In this half-hour talk posted last week at re:Work, Google’s VP of People Operations Prasad Setty discusses his experience leading the development of the search giant’s talent analytics program, and about the key difference he discovered between having data make decisions for people, and using data to improve the way people make decisions:
When Prasad Setty joined Google ten years ago to build its People Analytics team, he envisioned a workplace where all people-related decisions would be made by data and analytics. If algorithms were spitting out search terms, why couldn’t we use them to make decisions for and about our people?
Setty soon discovered that this was the wrong approach. Despite the ability of analytics to objectively predict outcomes with high accuracy, people were reluctant to rely solely on formulas when it came to making important decisions — especially decisions that involved people, such as a promotion. And so, Setty shifted his vision for the People Analytics team. Rather than using data and analytics to make all decisions at Google, the team’s mission would be to educate Googlers on how they were making decisions and to help them make better decisions over time.
What really stands out about Google’s approach here is that they chose not to use a quantitative focus, even though they had the analytic sophistication necessary to do so. At one point, Setty mentions how HR was able to create a logistic predictive model that was able to accurately predict promotion decisions with an error rate of only 10 percent based on a few easily measurable attributes. Despite this, the engineers involved in the hiring process made it very clear that they did not want to outsource such an important task away to an algorithm.
This is an important lesson in how organizations can effectively use data in managing talent issues, particularly culture change.
What Google does right here is that they recognize where data is helpful and where it is not, and are honest about the fact that data will not solve every problem for them. Even though they could predict promotion outcomes pretty well, this was not aligned with how leaders wanted to make promotion decisions.
This challenge is a key reason many organizations struggle to act on their culture based on their current data. Instead of recognizing the shortcomings of their current approach to measuring culture, too often do we hear organizations overextending their engagement and attrition data, which although useful, does not provide sufficient resolution to provide clear guidance as what what leaders should keep and what they should change to get their culture to perform more effectively. That’s partly why some of the key challenges facing talent analytics leaders are aligning analytics to business challenges and securing buy-in throughout the organization.
Another organization that successfully took a similar approach to data-driven decision making is Seagate, which CEB (now Gartner) examined in a 2013 case study. The challenge Seagate was solving for was the exact same situation Google ran into: When HR starts conversations with prescriptive insights drawn from data, it often causes line managers to take a knee-jerk defensive position and focus on criticizing the HR data, and then the whole analysis HR presents is undermined. So what Seagate does instead is:
- Develop simplified data visualizations that help leaders diagnose challenges themselves
- Provide implications of alternative leader decisions through decision scenarios
- Provide action owners with implementation guidance based on leaders’ decisions
It’s important to remember that in this age of growing talent analytics, people may say they want predictive data, but that doesn’t mean they’ll use it if it’s forced down their throats. No one likes a “know it all,” whether that refers to a person or a computer. Talent analytics teams would do well to take a page from Google and Seagate to give leaders more data-driven options instead of mandates.
CEB Corporate Leadership Council members can read our Seagate case study here.