Despite the title, this blog is not primarily about baking. (Sorry!) Rather, it’s based on my personal observations as a data scientist on how our thinking and approach to People Analytics can make or break our chances of success. I found inspiration for this blog when I saw my wife working on the preparation for a wedding cake.
It struck me how alike her methods in reviewing and revising a recipe are to my methods in shaping an analytics case into something feasible. It makes sense: in both cases, you’ll identify snags that you will need to tackle to get results. It drove me to dive deeper into the analogy: what can we, as HR professionals, learn from baking cakes?
Let’s start from the beginning with an example. There was a situation – a wedding – from which a specific need or wish arose: a wedding cake. Of course, the bride and groom had specific requirements and desires. They brought images to illustrate their vision for the cake. This is where the first obstacles present themselves. My wife took one look at the pictures and sighed: “Fresh raspberries? No, those would be out of season.”
How often have you seen organizations turn to an analytics team with great questions, if only there had been data to answer them? For example, an organization struggles with increased attrition rates among client-facing employees and worries about the impact on revenue. They want to know why these people are leaving. Unfortunately, exit interviews are conducted sporadically and reasons for leaving are not recorded well in systems.
Would proposing an analytics case under these circumstances be parried with “don’t bother” in your organization? In our analogy that would mean never getting a wedding cake, for lack of specific ingredients.
“Those raspberries look great and form a nice contrast to the dark chocolate. I could use seasonal fruit that is alike in color and pick different flavors. Or I could use a raspberry gel inside the cake, keeping the flavor and still work with chocolate.”
My wife conferred with the bride and groom: did they want the out-of-season raspberries for flavor or for looks? They opted for both different flavors and another look with seasonal fruit. The final cake was unlike the original vision, but the newlyweds were ever so happy with the outcome!
Keep an open and creative mind
Have you ever heard a great question, followed by a general consensus that the answer will always be “out there” and an analytics case would be futile? In my experience, that conclusion is generally reached through one false assumption:
“If we cannot answer the exact, original question, we need not bother.”
When I am under the impression that a case gets discarded prematurely because of the above, I tend to ask two questions:
- Can we get by with part of the picture?
- Can we use a proxy?
In other words: do we really need that exact cake, or are we still happy when we (1) have only specific parts of it or (2) change a few ingredients to render it feasible? Added bonus: the answers tend to reveal how badly results are needed. After all, if you prefer to dream your impossible cake over having a real one, how big is your need for cake to begin with?
Often, we will have to improvise, adapt and (hopefully) overcome the impossibility of a question and get to a feasible case that caters to the underlying need. Back to our HR example: let’s say we have an issue with attrition and want to improve retention. The question asked is: “Why are our employees leaving?”
The people analytics team answers: “We would love to help you, but we do not have reliable data on reasons for leaving. Exit interviews are not structurally conducted nor recorded organization-wide. However…”
Can we get by with part of the picture?
Do we need to know all the reasons for leaving across the entire workforce, or can we use the available data to some extent?
We do not fully trust the data on exit interviews, but at least we have something that might show recent changes in reasons for leaving. Sudden changes in distributions over time could tell us when the issue started. Underwhelming? Maybe, but it could be a start to ask more informed questions.
Still, by itself, this could give us results that are inconclusive due to missing values. In other words: we have parts of a cake that do not add up to anything we would consider a cake.
Can we use a proxy?
Perhaps the most relevant question in our example case is: can we use a proxy? Many academic publications provide strong indications of (cor)related factors that could be used instead of the data we thought of first. And sometimes an even simpler approach may suffice. For example: if we show common characteristics of who is leaving, could that answer enough of why they are leaving to take action?
We have many options at our disposal. We can investigate department, tenure, time under a manager, size of team, salary, performance scores, employee engagement, and more to derive profiles analytically. Such profiles may imply reasons why people leave. Zooming in on the reasons for leaving related to those profiles may uncover themes. And if not: could it be that absence of such reasons for these profiles is a clue by itself?
And why limit ourselves to standard HR data? We could consider other ways, such as looking into communication with Natural Language Processing or Organizational Network Analysis.
As icing on the cake, we can conduct interviews, informed by our analysis results. This way, we supplement data with expert views and fill potential gaps.
Concluding: a taste may suffice
Generally, the results you get using the approach above, even intermediate results will provide useful insights. And that, in turn, sets things in motion to tackle the issue at hand.
Think about the need and how that may be answered rather than the question asked. It is too easy to disregard a question because it cannot be answered in full or exactly as phrased. In my experience, the underlying need can often still be catered to using an incomplete picture or a few different ingredients.
In short: do not just follow recipes, but improvise. Result trumps recipe. Chances are that once your organization gets a taste, you’ll still be catering to its needs while building up that analytics appetite.