Both camps in the remote-work fight share a hidden assumption. The office loyalists believe location drives performance, so they mandate the office. The remote advocates believe the opposite location drives performance, so they mandate home. The experimental evidence embarrasses both, because its most striking result is not about where people worked. It is about who got to decide. The gains grew when workers chose their own setup, and that changes what the debate is really about.
The number that doubled
The seminal experiment ran at Ctrip, a Chinese travel firm with 16,000 employees. Call-center staff who volunteered were randomly assigned to work from home or in the office for nine months, and the home group came out about 13 percent more productive, most of it from working more minutes per shift and the rest from a quieter room (Bloom, Liang, Roberts and Ying, 2015). That result alone is usually where the story stops. It should not. After the trial, Ctrip let the whole relevant workforce choose their arrangement and allowed people to switch. When they did, the measured productivity gain roughly doubled, to around 22 percent (Bloom, Liang, Roberts and Ying, 2015).
That jump is the most important and least quoted finding in the literature. The extra gain did not come from more people working home. It came from people sorting themselves into the arrangement that actually fit them. Those who thrived at home stayed home. Those who did not returned to the office. The authors attribute the improvement to selection and learning, the workforce discovering and choosing what worked for each person. The mandate captured half the value. Choice captured the rest.
Why autonomy is the active ingredient
This reframes location as a container for something else. What the second phase added was not a place but a decision right, and decision rights over one's own work are the definition of autonomy. The practitioner literature has been insisting on this for years. Matt Mullenweg, who runs the distributed firm Automattic, says his organization was designed around intrinsic motivation, the autonomy, mastery and purpose framework popularized by Daniel Pink, and treats the granting of autonomy as the engine of distributed performance (Mullenweg, 2020). That is a founder's stated philosophy rather than a controlled test, and his stronger claim that a distributed firm beats any in-person one carries no evidence. But the Ctrip choice result gives the autonomy thesis something the opinion alone lacks, which is an experimental fingerprint. Value appeared precisely when control was handed to the worker.
The later Trip.com hybrid trial is consistent with the same reading. It found that two days at home cut resignations by about a third with no loss in performance (Bloom, Han and Liang, 2024). Retention is what you would expect autonomy to move first. People do not quit an arrangement they helped choose at the same rate they quit one imposed on them. The pattern across both studies points the same way. Impose a location and you get part of the benefit. Grant a choice and you get more.
The cost of mandating either direction
The practical consequence is uncomfortable for confident leaders on both sides. A blanket return-to-office order forfeits the choice dividend by definition, forcing office-suited and home-suited workers into the same box. But a blanket work-from-home order does the same thing in reverse, and the Ctrip data suggests it too would leave the doubling on the table. The value was never in the destination. It was in the matching, and matching requires that the person doing the work has a say.
The caveats are real and worth stating. Both experiments ran inside single Chinese firms, one in call centers and one among graduate professionals, so the exact percentages are not universal constants. And the choice phase at Ctrip mixed selection with learning, so no single mechanism gets full credit. What survives every caveat is the direction of the effect. Handing people control over their arrangement raised the measured payoff rather than lowering it.
The practical read
A firm chasing the productivity numbers from the headlines is often chasing the wrong half of them. The larger figure came from autonomy, and autonomy cannot be mandated into existence, only granted. The design implication is straightforward. Set the guardrails the evidence requires, a real measure of output and coordinated schedules, and then let people choose within them. The leader who insists on picking everyone's location, in either direction, is not maximizing performance. They are collecting the smaller number and calling it the whole result.
There is a quieter benefit to this that boards tend to miss. An arrangement people chose is one they are less inclined to resent and less inclined to leave, which is why the hybrid trial's clearest effect was on retention rather than raw output (Bloom, Han and Liang, 2024). Autonomy does not only raise the productivity ceiling. It lowers the exit rate, and a firm that grants it collects both. The instinct to control location feels like rigor. Against this evidence it looks more like leaving money on the table in the name of looking decisive.
Sources
- Bloom, Liang, Roberts and Ying (2015). Does Working from Home Work? Evidence from a Chinese Experiment (QJE 130). Productivity gains roughly doubled to 22% once workers could choose and switch arrangements, the central evidence that choice, not location, drove the payoff. https://academic.oup.com/qje/article-abstract/130/1/165/2337855
- Mullenweg (2020). Five Levels of Autonomy. States Automattic was designed around intrinsic motivation and autonomy, the philosophy the choice result gives an experimental fingerprint. https://ma.tt/2020/04/five-levels-of-autonomy/
- Bloom, Han and Liang (2024). Hybrid working from home improves retention without damaging performance (Nature 630). The one-third drop in resignations under a chosen-feeling hybrid arrangement is consistent with autonomy driving retention. https://www.nature.com/articles/s41586-024-07500-2