I can hear it now: "Oh, Great! Yet another article about analytics saving the planet, delivering world peace, and moving everyone to a state of higher being!” While I don’t really believe any of that, many of us in HR nevertheless tend to treat analytics like something akin to nuclear physics, as something “special,” and ultimately something best approached only by experts.
To help eliminate some of these barriers to understanding, I want ask and answer five fundamental questions that people typically have about analytics but are afraid to ask. Once you understand these foundational concepts, you will have a new way to understand analytics, how it works, and what it can truly provide you, your colleagues, and your organization.
1) What exactly does “analytics” mean?
While answers might differ depending on whom you ask, the core of analytics for HR really comes down to this: a systematic analysis of available data to drive strategic and operational talent decisions.
In this context, “systematic” means using established, consistent methods for handling and analyzing data with statistics and visualization. If the process is not systematic, then it will be difficult to consistently interpret the analyses in a way that makes sense.
For example, if the survey items used to measure engagement change from year to year, then it will be all but impossible to accurately determine whether engagement has truly improved, declined, or stayed the same. Similarly, if the core methods of analyzing and visualizing the data change, then starkly different conclusions may be drawn irrespective of any “true” changes in behaviors or activities.
Why did I also highlight “available data”? Consider this: HR now not only has ready access to its typical slew of measures (e.g., compensation, employment history, performance measures, etc.) but also a wealth of potentially new information including email network analysis, geographic tracking, website usage statistics, and a host of predictive models (more on that later). This not only means more raw data, but also more ways to integrate and identify relationships in that data. In short, what is “available” today differs from what will be “available” tomorrow. Making sense of all it is another matter of course, but the key is to remember that nothing is off limits anymore.
2) What are the different kinds of analytics?
I find it most helpful to break down analytics into three layers.
Level 1- Descriptive analytics: The bottom and most critical layer is “descriptive analytics.” This is best summarized as asking, “What happened?” Average time to fill a position? Turnover and retention rates? Performance measures? All of these and many others fall under Descriptive Analytics.
Note that “What happened?” doesn’t just mean what happened last month. It also means what happened the preceding month, and the month before that, and the month before that. Said differently, understanding demands context. Good descriptive analytics over extended periods of time provide a key part of that context.
Level 2- Relational analytics: If descriptive analytics means asking “What happened?” relational analytics means asking “What else happened?” As its name suggests, relational analytics is about finding informative relationships in the data. For example, instead of simply reporting on average performance for a given role, relational analytics further asks how other factors such as education, experience, or leadership might also be related to performance. Relational analytics is about finding relationships that go beyond the basic descriptions to create additional questions.
Level 3- Predictive analytics: Predictive analytics asks “What will happen?” Ignore the jargon for now and just remember that predictive analytics really tries to predict only one of two possible things: a category or a number. Predictive analytics wants to say either “I know your kind” (category) or “I got your number” (number). If you can remember that then you understand the fundamental goal of predictive analytics.
3) How does predictive analytics work?
In a nutshell, data scientists take a subset of data and train a computer using “machine learning” algorithms to predict some outcome. For example, let’s suppose I am interested in predicting which registered nurses are likely to stay (stayers) and which are likely to leave (leavers) within the next year (an “I know your kind” prediction). To do this, I might first select some measures like years of experience and patient satisfaction that I think predict whether a nurse will be a “stayer” or a “leaver.”
If I have picked the right predictors, then my model should first learn to correctly predict the stayer vs. leaver status for the nurses in the training data. If my model does indeed predict who will be a stayer and who will be leaver, does that mean am I done? Nope!
Even if my predictive model works for the examples in my training set, I still don’t know how accurately it will predict stayers and leavers using a new set of data from a different group of nurses. As a follow-up then, I need to use my model to next see how well it generalizes to examples it has never seen before using a fresh dataset called the test set. If the model still successfully predicts leavers and stayers with the new examples in the test set, it’s a keeper. If not, we try some other predictors and start over. Predictive modeling for the “I got your number” scenario works the same way.
4) What do data scientists really do all day?
Popular accounts of analytics in action typically project an image of an all-knowing, all-seeing team of engineer-/data scientist-types churning out models that predict when you will wake, eat, drive, get sick, or get the mail.
The unsexy reality? Analytics people spend upwards of 80 percent of their time engaged in “data wrangling.” Also known as “data munging,” this is essentially taking raw, messy data (imagine raw HR survey results in different Excel files) and converting that mess into tidy columns and rows of properly formatted data appropriate for analyses. Conceptually, data wrangling is like an old-fashioned “cut and paste” session in Excel, only with a few million rows of data instead of a few dozen. Replace your mouse movement with some programming and, voila, you have “data wrangling.” Fun stuff.
5) I don’t do analytics. Why should I care about any of this?
Even if you don’t officially “do” analytics in your day-to-day HR function, there is a constant flood of data that directly impacts leader decisions and the bottom line. The better you can process and understand this information, the better off you will be.
First, just start by asking yourself “What level of analytics is this?” in your next data-filled meeting. This simple step will help you see the kinds of descriptive and relational insights that are not even being considered. It will also suggest some striking predictive possibilities. If you are a more informed and challenging analytics consumer, you will be a more valuable member of your organization.
Second, realize that you are now in a better position to truly collaborate with your analytics team members. You might not speak the same language yet, but you have at least eliminated some of the mystery surrounding “analytics”. Work with them, talk with them, and let them leverage both your expertise and theirs to uncover the talent insights you need.
Third, and relatedly, demand more, not less from your HR analytics team. And by more, I do NOT mean more reports. I mean thinking through a problem with them by asking what the outcome would look like at the descriptive, relational, and predictive analytics levels. Then ask “What would the next action be?” to insure you are digging into the actionable stuff that matters.
You may not know it, but many HR analytics teams are tired of producing static reports of limited value. Most would rather be showing off, providing the insights that drive better operational and strategic decisions, and perfecting the models that predict talent outcomes before they happen. If they can’t provide that, well, then find someone who can, and send them a clear signal that they need to bring their “A” game. In most cases, though, I suspect that after making your ambitious request, your analytics partners will be actually thinking to themselves “Finally, someone who knows what we can do and wasn’t afraid to ask.”
John Lipinski is a data scientist in human capital analytics. John is the co-founder of Ten Minute Talent, and HRanalytics101.com. He can be reached at firstname.lastname@example.org