How to Predict and Tame Attrition with Data Science

An organization typically invests considerable resources into recruiting, onboarding and nurturing a new hire. More often than not, we find that all this investment ends up counting for nothing when one fine day the employee chooses to quit!

Though tangible statistics on the loss of talent may be hard to come by, surely most people managers would agree that attrition (not churn) costs our economy billions each year.

Retention is one of the biggest challenges an enterprise can face in the intensive talent wars of today. Maybe, we have to start treating employees as ‘Internal Customers’. Recently, a key enterprise announced that their employees come before their customers! Infosys also announced an application towards predicting attrition.

A lot of operation research techniques are being applied to predict attrition. Many enterprises are running pilots and POCs using data science to reach at least 50-60% accuracy level in prediction. Any analytical model prepared in this direction should have a good learning capability and be capable of adding any trait based on response variables for that industry and domain.

Equally pertinent is spending time to define various data sources, factors, weightage, etc. Human emotions play an important role and is the major impulse. Continuous training of the model from multiple sources of data and inter-correlations is the minimum required for the model to take off. This also requires taking care in terms of usage of personal data (if any) with required permissions.

In some ways, this will serve as the judicial system to the current appraisal/performance management process, which is one of the major factors contributing to attrition.

The benefits to prevent attrition is multi-fold and can range from over-hiring, role deficiency, predicting cross-location movement, knowledge loss, building second-line in time, etc. and thus prevent the huge loss to business.

The purpose of this article is to have more thought-provoking ideas in the field of attrition analytics. Based on various studies, attrition rate for an organization of ~10,000-50,000 revolves around 15-20% and may peak up to 25%! And lead time to hire and onboard an employee is suggested to be three to five months.

Listed below is a matrix highlighting a few insights and measures to understand early in-cycle the customized push factors originating for such cases. This will also help in defining and streamlining the generic pull factors, as well.

Does a combination exist across Fast Trackers, Trusted Positions or Rockstars? Humans are the biggest asset, and a combination of the above categories could be just the growth injection an enterprise needs.

Forecasting attrition is a blend of employee behavior, social media, insurance patterns, location and skill analytics. These avenues already have enough pointers to suggest any potential drop-outs, but this is practically unfeasible for a human to keep tabs and correlate, thus, we need a smart analytics service. This will also take away any human biasing in the decision and evaluation process.

It would be great to hear more views. Feel free to share with me: akshey58 at hotmail dot com.

Article written by Akshey Gupta
Image credit by Getty Images, Taxi, Ezra Bailey
Want more? For Job Seekers | For Employers | For Influencers