Predictive Analytics in Human Resources: Tutorial and 7 case studies
Predictive analytics is an upcoming trend in Human Resources (HR). Recruitment tools predict high performers, and increasingly companies are able to predict which employee is likely to leave. In this article, we will explain what HR predictive analytics are and how they can be a real game-changer for HR departments.
We will also discuss seven real-life examples of predictive analytics in HR, two of which are detailed case studies.
The logic behind HR predictive analytics
Do you know what your personal credit score, the Oakland Athletics baseball team manager Billy Bean from the movie Moneyball and your Match.com profile have in common? They all combine big data and predictive analytics in order to predict the future.
Predictive data analytics are everywhere. It is in its essence a technology that learns from existing data, and it uses this to forecast individual behavior. This means that predictions are very specific. In the movie Moneyball, predictive analytics were used to predict the potential success of individual baseball players.
So how do these predictive analytics work? Predictive analytics involves a set of various statistical (data mining) techniques that analyze historical data and outcomes. These techniques then try to create a formula, or algorithm, that best mimics these historical outcomes. This algorithm then uses current data to predict outcomes in the future.
Related (free) resource ahead! Continue reading below ↓
51 HR Metrics cheat sheet
Data-driven HR starts by implementing relevant HR metrics. Download the FREE cheat sheet with 51 HR Metrics
Predictive analytics in practice
Say there is a playground next to your house. For the past two weeks, you wrote down if there were kids playing on the playground or not. You also wrote down if it was sunny, rainy or cloudy, the temperature and the humidity. Based on the data you collected, would you be able to predict if kids will be playing on the playground on a specific day?
This is a tricky question. Obviously, these weather conditions have something to do with whether kids are playing outside or not. If the weather forecast is rainy, it will probably rain, meaning that kids are less likely to play outside. When it is hot, kids probably will play outside. But does your spreadsheet with information of fourteen consecutive days hold sufficient data to make an accurate prediction on whether or not kids will play outside?
Making predictions using a decision tree
A common and rather simple method of creating a predictive model is the decision tree. A decision tree is a tree-like model consisting of decisions and their possible consequences. In the decision tree, every node represents a test on a specific attribute, and each branch represents the possible outcomes of this test.
I made a decision tree on our weather data set by applying some simple data mining techniques. The decision tree was computed using a specific decision tree algorithm, called C4.5. This decision tree model fits the data well: it is able to predict whether kids will play on the playground with a 71% accuracy. This is much better than guessing, which has a 50% accuracy.
The decision tree is self-explanatory if you take a closer look. I asked our designer to make this a pretty picture – the original output is a much more boring black-and-white version.
There are two strong predictors in the decision tree. The outlook is the first predictor. Kids will play on the playground 4 out of 5 times when the weather outlook is sunny. When the forecast is rainy, the kids do not play outside. In case the outlook is cloudy, humidity is the second predictor. Kids are not likely to play outside if humidity is high (which it usually is when it rains). However, when humidity is normal, kids are likely to play outside.
In other words: the weather forecast and humidity can be used to predict whether kids will play on the playground outside accurately. Of course, in an organization, you’re not interested in predicting these specific outcomes. Rather, you will use HR metrics to predict business outcomes.
This simple example is an excellent showcase of how predictive analytics work. It all boils down to algorithms that learn from existing data to make predictions about the future. Eric Siegel (2013) compares this to a salesperson. Positive and negative interactions teach a salesperson which techniques work and which do not. In a similar way, predictive analytics enable organizations to learn from previous experiences or data.
How HR predictive analytics apply in practice
Now, how do predictive analytics apply to HR? HR possesses large quantities of people data, usually managed in a Human Resources Information System. By applying predictive analysis to this data, HR is able to become a strategic partner that relies on proven and data-driven predictive models, instead of relying on gut feeling and soft science. HR predictive analytics enable HR to forecast the impact of people policies on well-being, happiness, and bottom-line performance. An example is the role it can play in preventing expensive employee turnover.
