#1354 2010 · Ctrip 携程 · Travel services
Ctrip ran a coin-flip trial on working from home — then let workers choose, and gains doubled
the problem
A Shanghai call center was bleeding money on office space and losing half its staff a year, but nobody knew if home-working would crater output
background
Ctrip, China's largest travel agency, faced expensive Shanghai real estate and ~50% annual attrition in its call center — every departure burning four months of hiring and training cost. Home-working promised savings on both, but management feared the obvious: unsupervised agents shirking. Opinion was split exactly the way it is everywhere, on anecdote rather than evidence.
Instead of arguing, co-founder James Liang — then a Stanford economics PhD student — ran the question as an experiment: volunteers with an even birthdate worked from home four days a week for nine months, odd birthdates stayed in the office as the control, identical pay structure, performance tracked by the phone system that already measured everything.
what everyone would do
Decide by management judgment: pilot home-working with a hand-picked team (whose results prove nothing), survey employees (who tell you what they think you want), or copy whatever policy peers announce. Every version leaves the actual productivity question unanswered and the policy vulnerable to the first bad quarter.
what they saw
The call center already logged every call and minute — the measurement infrastructure for a real experiment was lying there unused. And the trial's biggest finding wasn't the 13%: it was that letting workers sort themselves afterwards doubled it.
the move
The randomized trial (with Stanford's Nicholas Bloom) settled the question with the company's own data: home workers took 13% more calls — more minutes per shift (fewer breaks, no commute friction, less sick leave) and more calls per minute (quieter environment) — while quits fell by half and reported satisfaction rose. Then came the move most companies skip: Ctrip let everyone re-choose. Half the home group came back (lonely), office workers who wanted home went home — and once workers sorted themselves by what suited them, the performance gain roughly doubled. The experiment was published in the Quarterly Journal of Economics and became the standard evidence on remote work.
why it works
Randomization removes the selection bias that makes every anecdote-driven workplace argument circular ('our best people would...'). Using the phone system's existing metrics made the trial nearly free and the results impossible to dismiss internally — the same dashboard management already trusted. The re-choice phase then captures what no mandate can: workers know things about their own homes, temperaments and families that no policy can observe, so voluntary sorting allocates each person to the environment where they produce most.
the payoff
+13% performance, quits halved, ~$2,000 saved per employee per year — and self-selection after the trial nearly doubled the gain (QJE 2015).
where it breaks
It fails where output isn't individually measurable (team production, creative work) — the call center's clean metrics are why the result was clean, and generalising it to jobs without such metrics is exactly the mistake the authors warned about. Mandated remote work forfeits the selection gain entirely, and thin slices of volunteers can't reveal effects that only appear at full scale (culture, mentoring, promotion gaps — which later Ctrip data did show for the always-home).
what came after
Became the canonical pre-pandemic evidence on remote work, cited worldwide when COVID forced the question; made 'run the experiment, then let people choose' the reference design for workplace policy.
references
- [1]Does Working from Home Work? Evidence from a Chinese Experiment (NBER WP 18871)NBER (Bloom, Liang, Roberts, Ying), 2013nber.org
- [2]Does Working from Home Work? — Quarterly Journal of Economics 130(1)Quarterly Journal of Economics, 2015academic.oup.com