A manufacturing company was struggling to attract and retain employees in a manual production role. To support business growth objectives, the company would need to hire approximately 900 people in this role over the next three years. Owing to a 25% attrition rate for this role, replacement hires made up a large proportion of this target. The cost of replacement was about R129 500 per employee – R48.6 million a year.
This presented a high-impact opportunity to apply workforce analytics and demonstrate significant business value. The Mercer team summarised the premise of the problem as follows: Can we apply predictive analytics to identify high-risk attrition targets so the organisation would have enough time to determine if a high performer was worthy of a retention effort?
A common challenge to applying workforce analytics is the misapprehension of ‘not having enough data’. But the question of data sufficiency can’t be answered without actually conducting the analysis. Only then can we determine whether the results can be generalised beyond the sample.
In this case, the company had a human resource information system (HRIS) which had been storing employee transaction data since day one. The team took three years’ worth of data, such as hire and exit date, pay changes, manager changes, job changes and compensation, to conduct the analysis.
Using this data extract and additional insights gathered from employee focus-group discussions and manager interviews, they applied a disciplined data science approach to develop several predictive models of varying complexity and accuracy.
Model predicts employee churn with 83% accuracy
The best model had an 83% accuracy rate in predicting which individuals would leave two months before their actual resignation.
It’s typical to find these models have a lower predictive power in reality than in simulations. But even at a 30% accuracy rate, the organisation would be able to save close to R3 million a year through reduced attrition.
Interestingly, the model with the lowest accuracy in predicting churn revealed the most compelling features leading to an employee’s exit from the organisation. These insights could be used to update HR policies and programmes in order to address this.
Model identifies determining factors for employee attrition
- Employees who took more days to complete probation were at greater risk of quitting.
- Anyone who took more than six months to complete probation eventually quit (nearly 50% of total attrition).
- Significant costs could be avoided by terminating employment earlier (e.g. by having a maximum time to complete probation).
- As pay increased, the likelihood of quitting reduced (you don’t necessarily need to pay top dollar for retention; just enough will suffice).
- The optimal range for scaled pay appears to be between -1.0 and 0.5 (scale pay is compensation moderated by tenure within the organisation).
The organisation’s leaders saw the business value that could be generated through solving the attrition problem. By investing a relatively small number of resources, they could solve a large business problem – all thanks to the application of workforce analysis.
Workforce analytics is a learning journey
Technology is providing huge opportunities to advance the way we gather, combine and analyse data about people at work. We now have exponentially more data (and, more importantly, better access to data) than ever before – both structured and unstructured. However, we don’t yet know how all this data will help us.
Workforce analytics is an experiment, and not all workforce analytics projects will lead to solutions. But failures will lead to insights and actions that will further our ability to harness value from data. An empowered workforce needs good data to drive decision making.
Just as marketing data and buyer insights are leading business transformation efforts, talent insights have the potential to deliver accelerated success on the people agenda – both to enhance the employee experience but also to drive better decisions around how to retain and engage people.
But are companies getting these insights?
Compared to last year, companies – both globally and in South Africa – didn’t move the needle an awful lot on the type of analytics that HR is able to provide. Very few report being able to translate data into predictive insights, and nearly one in three is still only able to produce basic descriptive reporting and historical trend analysis.
Companies in the financial services and logistics industries globally are ahead of the curve but still have a long way to go. Even with all of the data that is being collected, senior executives globally are not getting the kind of talent analytics they need to make business decisions.
For example, executives say that understanding the key drivers of engagement would be most value-adding, but only 35% of HR leaders globally are able to provide this information. This is especially surprising given that most companies today have at least some sort of engagement survey in place. Predictive analytics, such as identifying which employees are likely to leave in the next six to 12 months, are even less common and productivity outcomes even less likely.
HR and employees recognise that the disconnect may be due in part to a capability gap — both groups ranked data analytics and predictive modelling among the top three in-demand skills for the next 12 months.
The risk of not leveraging talent data is especially acute when there is so much organisational change on the horizon. When decisions are informed only by financial and marketing data, there can be unintended people consequences. For example, the World Economic Forum’s Future of Jobs study1 found that “women are at risk of losing out on tomorrow’s best job opportunities” as disruption and displacement are likely to occur in job families with the largest share of female employees. When HR is able to partner with the business to facilitate an evidence-based decision-making process, they help to mitigate these risks and ensure that the talent implications are being considered, especially when companies are making structural changes that will see great talent – along with the roles – leave.
The missing age dividend
The issue of increasing longevity and whether retirement ages have been set too low would seem like a low priority issue for South African employers and policymakers. Surely addressing the crisis of youth unemployment warrants greater attention? The problem here comes when we see these two concerns as mutually exclusive. By promoting the needs of youth at the expense of the aged, we ignore the interconnectedness of intergenerational dynamics. We prioritise because we believe limited resources dictate prioritisation, but occasionally we find that suppressing the interests of one group only exacerbates achieving the needs of another.
What follows is a discussion of the missing age dividend. Framing our understanding of the value of older workers should give employers pause for thought – and a deeper appreciation of what stands to be lost.
Taking age off the table when deciding who should work where
As we’ve seen, making HR policy decisions based on hunches and intuition can result in some disastrous business choices. With transformation a priority in South Africa, employers tend to assume the most obvious way to achieve it is to have older employers step aside to create opportunities for a new generation of workers. On what basis have we made that assumption? The power of HR data analytics is that we can test these assumptions empirically.
As we discussed in last year’s Benefits Barometer, studies from Scandinavia and Germany are challenging these conventions by suggesting that productivity and, by extension, job creation can actually increase when older, experienced workers are retained. But why rely on secondary information when we can do the research ourselves?
