How to use data science to retain top talent.

Decoding the digital footprint: how to use data science to retain top talent

You recruit, train, and nurture talent, and then one day the person resigns. 

Were there warning signs before they resigned?  Almost definitely.  

Were they slowly disengaging? Were their working hours stable? Did they respond to their manager with the same frequency? Probably not.

Then these are your signs.

In retrospect, these signs were obvious, but in real-time, it's difficult to respond to these types of changes in behavior.  In this article, we are going to explain the data science used to predict the likelihood of.  

The Digital Footprint

When an employee and/or a team interacts with a company's IT and Communication systems, such as email or Slack, they create a digital footprint that is unique to them. This digital footprint includes information such as the frequency and times of Slack messages and emails they send and receive, accumulated length and frequency of their meetings, and many others . Among others, the digital footprint tells us when the workday started, and ended, and how much time was spent on email and in meetings. 

Supervised Machine Learning (not as scary as it sounds)

In data science, there is a concept of Supervised Machine Learning.  In simple terms, this means that we train an algorithm to recognize an object or a pattern, and when the algorithm is shown something new, it decides whether it should be classified as the object that it was trained on.

Let’s use the example of different types of vehicles.

We first train the Machine Learning algorithm on how cars and airplanes look. For each vehicle we have, a picture is labeled as belonging to a particular category(sedan, SUV, sports car, commercial airliner, fighter jet, etc.). 

The algorithm looks at these pictures and tries to find patterns that distinguish one type of vehicle from another. For example, it might notice that sports cars tend to have a lower body and more streamlined design than SUVs or sedans, and that commercial airliners tend to have a wider wingspan than fighter jets. 

Once it has learned these patterns, it can then name pictures of new vehicles. When it is shown a picture of a vehicle it had never seen before, it analyzes the patterns in the picture and makes an educated guess as to what type of vehicle it is.

How this works for employee resignation 

In the case of predicting resignation, a similar method is used.  

We start with a list of historical employees who have resigned in a 24-month period.  The algorithms are trained on the employees’ digital footprint, analyzing about 150 different variables.  

The algorithm looks at variables for all employees and searches for changes in behavior similar to those exhibited by people who resigned.  It’s important to note that the algorithm does not look at each variable independently.  For instance, a person who has a tenure of 6 months and is exhibiting the same behavior as an employee with a tenure of 6 years, is likely to have a different classification.  

The prediction can be done at an individual employee level (based on the priva provided by the client) or at a team, division, or geographic level.  In the graph below, the analysis is done for various population groups.  It is valuable to understand resignation predictions that apply to specific population groups in companies that prioritize diversity and inclusivity. It is also important to identify the divisions in the company that are at risk so the loss of talent can be prevented.

Another prediction use case that is gaining interest is following the onboarding process of new employees. Proper onboarding has a direct impact on retention and by following the resignation risk of new employees, the organization can detect problems in the onboarding process.

Resignation is not a perfect tool because of the complexity of human behavior. Some resignations will be missed by even the most sophisticated of algorithms.  For instance, an employee can decide to resign because of an external factor (such as a spouse who needs to relocate for work).  

Furthermore, predicting the likelihood of an employee’s resignation does not mean that the is inevitable.  If we can identify that certain signs of disengagement or burnout are precursors to resignation, we can intervene to address the underlying cause of these behaviors. 

Ultimately, resignation predictions work best when they are incorporated into management practices to foster inclusive work environments where employees are recognized and valued for their contributions.