Where the time comes back, and how much of it is real
The recovery is about an hour a week at first, it comes from the dullest parts of your diary, and a fair slice of what gets called a saving is nothing of the sort.
Claims about AI productivity for executives usually arrive with a multiplier attached. Ten times the output. A working week compressed into a Tuesday. I train senior people on Claude for a living and have never seen that. What I do see, reliably, is about an hour a week at first, rising a little by the second month. It sounds like a poor return until you look at where the hour comes from.
The hour is made of unglamorous parts
Almost none of it is the interesting work. The hour comes out of the sixty-page pack that lands at eleven at night before a nine o'clock meeting, read properly for once because you asked for the five decisions buried inside it. It comes out of the workshop notes nobody has typed up, turned into something you can circulate while people still remember what they meant. Some of it comes out of the reply you have been avoiding for four days.
Anthropic's Economic Index puts numbers on this. Across all conversations, explanations account for 17% of what people produce, documents and reports 15%, and guidance 11%. Narrow it to work conversations and documents rise to 20%, ahead of explanations on 9%, email drafts 7% and analyses 6%. Nothing on that list would go on a CV, and that is rather the point.
Some of the savings are imaginary
Two corrections, both of which cost me money. The first is that a good deal of claimed saving has only moved. You did not spend forty minutes drafting the summary, you spent twenty-five checking it, and checking is duller work than drafting was. Fifteen minutes net, and a slightly worse afternoon.
The second is harder to hear. A task you did not need to do faster is not a saving. Producing an unread weekly update in half the time accelerates waste. I would rather a client dropped three things than did nine things quicker, which is an odd sentence for someone who sells training by the day.
The savings that hold up share a shape: you would have done the task anyway, at length, and you can tell within a minute whether what came back is right. Teams running steady, repeatable process work tend to find more of these than anyone else, which is why strategy and operations functions often show a return before the board does.
The hour is not really the prize
What changes is the threshold at which you are willing to attempt something. There is a real difference between a person who thinks an analysis would take a day so it will not happen this quarter, and a person who thinks it is ninety minutes and the first pass will be rough. The second person starts more things.
Some of those attempts turn out to be worth many times the hour that made them possible. This is also where the gap between colleagues starts to open, and it opens without anyone announcing it. Built In's write-up of Anthropic's research puts the consequence uncomfortably: the risk to your job is less the technology than the colleague who has learned to work with it.
What to do with the hour
Do not fill it immediately. Spend a few weeks noticing which tasks returned time and which ones only felt faster, because that difference is invisible from the inside.
Where the returns vary wildly from one colleague to the next, you are looking at a skills gap rather than a tooling one, and that is the thing to diagnose before you buy anything else. Point the hour at the work it is dependably good at, not the work you most wish it could do.
Nobody puts an hour a week in a business case. That is the only figure I have seen survive a second month, though, and I would sooner quote a small number I can stand behind than a large one I cannot.