landed on a different day, in a different gap, and the day that held it spent it. A saving you cannot gather is still a saving. It just behaves like loose change in a coat pocket, not like money in the bank.
How much of the saving gets spent checking?
Some of it, and the share is bigger than the pitch admits. The draft the tool wrote in two minutes takes six to read properly, because reading for errors is slower than reading for sense. A figure needs its source checked.
A name needs confirming. The tone needs one pass so the reply still sounds like your firm and not like a brochure. Nobody warns you about this at the demo, because at the demo nobody checks the output at all.
Then comes the habit that eats what is left. One last prompt. One more version of the paragraph, one more angle on the quote, one more polish on the summary, each one cheap because the tool is fast and asking costs nothing. Twenty one-last prompts later, the minutes the tool saved are sitting in versions nobody will ever send.
The tool saves minutes with one hand and bills part of them back in checking with the other.
Why does the saved time fill back up?
Because open time in a working week never stays open. The report that took an afternoon now takes an hour, and by the next week the calendar holds two more reports. Work expands into the room it is given, and AI gives it room.
The refill is not all work you chose, either. Some of it was never yours. The task that used to sit with a supplier or a junior now lands on your desk because you are the one with the fast tool. Being quick makes you the path of least resistance, and the week routes its errands through you.
What should you buy the tool for, then?
Buy it for what the workflow puts out, not for the minutes it shaves off. A workflow is the whole run of a job, from the request landing to the finished piece leaving the building.
If that run ships more good work at month end, with fewer stalls and fewer late nights, the tool earned its money even if the stopwatch story never came true. That’s the test I’d use on any tool you’re paying for.
It is the same discipline as pricing AI by the finished job. Count the piece that comes out of the end, finished and usable, and stop counting the motion in the middle. Minutes are motion. Output is the product.
Which workflow should you track for five days?
Pick one workflow your team runs every week. For five working days, write down where the saved minutes on that workflow actually went. Checking. Extra prompts. Work that refilled the gap. Work that was never yours to begin with.
Keep the list rough and honest, on paper, beside the screen.
Then set the judgement date. At month end, look at what the workflow shipped, not at the minutes anyone counted in the demo week. More finished pieces, fewer stalls, quality your name can still ride on: keep the tool, and buy the next one the same way. Output flat and the month no calmer: the saving was a story, and the renewal is where you stop paying for it. Measuring a workflow by what it ships is where my AI automation systems work starts.
Judge the month, not the demo.
Short answers on does AI save time
Does AI actually save time at work?
Yes, but in minutes, not the hours sellers promise. The savings land as small pieces across the week, four minutes here and eleven there. Part goes to checking the output, and the rest usually fills with new tasks. Judge the tool on what the workflow ships.
Why do AI time savings disappear?
Three things eat them. Checking the output takes longer than the pitch admits, cheap extra prompts use up what is left, and open time in a working week always refills. Being fast also makes you the path of least resistance, so other people’s errands land on your desk.
My take: track one workflow for five days before you judge the tool. More traps like that show up weekly in the AI Newsletter with ZAK.
2 Comments