However, only a few organizations are capable of producing predictive models for HR. According to Deloitte’s 2018 People Analytics Maturity Model, only 17% of organizations worldwide had accessible and utilized HR data. This is up from 8% in 2015, and 4% in 2014.
Of this 17% in 2018, only 2% qualified as having business-integrated data, meaning they use real-time, advanced AI-aided tools to collect, integrate, and analyze data. The other 15% is able to do predictive analytics on an ad-hoc basis.
This means that a lot of organizations still have a long road ahead of them before they can apply predictive people analyses. Despite this, early adopters already show some very interesting results. Let’s take a closer look at some of them.
Real-life examples of predictive analytics in HR
1. Predicting and preventing turnover at HP
Hewlett-Packard (HP) is a company with over 300,000 employees and has always been a leader in the HR predictive analytics field. HP’s management experienced a high level of employee turnover. Turnover rates of 20% were not uncommon in some of its sales divisions. This meant that people stayed on average between 4-5 years.
High turnover generally leads to high recruitment costs and lost revenue due to productivity loss and onboarding. Additionally, departing employees take their knowledge, network, and sometimes even customers with them. It is estimated that the cost of replacing mid-level employees is upwards of 150% of their annual salary. This can cost a company like HP millions of dollars.
Two scientists decided to try and solve this problem. They combined data of the previous two years and attempted to predict who would leave the organization. By using predictive models, they generated what they called a “Flight Risk” score. This score predicted the likelihood of leaving of each of HP’s 300,000 plus employees.
Their findings were groundbreaking. Based on the data, they could see why employees would leave HP. Higher pay, promotions, and better performance ratings where, for instance, negatively related to flight risk. However, there turned out to be intricate relationships between those findings. For instance, when someone received a promotion but did not get a substantial raise, this person would still be much more likely to quit.
As you can imagine, there were a number of practical and privacy-related problems with this Flight Risk score. This is why access to this data is only granted to a select group of high-level managers.
These managers could only see the scores of their subordinates. They also received training in interpreting Flight Risk scores so they would understand the potential ramifications and confidentiality issues that come with this data.
Additionally, the system informs these managers what the key risk factors of employee attrition are. This way, the system exerts pressure on managers to develop strategies to retain their staff. This helps to reduce costs and maintain business continuity.
Considering all the different factors a company could take into account when doing predictive analytics, it’s easy to get lost in details. It could, therefore, be of great value to create a dashboard with an overview of the most important metrics.
In the end, Flight Risk scores acted as an early warning system. It prompted well-trained managers to intervene before it is too late. Or, when the loss of an employee was unavoidable, to react accordingly. According to Siegel (2013), HP was able to save an estimated $300 million by applying predictive analytics to calculate this flight risk.
2. Predicting hire success at Google
In his book Work Rules!, Laszlo Bock, Senior Vice President of People Operations (HRM) at Google, writes that the most important instrument of Google’s People Operations is statistics. The questions interviewees get asked in Google’s hiring process are all fully automated, computer-generated, and fine-tuned in order to find the best candidate.
On top of that, Google estimates the probability of people leaving the company by applying HR predictive analysis. One of Google’s findings is that new salespeople, who do not get a promotion within four years, are much more likely to leave the company.
There are a lot of other relevant metrics in recruiting that you should keep track of to see whether you’re hiring the right people. These include advanced metrics like time to productivity and cost of getting to optimum productivity level.
3. Predicting revenue using engagement numbers
Employee engagement is often seen as the holy grail of HR. Employees who are engaged work harder, deliver better quality, are less absent and less likely to quit.
Best Buy, an electronics retailer, wanted to know how engagement impacted the sales in their stores. They analyzed their data and found that a 0.1 percentage point increase in engagement leads to a $100,000 increase in revenue per store.
After finding this astonishing impact, Best Buy decided to measure engagement multiple times a year. This also enabled them to measure the drivers of engagement. This helped to come up with HR interventions that would increase engagement and, in turn, store revenue.
