In 2010 Ctrip was China's largest online travel agency and, like every call centre in the world, it had two problems: office space in Shanghai was expensive and staff left quickly. The company's chairman, James Liang, was then doing a PhD in economics at Stanford, and proposed something rare in business: instead of arguing about whether working from home works, run a randomised controlled trial.
The company asked 996 employees in its airfare and hotel departments who wanted to work from home. 503 volunteered. After screening for tenure, broadband and a private room to work in, 249 qualified. They were split by whether their birthday fell on an odd or even day: half worked from home four days a week for nine months, half stayed in the office1.
Nobody chose their own group. That is what makes the result more trustworthy than any opinion survey.
What the first experiment found
The home group's performance was 13 percent higher1. What matters is where that came from: about nine percentage points came from working more minutes per shift, with fewer breaks and fewer sick days, and the rest from handling more calls per minute in an environment quieter than the call centre floor1.
Attrition in the home group fell by about half. Job satisfaction rose1.
But one number moved the other way, and it is the one HR should remember: at the same level of performance, home workers were promoted less often1. They did more, and were seen less.
After the trial the company opened home working to everyone and let the original participants choose again. More than half switched from their assigned arrangement, and once people could pick what suited them, the productivity gain nearly doubled, to around 22 percent1.
A decade later, another experiment
By 2021 the company had become Trip.com and the question had changed. This time the subjects were not call centre staff but graduate professionals: engineers, marketing and finance. And the arrangement was not four days at home but hybrid: home on Wednesday and Friday, in the office the other three days.
1,612 people in the Airfare and IT divisions took part, 395 of them managers and 1,217 non-managers. Again the split was by odd or even birthday, again it ran for six months, and this time the results were published in Nature in 20242,3.
Performance did not change. Review grades over the following two years were no different between the groups, and neither were promotions2. What changed was who stayed: quit rates fell from 7.2 percent to 4.8 percent, a drop of one third. Job satisfaction rose from about 7.8 to 8.22,3.
Trip.com estimated that each departure cost roughly 20,000 dollars in recruitment and training3. After the trial, it offered hybrid working to everyone.
Reading the two experiments properly
The limits need stating. Both ran at one company, in one industry, in China. One of the authors is James Liang, Trip.com's co-founder and chairman, which the paper discloses2. A good experiment in one place is not a law for every place.
Still, three things come out of them, and they stand on firmer ground than most of what gets said about this subject.
First, in both trials the biggest benefit of working from home was not productivity but retention. In repetitive work, output rose because people worked more minutes. In knowledge work, output held steady while quits fell by a third. If a company is deciding whether to allow hybrid, the number to put on the table is the cost of replacing people, not a feeling about diligence.
Second, the design matters more than the permission. The second experiment fixed the same home days for everyone, so the team still met three days a week. That is very different from everyone picking a different day and nobody overlapping.
Third, the most worrying detail is in the first trial: equal performance, fewer promotions. That is proximity bias, and it does not disappear just because a company allows remote work.
