How to move from descriptive to predictive analytics

According to the 2018 Global Human Capital Trends Report from Deloitte, 84% of respondents viewed HR analytics as important or very important, making it the second highest ranked item in the list of top HR trends. Importance attributed to HR analytics has been growing rapidly in recent years and more companies are moving towards making it a core component of how HR operates.

HR analytics in its essence is being data-driven when it comes to people management and HR decision making. When executed well, it will touch, and improve all areas of HR activity from recruitment, on-boarding, development and retention of talent.

Let us explore how analytics can help the HR department of a mid-sized company become a strategic partner to its overall success. “XYZ Inc.” is a research and development company. In the following example, we address one of its key priorities – retaining its highly skilled workforce.

Attrition at XYZ Inc.

Attrition is a pressing concern for XYZ since it is very expensive and time consuming to hire skilled scientists and high performing sales professionals. When a scientist leaves the organisation, they also take with them substantial knowledge and experience. There is also significant time spent on on-boarding new sales employees making it expensive to replace them. It is key to retain top talent for the success of the organisation.

Turnover rate at XYZ Inc. was 16% over the previous 12 months. This metric is often reported widely and tracked regularly as an indication of how well the company is able to hold on to its top talent.

Analytics 1.0 – Descriptive Analytics

Analytics begin when we start to tell a story with the data available to us. A story that can inform strategy. Let us create a custom dashboard for XYZ Inc. focused on turnover so that more patterns emerge:

Fig 1: A sample custom turnover dashboard built for XYZ Inc.

Fig 1: A sample custom turnover dashboard built for XYZ Inc.

When we slice the metric by different dimensions, more patterns emerge. Using such a dashboard, the HR leadership has more context to develop retention strategies.

To take this to the next level, XYZ Inc. can tell a better story about the state of their attrition by leveraging all the data it has or can capture. For example, HR business partners routinely conduct exit interviews with departing employees and capture their notes in Excel sheets. As this is done more from a compliance standpoint, this data is almost never analysed. Analysing it is also not straightforward as it is mostly text. By making improvements to how exit interview data is collected and employing techniques such as sentiment analysis, we can help XYZ combine this data to other data from the HR system to generate more insights on attrition.

Bringing all these streams of data together, XYZ is in a better position to understand the causes of attrition in the company and is now able to devise retention strategies specific to problems faced by different departments and groups of employees at the company

A major drawback of analytics 1.0 is that even though it combines data from different sources to tell an insightful story about attrition, it remains reactionary. Although retention strategies that rely on analytics 1.0 will be more specific than generalised ones based on turnover rates alone, predictive analytics can make the HR organisation more proactive and nimble.

Analytics 2.0 – Predictive Analytics

To understand the combined effect of variables such as tenure, age, monthly income, salary hike etc. by eye from a dashboard is extremely difficult. However, we can put advanced analytics techniques to work, to help us understand these complex relationships and build an early warning system for attrition.

There are a variety of machine learning algorithms to choose from and the HR analytics team should choose a model that works best for the business problem. At XYZ, a Random Forest classifier algorithm was deployed on the available employee data and we get a glimpse of how different factors play at varying degrees into an employee’s decision to leave.

Fig 2: The different attributes used by the random forest model by their importance in predicting attrition

Fig 2: The different attributes used by the random forest model by their importance in predicting attrition

This model is able to predict if an employee is an attrition risk with accuracy close to 90%. Below (Figure 3), we show an example of how a model like this can inform line managers of various departments about the number of employees in their teams who are at high risk. For example the manager of the R&D team knows that 20% of employees in his department who are likely to leave are high-potential (HIPO) employees.  XYZ Inc. can now use this early warning system to enable its line managers to proactively devise intervention strategies for at-risk employees to reduce the likelihood of attrition. At the same time, we will keep fine tuning the model and letting it learn from future data.

Fig 3: Number of employees at risk of attrition by department and high-potential (HIPO) employees

Fig 3: Number of employees at risk of attrition by department and high-potential (HIPO) employees

It is useful to see how XYZ structures its processes around managing retention of its high performing employees (Figure 4). The model is deployed throughout its workforce and is constantly updated so as to identify at-risk employees at each department. Line managers and HR monitors the output to identify high performers who are likely to leave and is in a position to develop personalised retention strategies to retain them.

Fig 4: How XYZ uses the retention model

Fig 4: How XYZ uses the retention model

From a point of applying textbook retention strategies, being data-driven has enabled XYZ to be more intelligent in terms of managing employee turnover. Starting with identifying what data is captured and understanding how insights can be drawn from it, HR organisations can move up the maturity curve starting with reporting, descriptive analytics and predictive analytics. Data-driven talent management is HR’s seat at the strategic table. The time to embrace HR analytics is now.

Online training on People Analytics

If you are looking to learn more about how to get started in people analytics then check out the myHRfuture academy online course on people analytics titled 'An Introduction to People Analytics' that is taught by David Green and Jonathan Ferrar. It's a great introductory course for anyone interested in learning more about people analytics.



Sunil Meharia is a Data Science specialist with a deep interest in how technology can transform Human Resources. He has over 8 years of experience across the Public sector, large manufacturing and research-based conglomerates. His primary focus has been on data analytics in human resources, business strategy, and digital transformation.