Forging a Predictive Path

For many talent managers, 2014 was going to be the year of predictive analytics.

The year was rife with predictive analytics conferences and research reports. Some corporate partnerships — including one between professional services firm KPMG and technology firm McLaren Group — foretold of a business environment in which predictive analytics would radically transform the insight available at talent leaders’ disposal.

And to some degree, it has. Still, as the industry forges ahead in 2015, evidence suggests the prognostications on predictive analytics in human resources may have been too optimistic.

“If predictive analytics were a baseball game, we’d be in the second inning,” said Brian Kelly, president at Vestrics, a predictive analytics firm in Durham, North Carolina. “Everyone’s appetite is quite large, but they are just learning how to do it.”

Despite the slow progress, some companies have shown great strides in using predictive analytics in talent management. These firms, profiled here, have used analytics to identify competency gaps, pinpoint reasons for sales slumps and improved hiring decisions.

What makes these companies successful, said Bill Schmarzo, chief technology officer of EMC Corp.’s global services for big data practice, is that they don’t begin with talent management — they end up there.

“Most predictive analytics efforts start by focusing on strategic goals to do things like create new revenue or monetize insights, vs. building organizational capabilities,” Schmarzo said. “They have to tie talent management back to that strategic goal.”

Sears Gets In the Mood

This is how Ian O’Keefe, head of talent analytics at Sears Holding Co., said he approaches the job. O’Keefe leads a team of analysts at the Hoffman Estates, Illinois-based retailer that has developed a complex analytical process for using people data, sales data, customer data and operations data to predict business performance.

“Sears is a deeply analytical company,” O’Keefe said. Predictive analytics “enables us to understand how things like collaboration, engagement and leadership influence outcomes that our business leaders care about.”

A recent example of this involved tracking retail staff mood and its influence on sales and customer experience. O’Keefe’s team installed mood-tracking systems on all of the company’s time clocks at its stores. When employees punched in and out, a digital screen asks them to rank their mood on a five-point scale, ranging from “unstoppable” (one) to “frustrated” (five) or “not in the mood to answer” (six).

“We collected 80,000 moods per day over 15 months,” O’Keefe said.

Using the data, the data team built a model comparing overall mood ratings to sales numbers, including sales per hour and per store.

Among the more interesting findings: When people come to the store to buy a specific item, like a television, the mood of their salesperson won’t directly affect the sales outcome. But when O’Keefe’s team narrowed the analysis of the data to look at what else customers purchased, a trend emerged. When moods are positive, customers are more likely to purchase peripheral products, like surge protectors, extra cables and protective warranties.

“There is a lot of margin in those items, which leads to a more profitable basket,” O’Keefe said.

Because O’Keefe and the Sears data team can now predict the mood of a particular store on a particular day — busy traffic days tend to have lower moods than slower days — they are now exploring strategies to influence mood on high-traffic days through manager engagement.

For example, the team is piloting an automated email program that sends store managers videos and short training programs on how to motivate people. The email will also send reminders at key times on busy days to walk the floor and encourage team members. This is especially the case near the end of a shift, when moods tend to fall.

“It’s a great example of how we are using analytics to understand the things that incrementally influence sales outcomes in the store,” O’Keefe said.

Happy Nurses at Dignity Health

Not all organizations have to have a mature analytics process to make a difference, however.

Todd Horton, director of workforce intelligence at hospital system Dignity Health in San Francisco, said his team is just developing predictive analytics skills and has still produced valuable results. To help the team bolster its analytics capability, it partnered with talent management analytics firm Strategic Management Decisions HR.

Together, the team is consolidating data from employee surveys, satisfaction results and other internal data sources to identify correlations between employee behavior and patient outcomes. Working with a vendor has helped the team contribute now while building in-house expertise for future analytics efforts, Horton said.

