Return-to-office memos keep coming, and so do the defections. The argument is framed as a choice between presence and flexibility, yet the evidence points somewhere else. Where hybrid works, it is not because people sit in different places. It is because management learns to run the work across time.
What the cleanest experiments say
A randomized trial inside Trip.com, published in Nature, assigned more than 1,600 university-trained employees to two days a week at home or five in the office. Attrition fell by about a third for the hybrid group, from 7.2 percent to 4.8 percent, with no detectable difference in performance reviews or promotion rates over two years (Bloom, Han and Liang, 2024). This is causal evidence inside a real firm, not a convenience survey. It is also bounded. The sample is a single technology company in China, and the treatment was structured hybrid rather than fully remote.
The classic counterpoint comes from a prior experiment at Ctrip, Trip.com’s predecessor. There, call center employees who volunteered for full-time home working were randomly assigned to home or office for nine months. Working from home raised performance by 13 percent, largely from more minutes per shift and a quieter environment. Job satisfaction rose and attrition halved. Yet conditional on performance, promotion rates fell for home workers (Bloom, Liang, Roberts and Ying, 2015). When the company later let everyone choose their setup, and people re-sorted, the performance gains roughly doubled to 22 percent. The lesson is not that remote is a magic switch. It is that design and choice matter, and visibility biases are real unless you counter them.
In a managerial translation of the Trip.com trial, the authors argue that three conditions explain why hybrid worked: rigorous performance management, coordinated team schedules, and visible support from the top. They also recommend that firms run their own A/B tests to calibrate policies to local context (Bloom, Han and Liang, 2024). Treat hybrid as an operating model with constraints, not a benefit to be rationed.
Why async is the mechanism, not the vibe
Hybrid fails when it shrinks the office without changing how work flows. The result is calendar sprawl and delayed decisions. The shift that makes hybrid durable is asynchronous work. Async is not a Slack timestamp. It is a system that lets work advance when colleagues are not simultaneously online, which forces clarity about outputs, service levels, and handoffs.
Practitioners who run large distributed operations have converged on the same mechanism. GitLab describes its operating model as “handbook-first,” with a public manual that would exceed 2,000 printed pages. The company asserts that async is impossible without strong documentation because a single follow-up question can block a colleague for hours. The handbook exists so teammates can self-serve answers instead of waiting for someone to come online (GitLab, 2026). This is not an outcome study. It is a playbook from a firm that chose to make writing, templates, and shared artifacts the backbone of coordination.
The causal evidence from Trip.com fits this mechanism. In the Nature trial there was no penalty to performance or promotion when hybrid was run with clear schedules and performance management (Bloom, Han and Liang, 2024). In their Harvard Business Review article, the authors are explicit about the conditions: define and measure output carefully, coordinate days to protect collaboration, and secure top-leadership backing so the model is not undermined by side deals or rumors (Bloom, Han and Liang, 2024). Async practices are how those conditions show up day to day. If you cannot watch inputs, you must agree on outputs and write down how to get there.
Matt Mullenweg, who leads a fully distributed company, frames this as a maturity curve. His “five levels of autonomy” describe Level 4 as asynchronous work evaluated by output rather than hours or process. It is a practitioner’s taxonomy, not a causal claim, and it helps leaders name what they are aiming to build (Mullenweg, 2020). The label matters less than the shift in managerial attention, from attendance and meetings to artifacts and delivery.
Guardrails against career penalties
The Ctrip experiment carried a warning. Even with higher measured performance, full-time home workers were promoted less often than equally performing peers in the office (Bloom, Liang, Roberts and Ying, 2015). Hybrid can dodge that trap, as the Trip.com RCT found no promotion gap over two years, but only when management is explicit about criteria and coordination (Bloom, Han and Liang, 2024). The implication is simple. If you move time and place, you must also move how you recognize contribution.
That means tightening performance management to focus on shipped work and customer outcomes. It means agreeing on team-level office days to protect high-value collaboration while keeping two days at home for focus, as the Trip.com experiment did (Bloom, Han and Liang, 2024). It also means documenting the evidence that feeds promotion decisions so visibility does not default to who was in the room. None of this requires software heroics. It requires writing, cadence, and clear decision rights.
How to start the shift
Leaders should borrow both the method and the mechanism. Method first. Run an experiment. Randomize team members to hybrid schedules for a defined period. Track attrition, output, customer metrics, and promotion outcomes. The Trip.com authors argue that the turnover reduction alone saved the company millions of dollars in recruiting and training, an internal estimate that speaks the board’s language (Bloom, Han and Liang, 2024).
Now the mechanism. Declare “async by default” for any work that does not require live interaction. Set response time norms for different channels so teams do not rebuild meetings in chat. Make documentation a production activity with owners and version control, not an afterthought. GitLab’s claim that async fails without strong documentation is a useful test: if the answer to a basic how-do-we-do-this is trapped in someone’s head or calendar, you are not ready (GitLab, 2026). Finally, expose promotion and performance criteria and review them for proximity bias. The 2015 Ctrip finding shows what happens when you do not (Bloom, Liang, Roberts and Ying, 2015).
The strategic read is clear. The best experimental evidence says that well-run hybrid retains talent without sacrificing performance. The operating lesson says that “well-run” means async at the core, with output measured, schedules coordinated, and the work written down. Firms that stop the conversation at location will get a quieter office and a louder resignation line. Firms that redesign how time works will get what the evidence already shows: more staying power with no performance trade-off.
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 by about one third with no loss in performance or promotions, grounding the case for structured hybrid. 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 where full-time home work lifted performance 13 percent and halved attrition but reduced promotion odds, a caution on visibility bias. 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. The authors identify three implementation conditions and report Trip.com’s internal estimate of millions saved from lower turnover, offering practical guidance for leaders. 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 asserting that async fails without strong documentation and describing a handbook-first system as the operational backbone. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/
- Mullenweg (2020). Five Levels of Autonomy (distributed work). A practitioner’s taxonomy where Level 4 defines async work judged by output, a framing device for leaders planning the shift. https://ma.tt/2020/04/five-levels-of-autonomy/