Predictive models in Human Resources?

In recent times, predictive models have become increasingly popular among businesses worldwide. The allure of being able to anticipate future events, detect trends and patterns, and make data-driven decisions has prompted businesses to embrace this technology. With the ability to plan and budget based on this information, companies can gain a competitive edge in their respective industries.

What is Predictive Modelling?

The Gartner Glossary defines predictive modelling as a statistical method that enables users to anticipate future behaviours by analysing past and present data and developing a model to forecast future results. Predictive modelling involves gathering data, creating a statistical model, making forecasts, and assessing or adjusting the model as new data becomes available.

There are several different techniques one could use to lift insights from data; these are just 2 popular techniques:

1.    Linear Regression- When two continuous variables show a linear relationship, a linear regression tool can be used to determine the change in a dependent variable based on the independent variable, this model is used to assess risk in the financial services and insurance industries; For example, In the credit card industry, a financial company may seek to reduce the risk of default among customers in their portfolio. To achieve this, the company intends to identify the five main factors that lead to default. Once these factors are identified, the company can develop appropriate interventions to address the specific needs of high-risk customers and minimise the risk of default. (Dusane, 2014); There are other forms of regression that are equally as useful under different circumstances for example Logistic Regression and Multiple Regression

2.    Decision Tree: A decision tree is a tool that facilitates decision-making by presenting a series of tests that branch out, leading to a specific path that matches a class or label. Think of it as a flowchart of if/else statements that guide you towards a desired outcome (Chamblee, 2021). As an example, consider Gerber, a renowned baby products company that faced a complex situation in 1999. The company was confronted with a report that a potentially harmful chemical, “phthalates,” was present in their products. With pressure from the Consumer Product Safety Commission and the national media, Gerber had to act swiftly and decisively. They employed a decision tree to determine whether to recall the products or wait for the results of an investigation. This decision-making process allowed Gerber to map out the probability of a specific occurrence and the corresponding estimated monetary gain/loss. Although the chemical was eventually deemed safe for use, Gerber wasted no time in devising a plan. The decision tree proved to be an invaluable tool, enabling Gerber to make informed decisions quickly and efficiently (Marovellio , 2021)

Using techniques like these together, led to the development of probably the most famous predictive model of all time, the weather forecast.

Predictive Models in HR

Predictive models have a significant potential to revolutionise Human resources departments globally, by enabling companies to make data-driven decisions about their workforce. These models can be used to predict employee performance; identify high-potential employees; and assess the success of recruitment efforts.

For example, in recent years Xerox utilised predictive modelling to improve employee retention rates. The company used data to identify factors that were likely to lead to employee churn or turnover, such as job satisfaction, distance from work, and tenure. By analysing this data, Xerox was able to identify which employees were most likely to leave and create targeted retention programs to address their specific needs, reducing turnover rates by 20% (Feffer, 2014).

In another case, In an effort to determine the impact of engagement on store sales, Best Buy, a renowned electronics retailer, utilised predictive analytics and models to examine their data and uncovered that a slight increase of 0.1 percentage point in engagement resulted in a $100,000 increase in revenue per store. This discovery prompted Best Buy to regularly monitor engagement levels and pinpoint the factors that drove engagement. By doing so, Best Buy was able to introduce tailored HR strategies aimed at enhancing engagement, which in turn led to a notable increase in-store revenue (van Vulpen, n.d.)

HR is home to large amounts of people data and by analysing that data in a directionally meaningful manner they will be able to lift insights that could drive business performance, increase NPS and solidify HR’s position as a business partner.

Pitfalls

As the common adage goes, not all that glitters is gold, and like most things, they are potential pitfalls and cons to using these models in HR.

1)    Access to data: The creature that is the predictive model feeds off of data, the more data is fed into it the stronger and more accurate it will become. However, not many organisations can create robust predictive models for HR purposes, as per Deloitte’s 2018 People Analytics Maturity Model. In 2018, only 17% of organisations around the world utilised HR data. Out of this 17%, only 2% had business-integrated data, meaning they use real-time, advanced AI-aided tools to collect, integrate, and analyse data. The remaining 15% can only perform predictive analytics on an ad-hoc basis (van Vulpen, n.d.), however, this number is growing year on year and early adopters are already demonstrating some interesting results.

2)    Biases: The models will produce to the data they are trained on. If the data used to develop the model is biased, it will produce biased results. This can lead to discrimination against certain groups of people. For example, a company’s past data may show a higher proportion of males in leadership roles compared to females, which can result in a model that is biased towards males, if this data is used to predict future leaders or to select a leader it can create serious issues. A way to work around this is to up-sample the minority class and down-sample the majority class, this may help increase accuracy. (The Human Capital Hub, n.d.)

3)    Counterintuitive action on the back of results: For example, if an employee is deemed to be a flight risk, instead of developing interventions to help retain them, they may be deprived of growth and learning opportunities by management who see the employee as a person who is going to leave anyway, and in that way, it becomes a self-fulfilling prophecy.

4)     Ethical concerns: The use of predictive models in HR can raise ethical concerns around privacy, data protection, and fairness, especially when the models are used to make important decisions such as hiring, promotion, or termination.

5)     There is a high chance they are wrong: “Putting too much trust in data will lead to problems. We all know that all models are wrong, but there are some that are useful. The useful ones are those that use data that represents reality well. But representing reality well may not be enough when it comes to HR issues such as diversity and gender balance.” (The Human Capital Hub, n.d.)

In summary, predictive models are a valuable tool for businesses to make data-driven decisions and gain a competitive advantage. Predictive modelling can also revolutionise HR departments by predicting employee performance, identifying high-potential employees, and assessing recruitment success. However, they are far from perfect and users must consider potential pitfalls.

This isn’t a call for HR to go build models but rather a gentle nudge for HR to begin thinking about how can they ask the right questions and what can the answers they get tell them.

A point to remember is that in your context, historical data may be a lot more accurate than a predictive model, e.g. employees most commonly leave after 2 years of joining the company, if that is an identified trend, begin to leverage and build on that information and make strategic decisions on the back of it, if the weather forecast says today is going to be sunny, but there are clouds in the sky and it has been raining all week, it would be wise to take an umbrella, but even that decision is a data-driven one.