Last year we introduced a workforce system that offers spare shifts to our drivers automatically. When a route opens up, it ranks the drivers who are available and qualified and sends the offer to the top few by text. The ranking uses hours already worked, distance from the depot and a reliability score built from attendance and accepted offers. Planners used to do this by hand, and the system is faster and, by most measures, fairer.
In the spring a driver at one of our northern depots complained to his union representative that he had received two overtime offers in three months while colleagues received a dozen. Neither his manager nor the planners could say why. It took us a week to establish that a period of sick leave the previous year had lowered his reliability score, and that the score never recovered because he was rarely offered the shifts that would have raised it.
The ranking was not malicious, and in most cases it was reasonable. But a decision that affected how much a man earned had been made entirely by software, and nobody in the company could explain it to him or correct it.
That is a legal problem as well as a human one. UK data protection law requires safeguards when a decision with a significant effect on a person is made solely by automated means: telling people it happens, letting them make representations, and giving them a route to a human who can change the outcome. The Data (Use and Access) Act 2025 widened the cases in which such decisions are allowed, and kept those safeguards.
I expect these decisions to multiply. Customer credit limits, route allocation and the first sift of job applications are already being discussed. Each one arrives as a feature in a product we buy, and the safeguards are rarely in the box.
So every automated decision about a person now has an entry in a register: what it decides, on what data, who owns it and how someone challenges it. The shift system shows each driver why an offer went elsewhere, scores decay after six months, and any driver can ask a named planner to review a ranking.
The next decade will automate many decisions about people. The work of making them explainable and correctable will stay with us.
– Serguey Shinder
