Hybrid cut quits by a third. Your policy says something else.

The policy says three days in the office. The engineering site runs two. The finance site runs four, because the director does. The resignation rate differs between them, and nobody has asked whether the policy, the practice or the director explains it.

The trial

In 2021 and 2022, a travel company randomised 1,612 employees by birthday into two groups: five days in the office, or Wednesday and Friday at home. Over the following months, quits in the hybrid group were 4.80% against 7.20% in the office group, a fall of one third. The fall was 54% for women and 52% for people with commutes over 90 minutes. Performance reviews, promotions over two years and, for 653 engineers, lines of code written showed no difference, and the statistical tests were designed to detect one. The 395 managers' belief about productivity moved from minus 2.6% before the trial to plus 1.0% after it.

It is one company in one country, and the authors say so. It is also the only randomised evidence of this size, and its result is in the direction that every honest observational study has pointed since.

The policy is not the practice, again

The written hybrid policy is a document. The practice is what badge records, calendar density and absence patterns say, and it is set by managers and sites far more than by the policy. Where the two differ, a quit-rate comparison between sites is a comparison between practices, and the policy is irrelevant to it. This is the gap we see most often when writing rules with HR teams: the rule they state is the document's, and the rule the organisation runs on is the director's.

Measuring your own number

1. Classify people by actual attendance pattern over a quarter, from badge or scheduling data, in bands: mostly office, hybrid, mostly remote. Aggregate only, above a minimum group size.
2. Compare resignations over the following twelve months by band, adjusted for site, tenure, job family and commute where you hold it.
3. Split by sex and by commute length. The trial says that is where the effect concentrates.
4. Report the gap between policy and practice by site and manager, as a fact for the leadership team, not as a compliance finding.

What a Coworker does here, and does not

Attendance data is monitoring data, and in Germany a system able to evaluate it needs the works council's agreement before it is used. The analysis above belongs in a scheduled brief, in aggregate, with the purpose written into the works agreement and the output restricted to the roles that own the policy. It does not belong in a manager's dashboard, and it does not produce anything about an individual. Read that way, the record answers a question the policy document cannot: what the organisation's own practice is costing, and for whom.