Common Sense Analytics

KISS (Keep It Simple Stupid)

When I read job postings for people analytics roles, I'm a little surprised by how much emphasis is placed on advanced modeling and machine learning. The ads often prefer or require someone with a Ph.D., even for organizations that are just beginning their people analytics journey. There are certainly a number of companies already doing advanced predictive and other types of modeling, and they are blazing trails for the rest of us. However, there are also many organizations that have been able to improve business performance and create competitive advantage through people analytics using basic mathematics and a lot of common sense. Simple can be sexy, too.

Explaining People Analytics in a Business Context

During an interview last year, someone posed the following scenario to me:

You're presenting results to a group of managers without strong math backgrounds and you need to explain Bayesian statistics. How would you explain it to them?

I told  him that if I found myself in a meeting with managers where I had to explain Bayesian statistics, then something had gone terribly wrong. I explained that even the most advanced mathematical model has to be reduced to a story we can tell about the data in the context of the business. If the model requires advanced mathematics to be understood and it can't be explained in business terms, it's time to create a new model.

I don't think the interviewer was very happy with my response. With the kind of analytics he's involved with, researchers might not completely understand why something works, but that's acceptable as long as the model is reliable and accurate. With people analytics, understanding why a model works is paramount; accuracy is often an illusion. (I didn't get the job)

Even if you have the most talented mathematician in the world on your team, and he creates the most cutting-edge, deep-learning neural network, it's not going to get you very far unless you can explain it in business terms that supervisors and managers can understand and relate to. In my experience, especially when you're starting with people analytics, some of the most insightful and meaningful analyses you can perform are what I call "common sense analytics." You begin with a question and determine how the data might answer it. You slice the data to look at it in ways that invariably lead to other questions, and you repeat the process. The data reveals a story as you refine your questions to uncover it. Or, perhaps you create a solid model based on machine learning algorithms, and now you have to translate that into common business sense. Regardless of the path you take, the data has to tell a story, and you need to put the story in the context of the business.

The skills required to translate math and statistics into business reality are advanced analytical skills, but they are different from the advanced analytical skills required to fine tune the parameters of a model using gradient-based optimization or to classify data using discriminant analysis. If it turns out that you need the latter, you can usually rent or borrow the technical expertise at least temporarily. Good story tellers are harder to find.

Painting a Picture of What Success Looks Like

The end of your story needs to paint a picture of the future when you've made a measurable improvement in business performance. The manager or supervisor has to see that the value of the end result is worth the time, effort and disruption to the business that it's going to take to implement this change. Finally, you have to provide supervisors and managers with everything they need to implement the change, including focused information that's directly tied to monitoring their progress. Resist the temptation to provide unrelated data because it will only divert their attention and add noise to the signal.

I think this was also the point of a recent article in Harvard Business Review that many have probably already seen. It's titled, "HR Must Make People Analytics More User-Friendly," and it has advice to help organizations organize and communicate these kinds of results for more effective change implementations.

MOTS: "Explainability" and "understandability" trump accuracy (which is probably just an illusion anyway).

Leave a comment! Let me know what your experience has been or how you've been successful.



Originally posted on HR Lens.