Employee Retention Case Study

Empowering HR Professionals to Retain At-Risk Employees

BACKGROUND

On average, employee turnover cost is 20% of their annual salary. Turnover is a major cost for many organizations due to the effort required to replace and train employees in addition to the impact on team productivity once an individual leaves. Assessing individual employee risk and understanding the driving factors are critical inputs to improving talent management and reducing turnover. EdjAnalytics and a national financial services company partnered to develop solutions to identify employees at risk of leaving the company and to further understand the factors associated with increased risk.

APPROACH

The client provided anonymized data from multiple systems for nearly 10,000 employees across the Sales and Operations divisions between the years of 2014 – 2017. Over 40 variables were analyzed including length of time with the company and within the role, performance reviews, job type and rank, compensation, and county of residence. In order to prepare for analysis and modeling, the Edj Data Science team cleaned the data and performed feature engineering to transform the data into more useful variables. Various model methodologies were then tested to best predict employee flight risk. The most successful models were the Random Forest Model (AUC = 0.73) and the Temporal Model (AUC = 0.78). (The closer the AUC is to 1, the more accurate the model is at predicting outcomes.) The output of the models provided a percentage chance prediction of how likely each employee is to leave the company in the upcoming quarter, and those predictions were translated into High, Medium, and Low risk categories to ease interpretation of the raw scores.

RESULTS

EdjAnalytics’ models demonstrated that years of service and salary were the most important factors in predicting turnover at the financial services company. Other indicators such as age, turnover rate by level, and variables relating to how quickly an employee has advanced through the organization also provided value in predicting which employees were most likely to leave.

Variable Importance

After determining that years of service was a primary variable in predicting an employee’s flight risk, EdjAnalytics revealed that employees with longer tenure were least likely to leave the company.

Turnover by Years of Service

Age followed a similar pattern with older employees showing lower turnover rates compared to younger employees. Annual salary, on the other hand, was more complex given its link to performance and tenure compared to other employees within the same role or level. When analyzing relative salary, those being paid at the higher end of their salary band were less likely to turnover.

To further illustrate how the models scored employees given the number of complex variables that must be considered, consider these two employees with customer service experience, Tim and Alex. (Their names and genders have been arbitrarily applied for illustration purposes.)

Tim vs Alex

Given Tim’s shorter tenure with the company, shorter time within the role, young age, and lower relative salary, Tim was considered to have a high flight risk of leaving. Alex, age 40, was considered to have a medium flight risk of leaving as she had been with the company for a number of years within the same role and was making a high salary for her position. Tim did, in fact, turnover while Alex was still with the company as of the most recent quarter of data.

NEXT STEPS

The client is in the process of integrating the Edj models within their internal data science and human resources teams and testing the efficacy of interventions in reaching retention objectives. Potential future enhancements to the models include expanding understanding of the employee’s socio-economic status using external data sources and adding manager performance details such as review ratings and turnover by manager.

EdjAnalytics and the company are continuing to focus on additional key initiatives including customer attrition and sales forecasting.