Special thanks to my student Harun Behroz for his great work on the modeling with SAP Analytics Cloud Smart Predict.
In most lines of business and across different industries, it’s standard nowadays to use advanced analytics to improve business processes. However, in human resources (HR), the potential of the collected data to generate insights is not even considered. And even worse: some companies still export their data to Microsoft Excel to calculate their women’s quota! Here is definitely a chance to simplify the process with an analytical dashboard that is based on live data and shows important HR key performance indicators.
The wealth and quality of HR data could also be used for more advanced analytics beyond simple reporting that only looks at past events. What about looking into the future and acting proactively on the issues HR is facing?
Let’s take a look at the issue of employee churn.
The Challenge of Employee Churn
Companies want to prevent their highly skilled employees from leaving the company. Today, many companies rely on the gut feeling of their managers to foresee employee churn. But this method is neither precise nor free of bias.
Think about some managers saying that their best guess for a high churn probability is an employee’s new haircut. A change of a haircut might indicate that a person is ready for more important changes. But could we really deduce from one factor an employee’s decision to leave in today’s complex world?
By considering more data and more complex patterns, we can make better decisions and we might even be able to find the ones leaving the company without changing their haircut.
I know there are people who argue that human behavior is too complex to be reflected in data sets. But I don’t prefer one approach over the other—I’m suggesting using both views: the opinion of the managers as well as the analytical and mathematical conclusions from the data.
The combination of both could lead to new insights. We could combine the empathy of the managers with the more complex patterns detected from the data. Consider:
- A high churn rate among new hires could indicate that the onboarding process needs to be improved and thus lead to a reflection of the company culture and a change in the treatment of new hires or young people.
- We could also have a closer look at internal employee turnover and their success rate or just make better learning recommendations based on data.
- Furthermore, data could help to reflect if we do our best to make the most out of the potential of experienced colleagues or by detecting high potentials.
Tackling Churn with Data Analytics—a Use Case
Let’s have a look at an example of how this can be achieved.
Last year, HR use cases became quite popular and one of my students, Harun Behroz, got interested. He started simple with building dashboards for HR since he isn’t a mathematician and had no background in data science or statistics at all. He was really motivated and was also looking for new challenges.
Just to show him what else you could do, I introduced Harun to
Smart Predict within SAP Analytics Cloud, which enables students and colleagues with an affinity for statistics to do their own predictions, even if they aren’t familiar with mathematics and algorithms.
He immediately fell in love with the tool and it gave him the long missing opportunity to find new insights in the given data. For example, he was now not only able to visualize how many people left the company last year and what were the statistical separators of these two groups,
he could also predict the churn score for each employee.
This is great, because when we first started working with HR use cases, Harun said that this BI analysis generalizes too much since each human behaves individually; he believed only looking at the statistics of the two groups might not give respect to this individualism. Sure, he argued, there are groups of people that behave similar, but there are definitely more than two.
This was one of the reasons I enjoyed working with him so much, since he always critically questioned everything. Also, with the predictions we made, he of course pointed out that it will not explain and show all influencing factors and not explain all the human behavior.
Let’s be honest: sometimes we just randomly take decisions, or we take decisions based on factors that were not described in the data. I can’t go into details about the influencing factors since they vary a lot, but there was one factor that was quite interesting. We call it the midlife crisis in a company.
A Common HR Factor—the Midlife Crisis
We found a certain point in tenure where the churn rate increased radically and dropped again afterwards. So, it looked like the employees started reflecting, after being a particular number of years with the same company, on whether they wanted to leave or whether they would stay with the company until they retire. It is sometimes good to know such things (I, for example, keep reminding my manager that my company midlife crisis is yet to come).
My super motivated student was keen on finding more influencing factors that describe the churn rate even better. Unfortunately, we couldn’t get all the data we wanted. In some cases, there have been restrictions and in other cases the data wasn’t appropriately collected and stored. Although this is an issue in many projects it is also a valuable finding, since it allows companies to improve the data mining process for future analysis and it shows that in most data science projects, data collection and model building are iterative processes.
One Prediction Begets Another—Building the Model
Of course, just doing one prediction was not enough: once you know how it’s done, it is easy to apply the same approach to a similar use case. This is what my student did. In the end, we had a nice model that predicted the churn rate of each employee.
Furthermore, we could predict a score for how quickly an employee would be successful in his new position after changing jobs internally. The idea behind the second use case is to help people who make a less common move to a new position that requires quite different skills.
I am afraid of giving a general example here since of course, it’s different for each move and each company. I can say that when I switched my position from development to presales I experienced that I had to learn a lot of new skills. But utilizing the data to reveal these insights lets HR plan accordingly. One way to support them could be to assign an experienced colleague. The other idea is to look at people who made a similar move and learn from their training history and for example suggest courses which will ease the transition.
Bottom Line
In the end, we had two great models and all we needed was a super motivated student, the right tool, help from the HR colleagues with business and process knowledge, and of course data.
So why not get started and make more out of your HR data?
Learn More
- Visit the SAP Analytics Cloud info page
- See my Hands-On-Tutorials if you want to try SmartPredict on your own:
