A new survey from Tech.co asked 300 US business leaders, all at companies under 500 people, a simple question: how many hours a week is AI actually saving you? The answers split the room.
At the top, 22% said AI hands them back six to ten hours every week — most of a working day. More than half, 54%, reported some productivity lift. But plenty of owners sat at the bottom of the range, saving under two hours a week. Same year, same tools available to everyone, wildly different returns. The obvious question is what separates the two groups.
It isn't access. ChatGPT and Claude cost the same whether you get ten hours back or one. The survey points at something less convenient: the owners winning the most time were the ones spending between $1,000 and $2,500 a month on AI, while those spending under $100 typically saved less than two hours. Spending more isn't magic — but it's a proxy for something real. The people getting value treated AI as a place they deliberately moved work, not a toy they occasionally poked.
Where the hours actually come from
Look at what the high-savers handed over and a pattern appears. Writing topped the list at 29%. Research came next at 26%, customer support at 19%, meeting notes and scheduling at 16% each, bookkeeping at 15%. These aren't exotic. They're the big, recurring, time-eating chores that sit in the middle of every week — the stuff that's easy to do badly and exhausting to do consistently.
That's the tell. The owners getting ten hours back didn't sprinkle AI lightly across forty small tasks. They pointed it hard at the three or four jobs that genuinely consume their week and moved those jobs over wholesale. The owner saving two hours is using AI to occasionally reword an email. The owner saving ten has made AI the first draft of every proposal, the researcher for every quote, the front line on routine customer questions.
The tool wasn't the variable. Where they pointed it was. Ten hours doesn't come from using AI more often. It comes from using it on the right, heavy, repeating work.
The customer-support surprise
One number stood out. Of the owners who handed customer support to AI, four in five reported a productivity improvement — the highest hit rate of any task. That's notable because support is the function owners are usually most nervous about automating. It's customer-facing; the failure mode is public.
But that's also exactly why it pays. Support is high-volume, repetitive, and full of questions you've answered a hundred times. It's the definition of work a machine should hold while a human handles the handful of cases that genuinely need judgment. The owners doing this well aren't replacing their relationship with customers. They're letting AI clear the routine eighty percent so they have time for the twenty that matters.
What to do with this
If you're getting less out of AI than you hoped, the survey suggests the fix isn't a better prompt or a newer model. It's a harder look at where your week actually goes. Pick the one task that eats the most hours — the writing, the research, the repetitive replies — and move that one over completely for two weeks. Don't dabble across ten things. Concentrate on the heaviest one.
Ten hours a week is the difference between drowning in admin and having a day back to actually run the business. The owners getting it aren't smarter or better funded. They were just more deliberate about which work they refused to keep doing themselves.