- June 6, 2017
- Posted by: Staff
- Category: Data
By Jon M. Jachimowicz
“My best employees are leaving,” Daniel told me, “and I can’t seem to figure out why.”
Daniel (not his real name) was a VP human resource manager at a Fortune 500 company. I asked him whether he had collected any data that could provide him with insights into systematic patterns. “I made sure we get exit interviews done with every single employee who is leaving us,” he replied. “I even personally conducted some myself! But no consistent pattern is emerging. I’m not sure how I can prevent my best employees from leaving us in the future.”
Here is the problem with exit interviews: People aren’t honest about the reasons why they quit. And even if they were, such post hoc rationalizations rarely reflect the true reasons employees quit.
Daniel’s conundrum is one many HR managers encounter in their organizations. Why are the best employees leaving the organization? Why are some employees more productive than others? How can employees become more creative? Often, the information that can help answer these questions already exists within the company, hiding in plain sight.
Although companies collect a great deal of data about their employees, most of them don’t do a great job of leveraging it for insights to answer these questions. If companies could improve their data practices at five important stages, they could become much more effective at solving some of the most pressing problems they face.
Step 1: Improve data quality. After listening to Daniel’s problems, I asked him what kind of data his company collects. A lot, it turns out. His department sends out a survey for all employees to fill out every six months. Managers conduct annual performance reviews that they log in a centralized system. The HR department keeps track of every promotion, while the operations department monitors which employees leave the organization.
However, when I asked to take a closer look at how Daniel’s department was collecting data, I was aghast. The survey did not collect data in a reliable, validated way. The performance reviews weren’t structured, and only 55% of managers filled them out. And the promotion and turnover data didn’t include dates.
Before you can use your data to get answers, you have to improve the quality of the data you collect. Design a more rigorous survey with better measures. Create a performance review system that makes it easier for managers to log their reviews. Think through what kind of data will be useful to collect, and then collect it — systematically. Have regular conversations with people from throughout the company to identify what questions are pressing and what kind of data you may need to answer those questions.
Step 2: Link different data. To answer a question like “Why are my employees are leaving?” you need to compare employees who’ve stayed with employees who’ve moved on. (Which is another reason exit interviews often don’t work — you’re only getting half the story.)
To do this, you need to link the data from different sources throughout your organization. In Daniel’s case, the data was championed by different departments. Performance reviews and employee surveys were managed by the HR team, whereas data on turnover was held by the operations team. Neither team realized which data the other team held, so they needed to change their processes to ensure they could connect the employees who’d left with their survey responses and performance reviews.
Find out what kind of data is being collected in the organization. Design processes that make it easier to connect the dots between individuals to get as many data points on each employee as possible.
Step 3: Analyze your data. Simply put, data analysis requires data processing abilities. For example, in some cases your performance outcomes might be at the group level: the success of a team project, or a successful outcome for a client team. Is it possible to deduce what made the project successful from the individual-level responses from each team member?
The answer is yes, but it’s not easy to do. In statistical terms, you may need to nest responses at the group level and run a random or fixed-effects model. What this means is you investigate the variability of individual-level responses to predict group-level outcomes. However, this goes far beyond the capabilities of what Microsoft Excel can do. To decide what kind of data analysis techniques to use, and more important, to conduct the analyses, you need skilled data analysts capable of using advanced data processing software, such as R or Stata.
The third step to leveraging your data, then, is to be competent in your data analysis. Think through what kind of analysis techniques are most suitable, given your type of data. Ensure that you have staff available who can conduct the necessary analysis; if you don’t, recruit or contract with experts who can help.
Step 4: Infuse your data with theory. Although many problems might seem pressing, you are not the only one faced with them. A lot of attention has been spent in the last few decades investigating the predictors of employee performance, turnover, and creativity. Academic researchers have documented what relationships exist and developed a wealth of theory that explains why they do.
This is important because theories can help us predict what will occur in the future, given a set of considerations. Hence, while data analysis is often retrospective, trying to understand after the fact why a group of employees left, strong theory can facilitate organizations forecasting who is most likely to leave in the future. In addition, strong theory can help identify what kinds of questions an organization should be asking when it is faced with a problem, such as rampant staff turnover.
The fourth step to leveraging your data is to infuse your data with theory. Investigate past research that has attempted to provide answers to similar questions you may be asking yourself. Look through what this research has investigated, how the researchers investigated it, and what theory they developed to explain the relationships they found. You may not have the time and resources to do this — that is fine. Academic researchers, in many cases, are more than happy to serve as advisors on projects and can help guide you along the way.
Step 5: Implement changes and keep track of outcomes. You have done it all: You increased your data quality, connected disparate data sets throughout the company, recruited strong data analysts, and consulted the research on relevant theory. You have a working model of why your employees are leaving the organization. Now that you have this insight, you need to turn it into an intervention. For example, in Daniel’s case, we found that many of the company’s best employees left because they did not feel they had sufficient autonomy over how they carried out their jobs.
This is a crucial step of the process: testing whether what you have learned can provide actionable insights that improve your organization. Daniel and I tried an intervention where we gave some employees, but not others, the chance to make their schedules more flexible. This type of split-testing was important because we wanted to have a control group. The intervention had no effect the first time we ran the experiment. That was obviously disappointing, but the good news was that our improved data practices allowed us to understand why that was the case and enabled us to optimize our intervention until it had the intended effect.
The fifth and final step is therefore to implement changes and keep track of relevant outcomes. The intervention may require several attempts to have the intended outcome. Some interventions might not work at all, and others may even backfire. But the best way to find out whether the insights you have gained are accurate is to put them to the test.