Wikipedia editors, or Wikipedians, create and edit articles to keep the world’s largest encyclopedia up-to-date. Each day, over 800 new pages are created, and 3,000 edits are made on the English Wikipedia alone (>3 per second worldwide). Wikipedia is able to predict who of its 750,000 editors is most likely to stop contributing.
I am not sure how Wikipedia acts on this information, but a kind ‘thank you for your contribution’ email could do wonders in appreciating and re-engaging these Wikipedians.
5. Making better hiring decisions using Facebook
Do your recruiters check the Facebook pages of applicants? Maybe they should. A study revealed that it is possible to predict someone’s personality and future work performance based on their Facebook profile (Kluemper, Rosen & Mossholder, 2012). In this study, a number of participants gave hirability ratings based on Facebook profiles. These ratings predicted 8% of manager-rated job performance for these people.
8% may not sound like much. For instance, a standard personality test has a higher predictive value for performance compared to looking at someone’s Facebook profile. However, the literature shows time and time again that the best predictive models for future job performance combine various predictors, such as IQ tests, structured interviews, and personality tests together. Looking through a Facebook profile could be an additional instrument to scan candidates.
6. Predicting impact at Nielsen
Nielsen, a data analytics firm, had a similar problem to HP. After talking to one of the presidents of one of the businesses, the SVP of People Analytics, learned that retention was a problem that needed to be worked on.
After doing a financial impact analyses, the People Analytics team learned that every one percentage point decrease in attrition resulted in a $5 million reduction of business cost.
The People Analytics team was then able to identify 120 key individuals who work at risk of leaving. Through lateral moves for 40% of the group, the team was able to reduce the attrition rate to zero for the first six months after implementation.
Through different initiatives aimed at reducing turnover, the team was able to reduce attrition by two percentage points for the global enterprise, resulting in a $10 million reduction of cost. Following this successful implementation, the project was rolled out to seven other countries (source).
7. Never hire toxic people
The last example is from a case study by Cornerstone about toxic employees. Cornerstone studied the impact of toxic employees in the workplace. Toxic employees are employees who are most likely to engage in toxic behavior. Examples of these behaviors are fraud, drugs or alcohol abuse, and sexual harassment.
These people are not only damaging to the company; they are highly toxic to the general work environment. Previous research suggested that one toxic employee in a team would cause productivity to decrease by 30% to 40%. On top of that, good employees are more likely to quit when they have to work together with toxic colleagues.
The company used a dataset of 63,000 employees. In this dataset, they marked which employees were involuntarily terminated due to workplace violence, falsification of documents, drugs, and alcohol abuse, and other policy violations. Based on these criteria, around 4% of all employees could be classified as being ‘toxic’.
After analyzing the dataset, Cornerstone identified a number of key characteristics of toxic people.
1. are self-proclaimed rule-followers;
2. score low on attendance and dependability;
3. and have a low service orientation.
Remarkably, the study did not find the previously reported high levels of productivity loss in the short term. However, it did find toxic behavior to be contagious. People who work together with toxic colleagues are also more likely to quit. Additionally, the study hypothesized that toxic colleagues contribute to long-term stress and burnout among other employees.
In the end, Cornerstone proved that hiring a toxic employee will cost the employer $12,800 on average, versus an average of $4,000 for a non-toxic employee. This excludes the long-term (and costly) productivity loss through burnout and other negative effects. By fine-tuning the hiring process, companies can prevent hiring candidates who are likely to become toxic and create a healthier working environment.
A game-changer for HR
As these previous examples show, the results of applying predictive people analytics can be astonishing. HR departments can potentially save (or earn) their company millions of dollars. Additionally, HR can help their managers and executives make better decisions by applying predictive analytics and using the right HR metrics.
The potential of predictive people analytics demonstrated by these business cases (and other HR analytics case studies) makes it clear that predictive HR analytics are here to stay. They are the game changer that enables HR to not only assess how employees work but also to predict and optimize the impact of people policies on both the employees and the business.
To learn more about the application of predictive analytics and HR analytics in general, check out these 11 online HR analytics courses, or our HR Analytics Academy in which we offer courses tailor-made for HR professionals who want to learn more about analytics.