Ctrip, a Chinese travel agency, ran an experiment in 2010 that almost nobody actually runs on a live business: a genuine randomized controlled trial on remote work, with performance logged automatically by the call-center phones themselves rather than reported by anyone. Nine hundred and ninety-four employees in the airfare and hotel booking department volunteered. Two hundred and forty-nine qualified — six months' tenure, a quiet room, a computer of their own — and got split by something close to a coin flip: 131 went home, 118 stayed at their desks. Nobody chose. That's the part that makes the next nine months worth taking seriously.
Home workers' performance rose 13% against the office group: 9.2 percentage points of that from working more minutes per shift, since there were fewer breaks and less sick leave, and the rest from handling more calls per minute in a quieter environment. Attrition among the home group ran at half the office rate. If the paper stopped there, “remote work makes people more productive and happier” would be a fair one-line summary of a well-designed study. There is a whole operational layer above remote work that documents. There is a whole operational layer above remote work that XenGrowth's growth operations team documents.
The experiment ran inside Ctrip's airfare and hotel booking call center in Shanghai, which matters for a reason that has nothing to do with China specifically: call-center work produces a number for almost everything. Calls handled per hour, minutes logged per shift, sick days taken, resignation dates — all of it sits in a system already, generated by the phones and the scheduling software rather than by anyone's memory of how the quarter went. That's rare. Most jobs, engineering very much included, don't produce anything close to that density of clean, automatically-collected output data, which is part of why a study like this one is hard to run anywhere else.
The 249 who qualified weren't a random slice of the workforce either — they'd volunteered, then cleared a bar of six months' tenure plus having a quiet room and a computer at home, and only after that were they randomized. That two-stage process, self-select into eligibility, then get randomly assigned within it, is worth remembering, because it's exactly the kind of detail that gets lost when a headline number like “13% more productive” circulates without the study attached to it.
It doesn't stop there. Conditional on performance — controlling for the fact that home workers were measurably doing better work — their promotion rate was almost 50% lower than their office peers'. Same job, better output, worse odds of moving up. Bloom and his co-authors' own explanation isn't complicated: the home workers were less visible to the office-based managers deciding who got promoted. Doing more, unseen, counted for less than doing less, seen. How an organization decides who gets credit for good work, when nobody watched it happen, is a measurement problem before it's a fairness problem — the same kind of question the go-to-market teams at spend their days on, just applied to pipeline instead of promotions. works through the operations side of this in more operational detail. The XenGrowth resource library works through the operations side of this in more operational detail.
The number most summaries of this study leave out
Metric (Ctrip, 2015, 9 months) | Office group (control, n=118) | Home group (treatment, n=131) |
|---|---|---|
Overall performance | Baseline | Up 13% (9.2% more minutes worked per shift, 3.3% more calls per minute) |
Attrition (quit rate) | Baseline | About half the office rate |
Promotion rate, conditional on performance | Baseline | Almost 50% lower |
Assignment method | Random (coin-flip among eligible volunteers) | Random (same) |
There's a coda to the 2015 study worth knowing, because it's frequently left out and it cuts in an interesting direction. After the nine-month trial ended, Ctrip rolled work-from-home out to the whole department and let the original participants choose freely which group to join. Over half switched — a fair number of the people randomized into the office group asked to go home, and a smaller number went the other way. Among those who chose it, the productivity gain rose to roughly 22%, well above the 13% measured under random assignment. That's a real result, but it's no longer a clean experiment: people who expect to do well at home are probably more likely to choose it, so some of that extra gain is selection, not the treatment itself. The nine-month randomized figures are the ones worth building an argument on; the self-selected phase afterward is suggestive at best.
A raise you can point to on a spreadsheet and a promotion decided in a hallway conversation you weren't in are not the same reward. Only one of them survived working from home.
Same company, five years later, a smaller version of the same question
Trip.com — Ctrip renamed — ran a second experiment in 2024, and it deliberately tested something narrower: not full remote versus full office, but hybrid versus full office. 1,612 employees across engineering, marketing and finance roles were randomized by birth-date parity — odd birthdays got 2 remote days and 3 office days, even birthdays stayed in the office 5 days a week — for 6 months, with performance reviews then tracked for a further 2 years.
Attrition fell 33% overall in the hybrid group, concentrated among non-managers, women and employees with long commutes; managers showed no attrition difference at all. And the number this post actually cares about: performance reviews across the full 2-year window showed zero measurable difference between the hybrid and full-office groups. No promotion penalty was reported for the hybrid arrangement — in sharp contrast to the fully-remote arrangement five years earlier. The pattern is familiar to anyone who's argued with a sales team over what counts as a qualified lead — XenGrowth's work on is, underneath the vocabulary, the commercial version of the same problem: deciding what counts, and who gets credit for it, when nobody agrees on how to look.
