Works councils and robots: the evidence the pilot needs.

The HR director is told that the works council will slow the AI pilot by six months. Two large studies of the last decade's automation wave say the opposite: where workers had a formal voice, the technology arrived on the same schedule, more of the existing workforce came through it, and the plant was more productive afterwards.

Robots in German plants

Findeisen, Dauth and Schlenker used linked employer-employee data to compare German plants adopting industrial robots with and without a works council. Councils raised the retention of incumbent workers by about four percentage points during adoption. Older workers in routine manual jobs earned about 3.5% more. The share of workers in training, already about five points higher in council plants, widened to about fifteen points. Councils did not slow robot adoption, and council plants showed higher productivity growth after adopting. The study is a working paper and should be read as such; its direction agrees with the peer-reviewed literature.

Voice at work in Finland

Harju, Jäger and Schoefer, in the American Economic Journal: Applied Economics, studied Finland's 1991 law that gave workers representation in firms with 150 or more employees. Labour productivity rose. Voluntary quits did not change. In recessions, firms with representation made fewer involuntary separations. Job-quality gains were small at most. The picture is of an institution that improves how change is managed rather than one that resists it.

What this means for an AI pilot

The parallel is not exact; a Coworker that reads records is not a robot on a line. But the mechanism the studies describe is the same one the law now expects. The EU AI Act requires employers to inform workers' representatives before a high-risk system is used at work, from 2 December 2027. Germany's Works Constitution Act gives the council a say over any system able to monitor behaviour or performance, and the Federal Labour Court reads that broadly. A rollout that starts with the agreement, rather than ending with it, is both the lawful path and, on this evidence, the productive one.

1. Bring the draft agreement to the first meeting: purpose, what is logged, who reads the log, retention, what the system will not be used for, a review date.
2. Offer the training commitment early. The German result is as much about training as about retention, and it is the part the council can hold you to.
3. Agree the aggregate-only outputs in writing. Team-level briefs above a minimum size, no individual rankings, names only with HR business partners.
4. Set the review date before the pilot ends, so the agreement can be adjusted to what the pilot showed rather than renegotiated around a tool already in use.

In our own deployments, logging, retention and access to the audit trail are set per client so that they can follow the agreement. The studies above are why we think that is the right order, not just the required one.