Posted by 
A retail chain wanted its apprentices to show up more reliably, so it paid them a bonus for attendance. Absences went up by about half, which came to more than five additional days per person per year, and they stayed up after the bonus ended. Jakob Alfitian, Dirk Sliwka and Timo Vogelsang ran it across 232 stores and 346 apprentices and published it in Management Science as “When Bonuses Backfire”.
The design contains its own control. A second variant paid in time off instead of money and produced no such effect. The backfiring was driven by the most recently hired apprentices — the people who had no view yet of what counts as normal here, and took it from the rule.
Antoinette Weibel, who researches control and trust in organisations, puts this case in the critical field of her framework, for a point that is easy to misread. The damage did not come from the fact that attendance was measured. It came from how the rule was built. The bonus measured from outside, attached itself to money, explained nothing, and let nobody it applied to shape it. The organisation put nothing of its own at risk; only the apprentices were tested. What it installed was a norm, and the surveys caught it in the act: absence had become acceptable behaviour.
Why This Is Suddenly Urgent
The experiment ran in 2018 and had nothing to do with AI. That is what makes it usable now: it isolates what a badly built control does to the people under it, with no technology to argue about instead.
AI is about to put a great deal of new control into a great many organisations, and it arrives from two directions.
Producing something that looks convincing has become nearly free. A market assessment, a supplier comparison, a draft policy: an hour of work, then a prompt. Checking that output and standing behind it costs what it always cost, because it still runs through a person who has to read the thing and sign it. Pressure builds on the checking side, and the obvious answer is to check more.
The same technology has moved control in the other direction. Control used to be expensive, and the expense rationed it. Comparing one person’s output against everyone else’s used to cost a supervisor a morning; it now costs nothing, runs every night, and covers everybody.
So the obvious answer became affordable at the moment it became tempting, and cheap control expands without anyone deciding to expand it.
Four Fields, and One of Them Holds
Weibel’s framework asks two questions, and organisations answer only the first.
Is the control enabling or coercive? The distinction goes back to Paul Adler and Bryan Borys, who showed that formal rules can equip the people they bind or merely constrain them. An enabling rule explains what it is for, and it can be argued with.
Does the organisation work at being trusted, or assume that it is? Passive handling expects trust and does nothing to earn it. Active handling costs the organisation something: it commits to something it can later be held to, in front of the people it is asking to trust it.
Coercive control with passive trust management is the critical field, and the attendance bonus sits squarely in it. Enabling control with active trust management is the field that holds under pressure. Less control is no answer either: a rule that binds nobody equips nobody.
Trust Is a Line in the Same Budget
An organisation with high system trust can work with samples. It can say: we examine a fifth of these, here is which fifth and why, and what an approval covers and what it does not. That is affordable, and it is honest.
An organisation without system trust has to examine everything, and it will not manage. What follows is the dishonest full check — the signature that says everything was looked at when very little was.
System trust is the only resource that lowers the cost of checking without an oracle to check against. Every other lever needs something reliable to measure the work against, and for most decisions worth making there is no such thing.
What Carries the Risk
Three things separate a rule that equips from one that only binds.
Rules state their purpose, and the people bound by them help write them. A rule nobody is allowed to argue with produces compliance and nothing else.
The organisation carries a visible share of the risk. A named sampling quota is an organisation saying out loud that it accepts what a sample will miss. It costs something to say, which is why it is believed.
And what gets measured stays deliberately narrow. Cheap measurement spreads into everything it can reach, and an organisation gets more of whatever it watches and less of everything else.
Where the Ingenuity Lives
Ingenuity happens in the space where somebody is trusted to use judgment before anyone has asked for a result. That space is the least legible part of the work, which makes it look like the easiest place to tighten up, and it is what cheap control occupies first.
An organisation can tighten it, and the apprentices in that retail chain show what comes back when it does.
The Norm Outlives the Instrument
The bonus ended and the absences stayed. Norms do not resign, and the people who learn them fastest are the ones who arrived most recently.
Whatever an organisation installs this year in answer to the flood of machine-made output will still be running when the flood is ordinary. It will have taught people what the organisation believes about them long after anyone remembers which quarter the tool arrived in. That is the decision in front of executives now: what a piece of cheap control teaches, and whether anybody chose to teach it.
About the author:
Guido Bosbach is a management consultant and author based in Bonn, Germany. He works with executive teams on decision quality, organizational clarity, and the cultural conditions for sustainable performance.