One of the team’s early dives into data explored whether employment satisfaction was negatively influenced by “span” — how many nurses answer to a single manager — and the effect it had on patient satisfaction scores. The team’s gut instinct said that the more people a leader has to manage, the higher the risk that employees would be unhappy, resulting in a lower score.

But the data showed otherwise. While span does increase the risk that employee satisfaction scores will be lower, highly trained managers were able to mitigate that risk, Horton said.

The team also discovered using analytics that nursing units that have a high rate of missed meal breaks have lower satisfaction scores. “That meant we were spending more money, because employees were missing those breaks and getting negative outcomes as a result,” Horton said. Further investigation showed the missed meal breaks were a broader indication that people were being worked too hard, and that their managers didn’t have the most effective leadership skills or training.

All of this analysis is enabling Horton to build a business case for change. He plans to roll out a pilot program in one division focusing on improving leadership development and coupling productivity goals with improved employee engagement efforts.

“It will be controversial because our industry is all about productivity numbers,” Horton said. “But the data shows us that productivity alone isn’t a magic bullet.”

Picking CCTV Executives

State television broadcaster China Central TV is another example of an organization harnessing simple analytics to predict future outcomes — in this case to make better hiring decisions for its global executive team.

“When I first brought up predictive analytics, a lot of people said, ‘No, that’s something only big tech firms do,’ ” said Glen Loveland, CCTV’s human resources manager in Beijing. “But the truth is that there are a lot of tools available to small- and midsized businesses that put predictive analytics within reach.”

When CCTV initiated plans to open new operations in Nairobi, Kenya, and Washington, D.C., Loveland knew they needed to scale the way the organization made hiring decisions. Many of the company’s leaders are expats — meaning they are notoriginally from the location in which they are working — and hiring them can be both risky and expensive. Loveland estimates hiring someone for a three-year expat assignment can run $1 million and take months to complete.

“We had to be sure the candidates we chose for these new positions would be the right fit,” Loveland said.

Previously, hiring decisions at the firm were based largely on gut instinct. But Loveland wanted to add a quantitative aspect to the process, so he implemented a candidate assessment tool from vendor Harrison Assessments to better measure those that would do well in these new roles based on the competencies of past high-performers.

“Like a lot of organizations, we need people who are risk-takers and who have a high tolerance for ambiguity,” Loveland said of his ideal candidate. This is especially true for expats, who need to be willing to uproot their families and immerse themselves in a foreign culture. “You can’t tell if someone has a high tolerance for ambiguity based on their résumé alone,” he said.

The assessment helped his team identify the characteristics of ideal risk-takers, as well as other critical competencies that suggest candidates will succeed in various positions. They used profiles of current high performers in the company when they were available and industry benchmarks for new roles, like social media engineer.

The assessments didn’t just identify whom to hire, but they allowed Loveland’s team to gauge strengths and weaknesses in prospective candidates and compare the profiles of multiple candidates to make the best hiring choice. It also let them know what professional development new hires would need to close any skill gaps so they would be better prepared to succeed.

Loveland said he discovered that not every piece of data is equally predictive, however. He initially planned to use the assessment to evaluate verbal reasoning and numerical reasoning for engineering candidates, which seemed logical for the highly technical roles. But the resulting analysis showed that how both candidates and existing engineers scored on these tests was largely irrelevant to their success on the job.

“We figured out that engineers are more likely to be successful in our company if they meet more of our cultural fit requirements,” Loveland said.

So while the tests didn’t deliver the expected value, the resulting analysis did help the company identify the measures that would have an impact on success. 

For the most part, Loveland said he sees this early foray into predictive analytics as a success. Since implementing the executive basements 14 months ago, CCTV has had no turnover among executive hires, and 80 percent of candidates have renewed their contract for another year.

“I wouldn’t attribute these results solely to predictive analytics,” Loveland said, “but it helped us add a quantitative step to something that used to be all gut and chemistry.”

To learn more about how to implement predicitve analysis at your organization without a sophisticated approach, read the sidebar that accompanies this feature.