Predicting Employee Attrition Using Big Data

 

2 weeks ago, the HR Head asked me a question – I want to know which employee is going to put in his papers – and when.

 

It was one of those rare moments when i was completely, totally stumped.

 

Here is a partial answer- How to use Big(and small) data together to predict employee turnover.

 

Factors that impact turnover

Lets start at the very beginning. What makes an employee quit?

 

When I conducted Exit interviews at our small 700 people IT organisation, i wouldnt ask them, “why are you leaving?” I would ask, “Why did you start looking?” I wanted to understand where the distance began, and why. The resignation is not what we are investigating. That action is the result of a disengagement that began weeks, months, even years before the actual resignation.

 

We are investing that disengagement. And the probability of its resulting in a separation. Two different things.

 

How do we measure something as intangible as disengagement?

 

I believe we may have some ideas here.

 

Pointers to Disengagement

 

• Employee Satisfaction Scores

This one is apparently a no brainer. Yet I am suprised to see that most ERP packages dont have a place to store the employee’s engagement score and compare that year on year. Then check for its correlation with their performance ratings and other behavioral actions. There are pointers there.

 

• Social Media Activity

This is where big data comes in. Have an internal IM program and an internal social media platform like yammer or internal discussion boards? Let your Big Data analysts do quick calculations on how often and with how many colleagues the IM was used. And how often the social media platform and discussion boards were used. Engaged employees will use more “connection points” to connect with the organisation and its people.

Notable Exception: Introverts. Introverts are people too. And they wont use Social Media. The end of this post says “Its the pattern, not the static data.” Read that section to know more.

 

• Meet the Parents

This is one of my favorite metaphors. When they bring the family, they are engaged. No exceptions. This also can be automated. Attendance at the family events is automated and can be fed into the giant supercomputer for automatic analysis. If you know when they stopped bringing the family, you know when they started thinking out.

 

• Access Card Patterns

Another big data beauty which needs individualised reading to make sense. How often was the access card used to go in and out? Whats the pattern? Has it changed lately?

 

Which access cards are used together? Are the breaks in groups, with one or two friends, or alone?

 

• Use of development resources on the learning portal

What kind of courses are being accessed? What was it earlier? Is it consistent with the expectations of the current role? What is the usage pattern?

 

• Correlate with Performance Ratings and the moneys

The higest risk categories are employees who have recently witnessed a fall in the rating, or whose difference from their maximum potential earning is very high. Let me explain. Suppose Mr. Alpha is paid INR 100 at the highest paying company. You are at the 60th percentile as an organisation, so you pay INR 60 for the same profile. But suppose the actual salary of Mr. Alpha is not INR 60, but INR 55, because of your internal compa ratio adjustments. Which means that Mr. Alpha is at a 45% discount from his maximum earning potential. That kind of gap is not sustainable.

 

And lastly, remember, its the pattern, not the static data. Big data will, over a period of time, establish patterns of behavior for each employee. When this pattern of behavior changes in a perceptible way, and for a consistent period, you know you should care enough to investigate more.

 

Does disengagement always result in attrition? Is it worth bothering with if it doesnt lead to attrition? What are some of the other pointers that can be used to arrive at behavioral disengagement? Anything we have missed out in the article above?