Essay

Time saved by AI evaporates unless someone decides where it goes

Two new measurements point the same way: companies are using more and more AI, yet it barely shows in revenue or margins. The technology works. What is missing is a decision about the time it saves.

November 2, 2026 · Levy van Bremen · Lees in het Nederlands

On October 24, the chief economist of the American investment firm Apollo published a short note with a blunt conclusion: outside the tech sector, AI has not yet delivered any financial payoff to American business. He compared the seven big tech companies with the other 493 companies in the S&P 500, the best-known American stock index. At the tech giants, revenue per worker and profit margins are rising. At the rest, both are flat, and he sees no trace of AI in the productivity figures either.

Nine days earlier, KPMG sent a similar signal from the Netherlands. In its quarterly survey of more than 2,100 board members and senior executives worldwide, the share of Dutch organisations saying they get meaningful value from AI fell from 82 to 68 percent in a single quarter. More telling still: only 4 percent place themselves in the phase where AI delivers demonstrable business results and growth. According to KPMG, technical obstacles are shrinking while organisations increasingly struggle with organisational ones.

The obvious conclusion is that AI is overrated. I draw a different one. The time AI saves is real, but it is spread thinly across individual minutes for individual employees, and that is where it evaporates. AI only shows up in the numbers when you decide in advance which clearly defined piece of work disappears and where the freed-up hours go.

Where the minutes go

This spring, the Dutch research organisation TNO studied what AI does in the daily work of four organisations. At LINKIT, which matches candidates with clients, AI writes offer letters and project descriptions. At Dutch Customs, an AI tool helps staff search the manuals faster. In both cases, TNO found mostly small time savings on specific tasks. And little thought had gone into that time in advance. The researchers say so plainly: how small time savings are filled with other tasks is often not considered beforehand.

TNO lists what can happen next. The time goes into more of the same tasks, the bar is raised over time, or the time is not filled at all. In that last case, work pressure probably drops, but the productivity gain is never realised. The report refers to organisations where more than half of employees use AI and save 4 percent of their time, while no more work comes out of it.

The mechanism is simple. Ten minutes saved per day, spread over five tasks, changes nothing about a schedule. Nobody takes on an extra client because of it, nobody hires one temp fewer. Payroll stays the same, revenue too, and in the annual figures the AI is invisible. The technology does what it promises. It has simply been deployed as a tool that makes everyone a little faster, without anyone deciding where the difference goes.

What does show up in the numbers

The fourth case in the TNO study shows the other pattern. At HelloPrint, AI automatically answers customer questions by chat and email. That is a whole piece of work, from incoming question to sent reply. According to TNO, customer service now needs roughly 80 percent fewer full-time positions. The work that remained became more varied but also harder: staff get the difficult questions and have to fix the AI's mistakes.

Two details make the difference. HelloPrint had written out all its procedures systematically, so the AI tool could work with them. And the gain came from work no human has to do anymore, which forced the company to decide what to do with the hours that freed up. That decision is exactly what is missing where AI is handed out to everyone as loose help.

This is where Apollo's argument fits, although I draw a different lesson from it than the economist does. He writes that AI may be a great calculator, but in business, intelligence is mostly judgment. That is true, and it explains why AI yields so little in work full of trade-offs: every employee still judges and checks, just slightly faster. The payoff lies in work that needs little judgment and keeps coming back, and in a deliberate choice about where human judgment stays.

What I see myself

I notice it in my own work. Every morning, AI agents run for me, programs that carry out a series of steps on their own, one for each area: sales, finance, marketing and releasing apps. At six o'clock an overarching agent follows that keeps track of what is still open and makes proposals. At the end of the day I receive an overview of what is done, what is still waiting and what should take priority tomorrow. On a normal day, something runs more than six hundred times.

The gain lies in work that has disappeared: nobody has to request and gather all that information anymore. But the same overview also shows me every evening which tasks are still waiting for me. The system has taken over the gathering. What I do with the time that frees up, it does not decide. That choice stays with me, and that is how it works in every company.

The strongest counterargument: give it time

The most serious objection is that it is simply too early to judge. Statistics Netherlands (CBS) also found last year that, after correcting for company size and other characteristics, there was no meaningful productivity difference between Dutch companies that use AI and those that do not. It expects the effects to become visible gradually and over a longer period. New technology needs time before organisations are built around it.

There is something to that. But waiting only helps if the gains pile up by themselves, and saved time does not. An hour that goes unfilled this week is not sitting in a savings account next week. Besides, the technology is no longer the bottleneck. TNO calls the applications it studied relatively cheap and quick to develop, and KPMG sees the problems shifting from technical to organisational. What is missing is a decision, and nobody makes that one automatically.

There is also a milder version of the counterargument: lower work pressure is valuable too. That is true, and TNO points it out itself. In a tight labour market, a team that works less frantically can be worth more than higher revenue per head. But then make it a deliberate choice, and do not count on a productivity gain that will never come that way.

What you choose now

If you are about to buy yet another AI subscription for the whole team, first pick one recurring piece of work with a clear beginning and end. Think of incoming quote requests, invoices that need to be booked, or customer questions that are the same every week. Settle three things in advance:

If you cannot answer the second question, the best you are buying is some calm on the shop floor. That is fine. Just do not expect to see it in your numbers.

Sources

Is there work in your business that comes back every week? That is what I build software for.

Book a call →