Hybrid is a retention instrument, not a compromise

Most hybrid schedules are settlements. Someone wanted five days in the office, someone else wanted none, and three became the answer because it sits between them. That is a negotiation outcome, not a design.

There is one study that lets you check what a designed schedule actually buys. Nicholas Bloom, Ruobing Han and James Liang ran a randomised controlled trial at Trip.com, a multinational travel company, and published it in Nature in 2024. Sixteen hundred and twelve engineers, marketers and finance staff were assigned by birthday — even or odd — either to the standard five days on site or to a hybrid schedule working from home on Wednesdays and Fridays. Randomisation by birthday matters: it removes the selection effect that makes almost every other comparison of remote and office workers uninterpretable.

Over six months, quit rates in the control group ran at 7.2 percent. In the hybrid group they were 4.8 percent. Attrition fell by roughly a third. The effect was largest exactly where a retention problem usually lives: 40 percent lower among non-managers, 52 percent lower among employees with long commutes, 54 percent lower among women.

The second finding decides the argument. Two years of subsequent performance reviews showed no difference between the groups, and no difference in promotion rates. The company got the retention without paying for it in output or in advancement.

We would not extrapolate this to every organisation, and neither did the authors. It is one firm, in one country, in one period, with two fixed anchor days rather than employee-chosen ones. The fixed days are probably load-bearing — a schedule where everyone is present on the same days is a different intervention from one where everyone picks.

But it is enough to retire a specific claim: that hybrid is a concession management makes and absorbs as a cost. In the best-identified study available, hybrid was a 2.4 percentage point reduction in attrition for no measurable performance penalty. Replacing an employee costs considerably more than that.

The design question is therefore not how many days. It is which days, fixed for whom, and what work is scheduled into the room when people are in it. A schedule that answers those three has a mechanism that could produce this result. A schedule that splits the difference between two preferences has no mechanism at all.