Insights
The Hidden Problem With AI Time Savings
A mid-sized accounting firm I have been thinking about ran the numbers on its AI rollout last spring and found something it could not explain. The tools worked. Drafting time on client memos fell by roughly a third. Reconciliation work that used to consume a junior’s morning was finished before the second coffee. By any measure the firm had recovered thousands of hours across the year. Revenue was flat. Client work looked the same. Nobody could say where the hours had gone. This is not a failure of the technology, and it is not a failure of the people. It is a failure of allocation. And it is now, I think, the most common and least discussed problem in AI adoption.
A failure of allocation
BCG’s 2026 global survey of workers puts the shape of it plainly. Among frontline employees who use AI regularly, 42% report saving eight hours a week — a full working day. Yet 66% say they receive limited or no guidance on what to do with the time they save, and more than half say they are not redirecting it into more strategic work. The firms that redesigned the work around the tool were 24 percentage points more likely to see measurable business improvement. The firms that simply bought the tool moved the needle by about five points. So the hours are real. What is missing is a decision about them.
The same six hours
BetterUp Labs found the same gap among managers, who report saving around six hours a week. Roughly 42% of that recovered time went back into polishing work that already existed, and 33% went into administration. Both felt productive. Neither changed team outcomes. The managers who instead reinvested the time in developing their people, in their own growth, and in thinking further ahead saw their teams’ performance with AI rise by 65%. The same six hours. Radically different returns, depending entirely on a choice nobody was asked to make consciously.
Here is what I find striking. Organisations run rigorous processes to allocate every other form of freed capital. If a department returned $400,000 to the centre, there would be a paper, a committee, a decision. When AI returns the equivalent in hours, there is no paper, no committee and no decision. The hours are simply absorbed — into more of the same work, into administrative sediment, into a slightly gentler day.
I want to be careful here, because there is an obvious and ugly reading of this: that recovered time belongs to the employer and should be immediately re-extracted. That reading is both wrong and self-defeating. Some of that recovered time should quietly become slack, and slack is not waste — it is the condition under which people notice things. An organisation with no unstructured time has no capacity for original thought, only for throughput.
Spent on purpose
The point is not to reclaim the dividend. The point is that it should be spent on purpose rather than evaporate by default.
The accounting firm did something small and effective. It stopped treating recovered hours as a personal windfall and started treating them as a line item with an owner. Each team was asked, quarterly, to say what its recovered capacity had been spent on, choosing from four categories: deeper client thinking, developing someone, learning something the firm does not yet know, or absorbing existing demand. Absorbing demand was permitted. It simply had to be named. Naming it changed the behaviour. In the first quarter, almost everything landed in the fourth category. By the third, roughly a third had moved into the first three — not because anyone was instructed to, but because the question had made the choice visible.
That is the whole intervention. Not a target, not a mandate. A question asked often enough that a default became a decision.
Most organisations will spend the next few years measuring how much time AI saved them. The more useful measurement, and the harder one, is what the organisation became with the time.