Return-to-office mandates aren't a headcount policy. They're a seniority filter.

The return-to-office mandate arithmetic looks fine until you check who's actually leaving

A return-to-office mandate reads, on paper, like a policy that treats everyone the same: badge in three or four days a week, or find another job. In practice, the departures it produces are not evenly distributed. The people who leave first are disproportionately the engineers who have the easiest time finding another offer, which correlates closely with seniority, recent track record, and how in demand their specific skills happen to be the week the mandate lands.

That correlation is the mechanism worth examining, not "people don't like commuting." Commuting dislike is close to universal. What separates the engineer who grumbles and complies from the one who updates their resume that week is whether leaving costs them anything. A staff engineer three years into vesting, with a strong recent shipped project and a few recruiters already in their inbox, is making a very different calculation than a two-year engineer still building a track record.

Seniority is a proxy for options, and options are exactly what RTO tests

Every RTO rollout is, whether framed this way internally or not, a test of who has outside options and who is willing to use them. Multiple workforce surveys from 2025 and 2026 point the same direction: departures accelerate after a mandate, and the increase skews toward more experienced and more skilled employees rather than spreading proportionally across the workforce.

A Forbes piece published this month put a name to what a lot of engineering leaders have been seeing anecdotally: RTO mandates function less like a return to a prior norm and more like a filter against whoever has the market power to say no.

For an engineering organisation specifically, that's a worse trade than it looks on a headcount spreadsheet. Junior and mid-level engineers are, on average, easier to backfill than the senior engineer who holds context on the one reconciliation job nobody else has touched in two years. Losing five generalists is an inconvenience. Losing the person who understands why a specific retry loop exists is an incident waiting to happen.

The bus-factor math nobody puts in the RTO memo

Bus factor, the number of people who'd need to leave before a project stalls, tends to concentrate in a handful of senior engineers on any team past a certain size. That's not a flaw in how teams are built; deep system knowledge takes years to accumulate and doesn't distribute evenly by design. It does mean an attrition policy that's indifferent to seniority is, in practice, a policy that raises the odds of losing a bus-factor-of-one person from exactly the group most likely to have somewhere else to go.

None of this shows up in the announcement memo, which typically talks about collaboration and culture. It shows up two or three quarters later, when a team tries to backfill someone whose actual job was never fully written down, because nobody expected to need the write-down.

The case for RTO isn't nothing, and it deserves to be stated

None of the attrition data makes the argument for in-person work disappear. Mentorship compresses badly over video: a junior engineer picking up a system by pair-programming next to someone, or by overhearing a hallway conversation about why a service is built the way it is, learns things a scheduled one-on-one rarely surfaces. Whiteboard debugging, the kind that happens when three engineers cluster around a screen because a deploy just broke something, is often genuinely faster in a room than split across separate video calls.

The problem isn't that this case is wrong. It's that a blanket mandate is a blunt instrument for capturing a benefit that's concentrated in specific moments, like onboarding, incident response, and project kickoffs, rather than in five days a week of ambient presence. An org that actually wants the mentorship and debugging benefit can design for those moments specifically. That's a different policy than requiring badge-ins on a fixed schedule regardless of what a given week actually calls for.

What the four-day-week data says about the alternative nobody's trying

If the goal behind RTO is genuinely about output and cohesion rather than office-lease utilisation, it's worth putting it next to the other workplace policy that's had a comparably large, comparably rigorous trial over the same period: the four-day week. The largest controlled study to date, run by Boston College researchers and published in Nature Human Behaviour, tracked just under 2,900 employees across 141 companies in six countries through a six-month trial with no pay reduction.

The headline results were about wellbeing more than raw satisfaction: burnout fell on a standard five-point scale, job satisfaction rose on a ten-point scale, and both mental and physical health measures improved. A twelve-month follow-up found the gains held rather than fading, which answers the usual "novelty effect" objection more directly than most workplace policy research manages.

The mechanism the researchers point to isn't simply "one extra day off." Preparing for a shorter week forced companies to strip out low-value meetings and hand employees more control over how work got done, and that restructuring is what carried through into the quality of the work itself. That's a different kind of lever than RTO, which asks employees to change where they work without asking the organisation to change how it works at all.

The levers that don't carry the same seniority skew

None of this is an argument that in-person work is worthless. It's an argument for being precise about which policy lever does which job, because not every lever costs a team its most senior people first.

The pattern across that table isn't "remote work is right and RTO is wrong." It's that policies which change compensation or workload evenly, or that leave the choice with the employee, don't produce the same lopsided exit pattern a blanket in-office mandate does. A mandate is the one lever on this list that specifically taxes the people with the most alternatives.

What to actually check before mandating anything

Before an engineering org rolls out an RTO mandate, three numbers are worth pulling that the initial announcement usually skips: the tenure and level distribution of the team relative to who's likely to leave first, the bus-factor map for anything the team would classify as a critical dependency, and the real cost of a senior backfill including ramp time, not just the recruiting fee. Run against those three, "badge in Tuesday through Thursday" often reads differently than it did in the slide that proposed it.

Model it against a mid-size team and the trade gets concrete. A twenty-engineer org that loses three senior ICs to an RTO mandate isn't just down three headcount; it's down whatever accumulated context those three held, multiplied by however many systems each of them owned outright, and a new hire typically needs the better part of a year to rebuild that context from nothing. That number is worth putting next to whatever the mandate was meant to save on office utilisation before calling the trade worth it.

The honest version of an RTO decision names the trade explicitly: some in-person collaboration benefit, purchased at the price of a disproportionate chance of losing the people whose absence costs the most to replace. Whether that trade is worth making depends entirely on the org. Skipping the arithmetic first is the actual mistake, not the choice of office days.