In early 2009, statisticians in Google embarked on a plan codenamed Project Oxygen. Their mission was to devise something far more important for Google Inc. than its next search algorithm or apps. They wanted to build better bosses.

So, it began analysing performance reviews, feedback surveys and nominations for top-manager awards. They correlated phrases, words, praise and complaints. Later that year, the “people analytics” teams at the company produced what might be called the Eight Habits of Highly Effective Google Managers.

Laszlo Bock, Google’s vice-president for “people operations”, which is Googlespeak for human resources and his team began ranking eight directives by importance. And this is where Project Oxygen took off.

For much of its 13-year history, Google has taken a simple approach to management: Leave people alone. Let the engineers do their stuff. If they get stuck, they’ll ask their bosses, whose deep technical expertise propelled them into management in the first place.

But what Bock’s group found was that technical expertise ranked dead last among Google’s big eight. What employees valued most were even-keeled bosses who made time for one-on-one meetings, who helped people puzzle through problems by asking questions, not dictating answers, and who took an interest in employees’ lives and careers.

Project Oxygen is noteworthy for a few reasons, according to academics and experts in this field. HR executives have depended in the long run on gut instincts more than hard data. But a growing number of firms are trying to apply a data-driven approach to the unpredictable world of human interactions.

“Google is really at the leading edge of that,” says Todd Safferstone, managing director of the Corporate Leadership Council of the Corporate Executive Board, who has a good perch to see what HR executives at more than 1,000 big firms are up to.

“Many companies adopt generic management models that tell people the roughly 20 things they should do as managers, without ranking those traits by importance. Those models often suffer “a lot of organ rejection” in companies, added Safferstone. “This happens because they are not presented with any evidence that will make a difference, nor do they prioritise what matters.”

Project oxygen started with some basic assumptions. People typically leave a company for one of three reasons, or a combination of them. The first is they don’t feel a connection to the mission of the company, or sense that their work matters. The second: they don’t really like or respect their co-workers. Also critically, most leave because they have a terrible boss — and this was the biggest variable. Google, where performance reviews are done quarterly, not annually, saw huge swings in the ratings employees gave to their bosses. Managers also had a much greater impact on employees’ performance and how they felt about their job than any other factor.

“The starting point was that our best managers have teams that perform better, are retained better, are happier,” says Bock. “So the biggest controllable factor we could see was the quality of the manager, and how they sort of made things happen.

In Project Oxygen, the statisticians gathered more than 10,000 observations about managers — from across more than 100 variables, from various performance reviews, feedback surveys and other reports. Then they spent time coding the comments in order to look for patterns.

Once they had some working theories, they figured out a system for interviewing managers to gather more data, and to look for evidence that supported their notions. The final step was to code and synthesise all those results — over 400 pages of notes and then they spent much of last year rolling out the results to employees and incorporating them into various training programmes.

“Reading and coding all the information was time-consuming. This was one area where computers couldn’t help,” says Michelle Donovan, a manager of people analytics. “People say there’s software that can help you do that,” says she. “It’s been our experience that you just have to get in there and read it.”

Given the familiar feel of the list of eight qualities, the project might have seemed like an exercise in re-inventing the wheel. But Google generally prefers, for better or worse, to build its own wheels.

“We want to understand what works at Google rather than what worked in any other organisation,” says Prasad Setty, Google’s vice-president for people analytics and compensation.

Once Google had its list, the company started teaching it in training programmes, as well as in coaching and performance review sessions with individual employees. It paid off quickly.

Bock tells the story of a manager whose employees seemed to despise him. They found him bossy, arrogant, political and secretive. They wanted to quit his team. “He’s brilliant, but he did everything wrong when it came to leading a team.”

Google gave him one-on-one coaching — the company has coaches on staff, rather than hiring from the outside. Six months later, team members were grudgingly acknowledging in surveys that the manager had improved. “He’s nowhere near one of our best managers, but he’s not our worst anymore. And he got promoted.”

Mark Klenk, an engineering manager whom Google made available for an interview, said the Project Oxygen findings, and the subsequent training, helped him understand the importance of giving clear and direct feedback to the people he supervises.“There are cases where some people who don’t necessarily realise they need a course correction,” Klenk says. “So it’s just about being clear about saying, “OK, I understand what you are doing here, but let’s talk about the results, and this is the goal.”

Google executives say they aren’t crunching all this data to develop some algorithm of successful management. The point, they say, is to provide the data and to make people aware of it, thereby helping managers understand what works and what doesn’t.

The traps can show up in areas like hiring. Managers often want to hire people who seem just like them. So Google compiles elaborate dossiers on candidates from the interview process, and a special group makes hiring decisions. “We do everything to minimise the authority and power of the manager in making a hiring decision,” Bock explains.

Google also tries to point out predictable traps in performance reviews, which are often done with inputs from a group. The company has compiled a list of “cognitive biases” for employees to keep handy during these discussions. For example, somebody may have just had a bad experience with the person being reviewed, and that one experience inevitably tramples recollections of all the good work that person has done in recent months. There’s also the “halo / horns” effect, in which a single personality trait skews someone’s perception of a colleague’s performance.

The true test of Google's new management model, of course, is whether it will help its business performance in the long haul.