People analytics: What the tech can, and can’t, do
The term “people analytics” is often used casually or even reverentially without more than a superficial understanding of what it means. But the technology is fast becoming a powerful tool to assist HR in strategy development and tactical decision-making. To learn more about the topic, Employee Benefit News recently spoke to Giovanni Everduin, a senior advisor for HR and organizational development at management consultancy Boston Global and a recognized people analytics expert. Edited excerpts of that conversation follow.
EBN: What is your concept of “people analytics”?
Everduin: It’s the use of data in a structured way to get insight and drive better decision-making about an organization’s workforce. That insight can come from hindsight--looking back and understanding what happened in the past, and why. But for most people real analytics is when it gives you foresight to start making some predictions to support decisions to optimize future outcomes.
EBN: Do you encounter misconceptions about what people analytics actually is?
Everduin: Yes, people very often confuse analytics with what I would best describe as advanced operational reporting, pivot tables in a nicer visual interface. When I was a chief people officer and told the leadership that my team was working on an analytics project, everyone was like, “Great, great. I want analytics.” And so I asked for some strategic topics, some hypotheses they wanted to test. And the basic response is, “I just want analytics to tell me what to do.” However, today a lot of CFOs have a clearer understanding of analytics probably because they’re a lot further along than HR in terms of using data for insight.
EBN: When people do have a clearer understanding of how people analytics works, what are the most common things people are trying to figure out?
Everduin: The most common one that I come across is what’s causing attrition, and whether there are any particular hot spots within the organization that are at a much higher risk of attrition than others. The next question is why? This is more compelling now because it’s relatively easy to quantify the financial impact of attrition.
EBN: Don’t standard HCM systems today already have the capability, with all the data elements they take in from the employee population?
Everduin: Companies typically don’t have in good shape the kind of qualitative data you get in an exit interview.
If you are able to get frank insights from people about why they left, that could be incredibly powerful and provide some context that really gives more meaning to the data you already have, like demographics.
EBN: What else?
Everduin: Another one is understanding the drivers of performance and productivity. Some have a higher impact than others, negative and positive. Is it a manager? The temperature in the building? Are we hiring the right people?
EBN: What kind of data does the analytics need to produce good answers?
Everduin: Employers increasingly are using psychometrics and information from structured interviews. These create a wealth of data to start building profiles. You look at what kind of people do well in the organization six months, 12 months, three years, five years down the line. Where are these people coming from, what traits do they have in common--personality, cognitive abilities, cultural factors, and so on? And, in the process, you can learn who is hiring your best people, and who isn’t.
EBN: How are people analytics systems able to discern characteristics like personality? Is this from somehow interpreting spoken language, such as in an exit or job interview?
Everduin: There have been a lot of advancements lately with natural language processing. Google and other companies have done a lot of cool work in this area. This capability allows you, for example, to look at how often a certain phrase or term is mentioned, and start attributing meaning and context to it. It can detect whether something is said in a sarcastic way or in a positive way. Once you start getting a machine to understand that it can actually start attributing meaning to words, it can start grouping them in themes and start understanding subtle nuance of interpretation, beyond just the word.
EBN: How does the machine ingest the raw material to interpret it?
Everduin: That depends. Ideally the HR manager or whoever does the exit interview has an iPad that is recording and automatically transcribing what the departing employee says. For the moment, it’s probably going to be either people filling in a questionnaire on a portal by themselves, or someone sitting with pen and paper during the interview and taking notes.
EBN: How big a data set do you need for an analytics system to generate insights on a question like the causes of attrition, within a tight margin of error?
Everduin: If you’re a small organization, let’s say 50, 60 people, and you have 10 percent attrition, that’s not a lot, that’s five, six people a year, right? So from a statistical perspective, that obviously wouldn’t give you a lot of confidence in terms of predictive power. That being said, I think it’s still more than enough to give you quick and dirty insight on what’s going on.
Everduin: Let’s look at it in a different way. I don’t think most organizations are remotely advanced enough to start going to that level of statistical analysis. So I think just getting basic data out, whether it’s 50 people, or 1,000, it’s good enough to give you some first level of insight, because at the end of the day I don’t think analytics ever gives you the final answer. It gives you a very good clue of where to look, it gives you some very good possibilities and potential scenarios, but it still requires someone to go in and kind of make sense of it all, put it into context, interpret it, do some validations.
EBN: When you’re doing an analysis of attrition and it points to problems with some managers at high levels within the organization, maybe it’s easier for HR to navigate the politics of the situation since they can say, “This is what the system is telling us.”
Everduin: Sure, often you’ll end up exposing things some people would probably prefer not to be exposed. For that reason, it’s important for whoever is doing the analytics to report directly to the chief HR officer. Beyond that, I don’t think it’s HR’s job to fix every problem. I think the role of HR more and more is to identify problems, help identify solutions and work with the business units.
EBN: Have you witnessed cases of organizations making bad decisions based on an over-reliance of people analytics data?
Everduin: Yes. I mentioned analytics using psychometrics as a variable in performance assessment. You can start benchmarking people within the organization, like being one or two standard deviations above or below the norm. It can be dangerous if you give that kind of data to people and they start making decisions with it that they shouldn’t be making, like firing people. That’s really not what the data is for; it just tells you this is the kind of distribution in your organization. A low number might just mean that they’re in the wrong job, or that they’ve never been coached or developed, or something else.
EBN: What about the fact that analytics systems are created by imperfect humans in the first place?
Everduin: That’s right. Analytics and artificial intelligence are not necessarily without bias. It still reflects the prejudices of the person who programmed it, who set the questions or the boundary conditions. It’s easy to do an analytical project and have it come out to support whatever your initial hypothesis was going in, depending on what questions you’re asking.
EBN: Garbage in, garbage out?
Everduin: That’s right. To lower that risk, you need to have a couple of people review the assumptions you’re making and check for bias. And you still need to watch out for “group think.” I have this rule that whatever we find out, we’ll do another round to see if we can find a way to prove the findings wrong, to look for data to the contrary, to test our conclusions. That helps to keep you honest.