The Return-to-Office Reversal: What the Data Actually Shows

Big companies have spent four years ordering workers back to their desks. Here is what the attrition, productivity, and hiring data actually show — and why the calculation looks different for a 15-person startup.

In September 2024, Amazon CEO Andy Jassy told employees they would need to be in the office five days a week starting January 2, 2025, ending the three-day hybrid policy the company had defended since 2023. JPMorgan Chase followed with a firm-wide five-day mandate for all staff by March 2025. Dell, AT&T, and the U.S. federal government made similar moves within the same eighteen months. If you run a twelve-person startup or a forty-person agency, you have probably had a version of this conversation already: should we do that too?

The honest answer requires looking at what actually happened when large companies enforced these mandates — not what executives said would happen — and then asking whether any of it applies to a company your size. The evidence, drawn from company disclosures, a National Labour Relations Board ruling, and peer-reviewed research, does not support a simple story in either direction.

What's actually happened

The push to end remote work started earlier than most people remember. In April and May 2021 — before most Americans were vaccinated — JPMorgan's Jamie Dimon and Goldman Sachs' David Solomon both said publicly that working from home was, in Solomon's words, "an aberration that we are going to correct as quickly as possible." Goldman required most employees back by June 2021. Tesla and Twitter followed a different route: after Elon Musk took over Twitter in October 2022, he ended remote work outright; months earlier, a leaked Tesla memo had told staff to log a minimum of 40 hours a week in an office or resign. Apple settled on a middle position in September 2022, requiring three office days a week over the objections of a group of employees who organised as "AppleToo."

The bigger wave came in 2023 and 2024, and it hit companies of very different sizes and cultures. Disney's Bob Iger announced a four-day-a-week, Monday-through-Thursday requirement starting March 2023. Grindr told its mostly remote staff in August 2023 that they had to relocate near an office or leave; the dating app lost roughly 46 per cent of its workforce — 82 of 178 employees — within a month, according to CNN and the Communications Workers of America. Dell told its sales staff to return five days a week starting September 30, 2024, and separately ruled that fully remote employees anywhere in the company would no longer be eligible for promotion, tracking attendance through badge swipes and VPN logs. Amazon's five-day mandate took effect in January 2025, AT&T's followed for staff near company hubs, and in January 2025 a new presidential executive order ordered federal agencies to end remote work arrangements for most employees.

What's worth noticing in that list: the policies are not uniform, and neither is the enforcement. Kastle Systems, which tracks building-badge activity across major U.S. office markets in its Back to Work Barometer, has recorded average weekly occupancy hovering in the low-to-mid 50 per cent range even years into this wave of mandates — meaning a meaningful share of employees at companies with a stated in-office policy still aren't showing up most days. The mandate and the actual behaviour it produces are two different data sets, and only one of them gets a press release.

What the measured outcomes actually show

The clearest large-sample research on this comes from Yuye Ding and Mark Ma at the University of Pittsburgh, who studied 137 S&P 500 companies that adopted return-to-office mandates. Their finding, published in 2023 and expanded in follow-up work: mandates did not improve financial performance — return on assets, profit margins, and stock returns showed no measurable benefit — while employee satisfaction, measured through Glassdoor ratings, dropped, and firms saw an abnormal increase in turnover of roughly 13 to 14 percentage points, along with longer time-to-hire for open roles.

A follow-up study by the same researchers, "Return-to-Office Mandates, Brain Drain and Gender Difference," found something that cuts against the stated logic of most mandates. Companies frame RTO as a way to raise accountability and, implicitly, to push out underperformers. The data show the opposite pattern: turnover among women rose roughly 20 per cent more than among men after a mandate, and mid- and senior-level managers — typically the most experienced, highest-paid, hardest-to-replace employees — left at higher rates than junior staff.

The most rigorous evidence on hybrid work specifically — as distinct from full five-day mandates — comes from a randomised controlled trial run at Trip.com, a Chinese travel company, and published by Nicholas Bloom and coauthors in Nature in June 2024. Across 1,612 employees randomly assigned to either a hybrid schedule (three office days, two remote) or a full in-office schedule, hybrid work cut attrition by 33 per cent overall, with the largest effects among non-managers, women, and employees with long commutes. Performance reviews and promotion rates showed no meaningful difference between the two groups. It's one of the few studies in this space that can claim a causal effect rather than a correlation, because assignment to hybrid or full-office was randomised rather than chosen by employees or managers.

What the research does not settle is whether full five-day mandates improve output or collaboration — the thing companies most often cite as the reason. No company that has implemented one has published performance data showing it worked; the public justifications from Amazon, JPMorgan, and others have been framed around culture and mentorship, not measured productivity. That's not proof the claim is false. It's an honest gap: the strongest causal evidence available favours hybrid over full remote, and separately, favours hybrid over full in-office on retention — but there is no equivalent randomised study isolating the effect of a five-day mandate on output.

It depends which policy you mean

"Return to office" gets used as one phrase for three different policies with three different evidence bases. Treating them as interchangeable is where most of the confusion in this debate comes from.

The one policy type with genuinely strong causal evidence behind it is hybrid, specifically because Bloom's team could run it as a true experiment. Full RTO and remote-first both rely more on observational data — comparisons across companies that chose different policies for different reasons — which makes it harder to separate the effect of the policy from the effect of the company that adopted it. Companies confident enough to go remote-first were often already well-run; companies desperate enough to mandate full RTO after years of drift were often already struggling with something else.

What actually matters at your stage

None of this tells a twelve-person startup what to do. It tells you what to check before copying a policy that Amazon designed for 1.5 million employees and a real estate footprint measured in the tens of millions of square feet.

Start with the talent market you're actually competing in. Fully on-site roles now make up the large majority of U.S. job postings, but remote and hybrid postings still draw a disproportionate share of total applications relative to how few of them exist, and women and younger candidates are more likely to apply to remote roles specifically, according to LinkedIn's talent data. If you're a ten-person company trying to hire a senior engineer or a first ops lead against companies ten times your size, flexibility is one of the few terms you can compete on that doesn't cost you cash.

Look at your own attrition data before borrowing someone else's conclusion. A company with 15 employees doesn't generate a statistically meaningful turnover rate — losing two people in a year is a 13 per cent attrition rate and also could be almost anything. Amazon's data comes from hundreds of thousands of employees, which is exactly why it can support conclusions yours cannot. If you don't have a real productivity or collaboration problem you can point to with specifics — missed deadlines, a project that stalled from lack of coordination, decisions that took too long because the right people weren't in a room — you're reacting to someone else's headline, not your own evidence.

Separate a collaboration problem from a trust problem. The Ding and Ma research on brain drain makes a point worth sitting with: mandates function as a blunt, company-wide tool applied to everyone regardless of individual performance, which is exactly the kind of policy an organisation reaches for when it doesn't trust its management layer to make individual calls. If the real issue is that you don't trust one or two people to be productive without supervision, a five-day mandate for the whole company doesn't fix that — it just makes the people who were never the problem start looking for other jobs, which is precisely the pattern the research shows.

Weigh what you'd actually be trading. A 40-person company doesn't have Amazon's redundancy. If a mandate costs you two senior people — the ones most likely to leave, per the gender and seniority research — you don't have five hundred more candidates in a pipeline to absorb it. The real estate maths is different too: most small companies don't own a half-empty building they're trying to justify, which was a real (if usually unstated) motive behind some large-company mandates.