Return-to-office commands have met their match in the data. The cleanest experiments we have show that when people work part of the week from home, they quit less and do not perform worse. The harder question, and the one most firms avoid, is how to make that flexibility reliable across time zones and calendar clashes. Hybrid can lower attrition, but only an asynchronous operating system keeps work flowing when no one shares a desk or a clock.
What the best experiments actually say
In 2021 and 2022, Trip.com ran a randomized controlled trial with 1,612 university-trained employees in Shanghai. People were assigned by birthday parity to two models for six months: five days in the office or three in the office and two at home. The hybrid group resigned at a rate of 4.8 percent versus 7.2 percent in the control, a one-third drop, and their performance reviews and promotion rates were statistically indistinguishable over two years (Bloom, Han and Liang, 2024). This is gold-standard causal evidence inside a real company. It tests the hybrid model many firms now argue about, and it finds the feared performance penalty did not materialize.
There is history here. A decade earlier, a randomized experiment inside Ctrip, the Chinese travel company later renamed Trip.com, assigned call-center volunteers to work from home or stay in the office for nine months. Working from home raised performance by 13 percent, much of it from more minutes worked and fewer sick days, and it halved attrition. There was a tradeoff that leaders still underestimate. Conditional on the same performance, promotion odds fell for the home group during the experiment. When Ctrip later let employees choose their arrangement, the measured productivity gains almost doubled to 22 percent, an effect the authors attribute to selection and learning (Bloom, Liang, Roberts and Ying, 2015). The lesson travels: flexibility pays, but design matters, and visibility biases can creep in unless managers are explicit about how advancement works.
Trip.com’s managers also learned that implementation quality determines whether hybrid helps or hurts. In their managerial write-up, the authors identify three conditions that supported the positive results: rigorous performance management, coordinated hybrid schedules at the team or company level, and visible backing from top leadership (Bloom, Han and Liang, 2024 HBR). They also report that lower turnover saved the firm millions of dollars in recruiting and training, an internal estimate rather than an audited figure, but the retention math is straightforward when knowledge workers are expensive to replace (Bloom, Han and Liang, 2024 HBR).
Async is the operating system that makes hybrid scale
Hybrid without a plan for how work moves is just a quieter office policy. The necessary upgrade is asynchronous work: organizing around written decisions, clear ownership, and outputs that move when colleagues are offline. Practitioners who have built large distributed organizations argue that async is not a communication preference, it is a management choice. Matt Mullenweg, the CEO of Automattic, describes a maturity path where level four is defined by asynchronous work evaluated by output rather than hours, and level five, fully distributed and output-led, is an aspiration rather than an evidence-backed claim (Mullenweg, 2020). Take the conceptual flourish with the right caution. The value here is the framing: async is a standard for how work is measured and handed off, not a tool setting in Slack.
The best documented case study of this practice remains GitLab’s public handbook. The company states that asynchronous communication is impossible without strong documentation because a single missing answer can block a colleague for hours. GitLab’s handbook, which would exceed 2,000 printed pages, is designed as the operational foundation of async work so that anyone can self-serve decisions, processes, and norms rather than wait for a meeting (GitLab handbook, 2026). This is not causal evidence of outcomes. It is a clear statement of mechanism from a scaled practitioner: documentation is the infrastructure that lets hybrid and globally distributed teams move without constant scheduling.
The managerial translation
What should leaders do with this mix of experiments and practice? First, anchor the hybrid debate in the retention and performance evidence. In the Trip.com experiment, two days at home cut quits from 7.2 to 4.8 percent with no loss in evaluated performance or promotions over two years (Bloom, Han and Liang, 2024). That is the baseline business case. Second, adopt the implementation guardrails that the authors emphasize: clarify performance metrics and review cycles, coordinate days in and out at the team level so collaboration is predictable, and signal top-level support so managers do not quietly sabotage the model (Bloom, Han and Liang, 2024 HBR).
Third, treat async as a system, not a slogan. That means shifting from meetings by default to documents by default. Define the artifacts that constitute work in your context: design briefs, pull requests, decision logs, customer notes. Set response-time expectations by channel and time zone, so urgency is rare and clear. Write down how promotions, visibility, and sponsorship actually happen, and audit the data quarterly for any penalty associated with location. The Ctrip trial’s promotion gap for full-time home workers was not destiny, it was a product of how managers saw and scored work (Bloom, Liang, Roberts and Ying, 2015). Closing that gap requires documented criteria that travel in writing.
Finally, copy the spirit of experimentation. The Trip.com team literally A/B tested hybrid. Most firms can run smaller within-unit pilots with clear metrics and pre-registered success criteria, then scale what works. The alternative is a slow drift back to presence-based management dressed up as collaboration. That model raises costs without proof that it raises performance.
The practical read is simple. Hybrid, done with discipline, is a retention engine. Async, built on documentation and output-based management, is how you make that engine reliable across distance and time. Start measuring managers on the clarity of goals and the quality of written decisions, not on the headcount in their meeting invites. If you do that, the debate about who sits where three days a week becomes less important than the fact that the work ships when it should.
Sources
- Bloom, Han and Liang (2024). Hybrid working from home improves retention without damaging performance, Nature. A randomized trial at Trip.com where two days at home cut quits from 7.2% to 4.8% with no loss in performance or promotions. https://www.nature.com/articles/s41586-024-07500-2
- Bloom, Liang, Roberts and Ying (2015). Does Working from Home Work? Evidence from a Chinese Experiment, Quarterly Journal of Economics. A randomized trial at Ctrip where full-time home work raised performance 13%, halved attrition, and lowered promotion odds conditional on performance, with larger gains after choice was rolled out. https://academic.oup.com/qje/article-abstract/130/1/165/2337855
- Bloom, Han and Liang (2024). One Company A/B Tested Hybrid Work. Here's What They Found, Harvard Business Review. Managerial summary of the Trip.com RCT that highlights three conditions for success and reports an internal estimate of millions in turnover savings. https://hbr.org/2024/10/one-company-a-b-tested-hybrid-work-heres-what-they-found
- GitLab (2026). GitLab — handbook-first & asynchronous work (all-remote handbook). A practitioner case stating that async is impossible without strong documentation and describing a handbook that would exceed 2,000 printed pages. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/
- Mullenweg (2020). Five Levels of Autonomy (distributed work). A practitioner's taxonomy where level four is asynchronous work measured by output, used here as a conceptual frame rather than empirical proof. https://ma.tt/2020/04/five-levels-of-autonomy/