2015: fully remote (5 days/week) | 2024: hybrid (2 remote days/week) | |
|---|---|---|
Attrition | Cut roughly in half | Down about 33% for non-managers; unaffected for managers |
Measured performance | Up 13% (more minutes worked, more calls/minute) | No measurable difference, across two years of reviews |
Career advancement | Promotion rate almost 50% lower, conditional on performance | No promotion penalty reported |
Two Chinese call centers are not a universal law
Both of these are one company, both are in China, and both happen to sit in roles with unusually clean, automatically logged output — calls per minute, formal review scores — that most engineering work doesn't have. That's not a reason to dismiss them. It's a reason to be precise about what they actually license you to believe. They are, as far as I can tell, the best controlled evidence that exists for how remote arrangements affect career advancement, specifically because almost nobody else has run a real randomized trial on this question at all. Most of what circulates about remote work and careers is survey data, which measures what people believe happened to them, not what happened. works through AI agents and marketing automation in more operational detail. XenGrowth on AI agents and marketing automation works through AI agents and marketing automation in more operational detail.
What these two studies together support is narrower and more useful than either “remote work is fine” or “remote work kills your career”: the size of the promotion effect tracked the size of the remote arrangement. Five days unseen cost real ground. Two days unseen, in the only comparable experiment that exists, did not. Whether that holds at three days, or in a software team instead of a call center, or under a manager who has never worked anywhere but an open floor, is a real question this data cannot answer — which is exactly where the mechanism, rather than the headline number, becomes the useful part.
What visibility actually means, mechanically
The mechanism both studies point at is visibility to the specific person deciding on promotions, not visibility in general and not “being active on Slack.” A manager who sees you daily forms an impression of your work from a hundred small, low-effort observations — a comment in a hallway, a whiteboard glimpsed mid-thought, being the person other people go to when something breaks. Remove the physical proximity and every one of those observations has to be replaced by something deliberate, or it simply doesn't happen, and the manager's picture of your contribution quietly shrinks to whatever made it into a status update. That gap is the entire subject of how remote engineers get visibility without performing busyness, and it matters more than this post has room to argue in full.
The same instinct that makes marketing teams obsessed with — measuring what a visitor actually does on a page, not how convincing the page looked in a meeting — is the one a manager needs turned on themselves when judging someone they rarely see: the promotion decision gets made on what was visible, not on what was true. If your team hasn't built a way to make good remote work visible on purpose, the Ctrip result is a description of what happens by default, not a special case. For the AI search, GEO and discovery angle, see . For the AI search, GEO and discovery angle, see XenGrowth on AI search, GEO and discovery.
Written proposals and postmortems that circulate under your own name, not summarized by someone else on your behalf
Asking directly to present your own work in a review or all-hands, rather than assuming the work will speak for itself
Being the person who answers in the open channel, in view of the whole team, instead of resolving things quietly in a DM
Volunteering for cross-team or on-call rotations that put your name in front of people outside your immediate manager
None of that is about performing effort. It's about making sure the artifact of good work reaches the specific person who decides on advancement, since the Ctrip result suggests that if it doesn't reach them by some deliberate route, it mostly doesn't reach them at all. An office puts that transmission on autopilot in a way a remote setup simply does not, whether or not anyone intends it that way.
Full remote without any deliberate visibility habit really did cost promotion odds in the one dataset built to measure it. Don't assume that generalizes automatically to every company — but don't wave it away either.
Hybrid, even a modest two days a week, showed no such cost in the only comparable randomized trial. The mechanism (manager visibility) appears to track the dose (days spent unseen), not remote work as a single category.
If you're mostly remote, build visibility on purpose: design docs and postmortems that circulate widely under your name, asking to present your own work rather than having a manager summarize it, being reachable in the open channel rather than only in DMs.
Push your organization toward promotion criteria anchored in artifacts — design docs, incident write-ups, review comments — rather than hallway proximity. The Ctrip promotion gap was a failure of measurement, not proof that remote workers deserve less.
Treat the specific configuration of an entirely office-based manager evaluating an entirely remote report as the actual risk factor here, not “remote work” as an undifferentiated category.
None of this makes the Ctrip finding disappear, and it shouldn't. A well-run randomized trial found that better, cheaper, happier work went with worse odds of advancement, and that's an uncomfortable result precisely because it isn't about effort or output. It's about who gets to see the work at all. The honest reading isn't that remote work is a career mistake. It's that career advancement, in this specific pair of studies, tracked whichever kind of work someone could actually watch happen — and a manager can't watch what a screen doesn't show them.
Further reading from XenGrowth
Where this work meets go-to-market
The same visibility problem shows up whenever a team decides what counts as good work without a shared way of measuring it. publishes operator guides on building exactly that kind of shared measurement into a revenue team's day-to-day — the commercial version of the promotion question this post has been circling.
Further reading from XenGrowth
Where this work meets go-to-market
The operational playbooks that sit alongside remote work live with .
Further reading from XenGrowth
The XenGrowth resource library — what you'll learn: how the commercial side of this work is run, across search, automation and revenue operations.
XenGrowth on AI agents and marketing automation — what you'll learn: how the teams who own AI agents and marketing automation plan and measure it.
XenGrowth on AI search, GEO and discovery — what you'll learn: how the teams who own AI search, GEO and discovery plan and measure it.
Where this work meets go-to-market
The operational playbooks that sit alongside remote work live with XenGrowth's operator guides.
Five questions on the two studies this post leans on. Both are real experiments, not surveys, and the details change what you should take from them.
