Google offered some insights yesterday into its innovative, data-driven HR process. “All people decisions at Google are based on data and analytics,” said Kathryn Dekas, a manager in Google’s “people analytics” team, speaking at O’ Reilly Strata. Those decisions cover compensation, talent management, hiring and all other HR issues. Google’s data-based HR may become a key factor in the company’s future success.
With 28,000 employees and a constant stream of new hires, HR is an important topic at Google. But true to Google’s engineering roots, HR, just like any other area of the company, needs to produce data to justify decisions and policies. This led to the creation of a people-analytics team, a hodge-podge of data miners, psychologists and MBAs.
One of the team’s better known endeavours is Project Oxygen, Google’s quest to build a better boss, or at least identify what makes a good one. Project Oxygen initially set out to determine if managers matter. In the very early days, Google got rid of all managers. Although they were later reintroduced, a belief persisted within the company that managers do not really make a difference.
So the analytics team looked at a combination of performance review data and employee surveys, where employees review their bosses to determine whether there were significant differences between the impact of the best and worst bosses. The answer from the data was a resounding “yes”.
The people analytics team pushed on to try to determine the common characteristics of the best managers and how to improve the skills of the worst. The conclusion was a list of the 8 skills of a good manager, at least according to Googlers. One surprise was that a boss’s technical expertise is much less important to employees than the ability to take a genuine interest in their lives and careers. The best bosses didn’t micromanage, had a clear vision for the team and were results-oriented.
The worst managers also had some behaviours in common. Googlers generally like having regular, one-to-one meetings with their managers. “One thing that was consistent among the struggling managers was that they were not consistent in who they offered one-to-ones to,” explains Dekas. “They may have been meeting with people who weren’t performing well, or with those who were performing exceptionally well”. One best practice that Google introduced, based on the insights from Project Oxygen, was to institute one-on-one meetings with all team members. The company also completely redesigned its training for new managers in line with the results. One year after project Oxygen reached its conclusions, “75 percent of our struggling managers have significantly improved,” Dekas reports.
Another project the team undertook was to forecast the future organizational structure of Google based on current hiring and promotion practices. It turned out that if Google continued to promote at the current rate, it would end up “fat in the middle”, with many middle-ranking employees and fewer opportunities for junior hires to advance. So Google implemented a new practice where the company doesn’t directly replace employees who are promoted or leave the company but instead hires new employees at a lower level. The people analytics team forecast that this would make career advancement easier for junior employees.
Debunking HR myths also turned out to be an important function of the analytics team. Like any company, Googlers had persistent, but often erroneous beliefs, about HR issues. Typical myths were that employees at Google’s headquarters were promoted more quickly than those in other Google offices, or that Googlers who worked on “shiny projects” were more likely to be promoted. The data showed that neither of these hypotheses was actually true, but the analysis did reveal that getting feedback from senior peers was the most important factor if you want to be promoted within Google.
Google’s HR process is currently the acme of data-driven people management. I asked Dekas if any room remains at Google for gut feeling or intuition. “You can’t have an algorithm for everything” says Dekas. “You use data to inform, but you don’t rely on the data to make the decision.”