A talk at the American-German Business Club (AGBC) Munich · 28 October 2026
Where did the time go?
AI saves people time, but that time rarely turns into extra sales. It goes into people learning new ways to work, which no report measures. Unless someone gives the saved time a job and an owner, it disappears.
This page sets out, in six pictures, the argument of a ten-minute talk to business owners and leaders. It assumes no background in AI. The slides and all 16 sources are at the bottom.
The promise, tested
The AI promise did not survive one real company’s accounts.
The case for AI usually runs like this. It makes people faster. Faster means more work done. More work means more revenue. Surveys back the first step: people using AI say they save 7.5 hours a week, about one working day [1]. A forecast by the bank Goldman Sachs says AI could add 7% to global GDP, the value of all the world’s output, within ten years [2].
Neither claims that any single firm will sell more. That step seemed a sensible assumption, so we tested it on a UK firm of more than 1,500 people, listed on the stock exchange, whose accounts anyone can read [3]. We took its most generous form: AI makes every working hour four times as productive, staff get a third of each day back, and all the extra work sells at today’s prices. Two-thirds of a day at four times the output is more than two and a half times today’s work. On those assumptions, its extra sales would be more than everything it sold that year.
Our own arithmetic on the firm’s accounts, at our most generous assumptions; not a forecast, and not a measure of the firm’s own AI use [3]
Revenue growth in its latest year (to 30 April 2026), from audited accounts [4]
Surveys reported the time saved. The extra sales never came. So where did the time go?
Speed only becomes money when someone checks the work is good enough and sells it. That is a separate job, and the pitch leaves it out.
What the research measured
Feeling faster is not the same as being faster.
When researchers measured the work, instead of asking people how it felt, three findings stood out.
A working paper is a study published before other researchers have checked it. Peer review is that check.
Experienced software developers, working on real tasks, took 19% longer with AI. They still believed AI had made them faster [6].
Small study, early 2025A software firm gave its customer support staff an AI assistant, trained on past support chats with extra weight on its best staff. In their first month, the newest staff solved 46% more problems an hour. The most experienced gained least: the tool was largely showing them their own best practice [9][10].
46% from the working paper; the pattern is peer-reviewedAcross more than 12,000 firms in Europe and the US, productivity gains from AI showed up only in firms with 50 or more staff. Below 50, no gain could be measured yet [11].
Not yet peer-reviewedWhy the gain gets stuck
Saved time has to get past three people before it becomes profit.
In a large survey, 80% of respondents said AI had improved individual productivity. Only 37% could point to any effect on their organisation’s profit [7].
A Bain & Company brief explains the gap through three people at different levels of a business [8]. In Bain’s words: “The CEO sees the destination. The general manager sees the obstacles. The individual sees a tool that works for some people but not yet for them.”
In plain terms, the leader has set the goal. The manager must get the budget and approvals to reach it. Most employees have yet to find out how the tool helps them. Our reading: if any one of the three does not act, the saved time stops there. In a small firm, all three are often the same person: the owner.
Who AI helps
AI helps most when you can already judge its answer.
Small business owners in Kenya were given an AI business adviser. On average, it made no difference. The strongest businesses did better and the weakest did worse. What differed was which advice they acted on [12].
Peer-reviewedConsultants given a task just beyond what AI does well were 19% less likely to get it right with AI than without it [13].
Peer-reviewedOur reading: AI sounds just as sure when it is wrong as when it is right. If you cannot tell a good answer from a bad one, you cannot push back.
How people use AI
Three people can use AI just as much and learn very different things.
Harvard researchers read the AI chat logs of 244 consultants at a leading firm and found three ways of working. They named them Centaur, Cyborg and Self-Automator [14].
An AI usage report counts how much each person uses AI, so all three look the same. Flip each card to see what the study found they were learning.
Centaur
Splits the job. Decides the approach, gives AI clearly defined pieces, and checks what comes back.
Cyborg
Works with AI in one continuous conversation: drafting, challenging, arguing back.
Self-Automator
Hands the whole job to AI, and passes on whatever comes back.
One isn’t learning at all.
In the study, 14% worked as Centaurs and 27% as Self-Automators [14]. Working paper, not yet peer-reviewed. The skill lines are illustrative: the study describes them in words, not numbers.
What to do about it
Three things a business can do so the saved time is not lost.
These are our suggestions, drawn from the research above. Tap or click each one for a question to ask your own team.
A question for your team
Where are we using AI on work nobody here can check?
A question for your team
Who is using AI to learn, and who is using it to skip the thinking?
A question for your team
What is the time we save for, and who owns it?
The answer
The real AI dividend is time, not revenue.
So where did the time go? Into people, learning new ways to work that nobody planned and no report can see.
For now, the payoff from AI is not extra revenue but time, which you can invest in your best people and in rethinking how the business works.
This is no longer theory: in October 2026, a Goldman Sachs executive said its newest hires now manage AI agents, software that carries out tasks on its own, from their first day [16].
Whoever runs a business or a team has to give that time a job and an owner, and decide who it benefits. They also decide whether their people become Centaurs or Self-Automators.
Take it with you
The slides, the sources, and where to go next.
All 16 sources, each checked against the original
Titles and pinpoints are given as published, in English.
- [1]
- [2]
- [3]
- [4]
- [5]
- [6]
- [7]
- [8]
- [9]
- [10]
- [11]
- [12]
- [13]
- [14]
- [15]
- [16]