HR data analytics demands sensitivity to business needs
Panelists from Chevron, Morgan Stanley and Wal-Mart explained how to build productive relationships between HR analytics teams and the business side.
Effective HR data analytics requires close collaboration between data scientists and business managers, according to experts from Chevron, Morgan Stanley and Wal-Mart who participated in a recent panel at the HR Technology Conference & Exposition in Las Vegas.
The panel was moderated by Brian Kelly, president of Vestrics, a maker of cloud workforce analytics software based in Chapel Hill, N.C. In the first part of this panel report, the experts explained how they use HR analytics for workforce optimization.
- HR data analytics and business teams must work together
Jeremy Shapiro, an executive director in HR and head of global talent analytics at New York-based securities firm Morgan Stanley, shared the nuances of getting HR data analytics staff to work effectively with the business side.
“In an organization with 55,000 people, there’s complexity, and it’s really hard to understand all the business drivers for that many businesses all at the same time,” Shapiro said. “Many times we don’t know what the business problem is [before] walking into a meeting with a senior leader.”
It’s almost as if there are three partners, he said: business leader, analyst and the sheet of statistics they’re going over. “Most of the time, we’re able to find nuggets together,” which can make business leaders more committed to the analysis and comfortable drawing their own conclusions from the data, he said.
RJ Milnor, manager of planning, analytics and reporting at Chevron Corp., an energy provider based in San Ramon, Calif., advocated a similarly soft touch when collaborating with business managers. “Sometimes we feel like we need to provide the answer for the business,” Milnor said. “The business really doesn’t like that. Most business leaders actually know their business really, really well. What works well for us is helping the business ask questions they may not have asked before.”
Chevron’s analytics group functions like internal consultants, Milnor said. “It’s providing the strategic questions back to the business for them to consider and providing them new options to achieve their business objectives. And that has worked extraordinarily well.” It helps to bring the HR data analytics team into the conversation at a senior level, he added.
Sometimes using HR data analytics effectively requires trial and error, Milnor said, recounting a time his group set out to analyze a learning and development program to understand its impact on Chevron, running multiple regression analyses on the effect on attrition and productivity.
“There weren’t any [effects], which kind of surprised us,” he said. The problem was Chevron had taken a data-mining approach — throwing a lot of data and seeing what came out — instead of a hypothesis-driven one employing root-cause analysis or some test of the hypothesis.
“We put our engagement data in there, and we saw a nice relationship between engagement and safety measures, which we didn’t know existed before,” Milnor said.
Kelly stressed the importance of understanding that predictive analytics can’t be precise, which CFOs know well from their financial forecasts.
“There has to be a mind shift to be able to walk in and feel comfortable with a dialogue,” Kelly said. “It’s not always about the math. It’s about the story the business is looking for.
“There’s really not a science to move from reporting to predictive to optimization,” he continued. “It’s more of a process and a dialogue and a comfort level with being incorrect with some numbers and going back and re-optimizing the models.”
- Choosing software, managing data
Kelly outlined several options for approaching HR data analytics. “There is enough packaged software that exists within your [human resource information systems] to handle the reporting. When you start to move into predictive, there are [a] number of ways that you can look to approach this.”
First, decide what types of statistical tests to apply to each situation, and whether to use hypothesis- or data-driven approaches. Build a team and bring in resources from finance, strategic planning or HR: It will be tougher to scale and not always efficient, but more learning is possible. Hiring a consultant is another route, but the methods will be harder to repeat.
The panel consensus was that data doesn’t have to be perfectly clean, and what analysts need instead is knowledge of how much accuracy is required for each problem.
“I’ve never worked in an area where the data is perfect,” said Elpida Ormanidou, vice president of global people analytics at Wal-Mart Stores Inc., the retailer based in Bentonville, Ark. “A really good analytics person will be able to assess what is good enough. … You can make a lot better decisions from good enough data than no data at all.”
Kelly said never to start with data cleanup because it’s a never-ending process.
It’s better to decide first how much the data inconsistency influences the model and tell senior managers why it’s worth cleaning up the data.
Analytics will always involve probability-based models, according to Kelly. “You’ll never get to the point where a software vendor will be able to say, ‘Here is the future, and therefore you should plan for this.'” Instead, “futurists” can propose what-ifs and ranges of likely outcomes.
“At the end of the day, it always goes back to the business person saying, ‘I believe this is the most likely path forward, and we’ll plan for this, and we’ll calibrate along the way.”