In January 2018 , the Harvard Business Review ran an article titled “Why People Really Quit Their Jobs” written by Lori Goler, Janelle Gale, Brynn Harrington and Adam Grant. The case study was Facebook and why employees left that company. The main take – away was that people quit jobs ‘when their job wasn’t enjoyable, their strengths weren’t being used, and they weren’t growing in their careers’. At Facebook, people don’t quit a boss — they quit a job – which is often designed by their boss / manager. The managers who retain star talent often make the job roles flexible enough to accommodate the growth aspirations of the employee. As the debate on why employees quit and how to retain the top performers continues, lets look at how analytics and data science helps companies understand the attrition well and, perhaps, predict it too.
Defining what we seek around Attrition analytics is quite a challenge and we could be looking at answering any or all of the questions mentioned below:
- Attrition trends predictions – how much attrition can be expected month on month for the year? Can this be predicted by location / department. I need to understand my attrition data better.
- What have been the significant segments from which my attrition came in the last quarter? Some skills, levels and designations, locations, engagement score bands, time spent in the organization, etc. which show up as important segments in attrition.
- Is the attrition in each segment in proportion to the contribution of the segment in overall employee pool? Which are the segments that show dis-proportionate attrition
- What are the leading indicators of attrition? Which signs should serve as warning indicators that the employee is considering moving out ? What are the signs of dis-engagement?
- Do I have a Defend, Grow, Exit matrix which helps me understand which attrition is acceptable / critical ?
- For preventive measures and policies, do I have a Probability of Attrition scorecard which will let me know the chance of an employee attrition in the next 3 months / 6 months. If I run this scorecard on all employees each month, I can ensure early interventions for retention.
How we do it :-
We follow the process of DCOVA and I . Starting by defining the problem, we then seek to look at all the data available around the parameters and organize the data into a correct format. This is followed by a visualization and then the statistical techniques. Starting from Cognitive analytics which helps in understanding the data and trends and supplementing it with Descriptive statistics, we move on to understanding correlations and measures of associations (Segments which behave alike) . Finally, the Predictive analytics takes care of Forecasting, predicting How many , how much and Probability of attrition at an individual employee level.
Insights form a very important link where the statistical outcomes are de-coded and understood in common language and solutions suggested and debated.
Data requirement: –
Often, the presence of valid data is the defining and governing factor for the decision on which problems the analytics team will solve for. Also important is the HR / Senior management goal around attrition.
Projects worked on: –
- Trend forecasting (Month on month) for an ITES major. The forecasts were broken up by department and locations. Also, the conditions under which the forecast would not hold true were also articulated (through a causality model)
- Deep dive attrition MIS was created for a Services company to explore the changes in the Attrition trends across FY 15 and FY 16.
- A Probability of attrition scoring model was created for a banking unit to enable retention efforts before the actual attrition. Also, a Defend Grow Exit matrix was created to decide the force of interventions made.