Human resources is more capable of collecting data than ever before and with that data, a wide range of decisions and observations can be made. In recent years, HR teams have familiarized themselves with the different types of analytics and have attempted to discover how their teams and organizations could use them, but one type of analytics remains difficult for many organizations to achieve: prescriptive analytics.
Prescriptive analytics refers to the type of data intelligence that allows organizations to combine the capability of descriptive analytics (what most are achieving now) with a view toward the future. Users can gain insight into what will happen next, but more importantly, prescriptive analytics provides insight into what the organization should do next.
You’re likely most familiar with this model through services you subscribe to in other areas of life, be it Spotify’s Discover playlist or the recommended for you section of entertainment streaming service platforms. It looks at your past usage data in an effort to predict what else you’d like and what you will like in the future. In the consumer market, the message around these can be customized to increase engagement and optimize how you experience whatever it is the model believes you will like the most.
Prescriptive vs Predictive
It’s important not to confuse prescriptive analytics with predictive analytics. Though they may sound similar, the two are actually quite different.
While predictive analytics is also valuable, providing the ability to identify employees that are most likely to quit, for example, prescriptive analytics could be used to build retention plans and plot actions to achieve desired outcomes.
For HR, that could have huge impacts for learning and development and workforce planning in the very near future, but questions over data quality and how to implement it without disrupting the workforce further remain. But there’s another problem according to Sarah Johnson, Vice President of Enterprise Surveys and Analytics at Perceptyx.
“It’s human nature to believe that we know something that can’t be accounted for in the data based on our experiences,” Johnson said. “The challenge will be twofold: First of all identifying the many, many variables that need to be included in the analytics, both internal to the company, individual on the part of employees, and the external world. Collecting these data and then determining how to account for them and weight them in the analysis and solutions will be a challenge. But let’s assume we figure all of that out; the challenge will then be the human decision to implement the recommendations. The human element may choose to override the recommendations from prescriptive analytics based on our own beliefs, expectations, and biases.”
Another challenge is determining the proper course of action that the analytics identifies would then require careful consideration. Current potential actions could be out of date or ineffective and there may be scenarios where the action needed isn’t clear.
“It seems to me that in the HR world, predictive analytics would be most effective where there are finite potential actions to be recommended,” Johnson said. “But even then, HR will need to continually monitor data to determine where there are scenarios where current solutions do not exist. The challenge will be to create recommendations that are specific to the situation and not just generic recommendations that can be interpreted and implemented in a variety of ways.”
Prescriptive analytics often resorts to optimizing accuracy over interpretation. Throw all the variables you can think of at an algorithm and it can tell you what is likely to happen with a high degree of accuracy, but it cannot necessarily describe the reasons why, which leaders long to understand prior to making decisions. And the issues with prescriptive analytics being an ideal fit for HR don’t stop there.
“Going deeper down the prescriptive analytics rabbit hole, algorithms by themselves do not care about the veracity of data nor what data is included or excluded,” Johnson said. “Take for example, Amazon: when feeding resumes into machine learning algorithms to identify top talent – it succeeded, saving time during the hiring process and improving quality of hire. But given that technical roles were predominantly held by males when analyzing 10 years of applications, black-box algorithms leveraged this insight to perpetuate the male-dominant bias. Amazon has since scrapped this prescriptive analytics tool.”
Finally, legacy systems remain in place that hamper prescriptive analytics efforts. In the end, new cloud-based technology is required to adequately support these efforts, but investment is tough at a time when the future is so uncertain with regard to the pandemic and business models going forward.
The Talent Pool
Analytics efforts in HR generally face another problem in the way HR professionals are trained. Most are not data scientists and many have no desire to be. As a result, the talent pool for advanced HR analytics most often comes from outside of the HR function. Few HR professionals receive the kind of training necessary to do this kind of work, so most organizations have to rely on data scientists currently working for the organization in other functions or work with an external vendor that can supply analysis. There are challenges with both.
“Internal data scientists will need to be educated in terms of HR,” Johnson said. “They will need to understand the limitations of HR data relative to other company data, and then be immersed in the role of the function and its governance, policies, and practices. An external vendor will develop models and solutions based on a wide spectrum of companies and practices, but often times these don’t feel specific enough to be of use to a given organization without extensive customization.”
Moving Toward Prescriptive Analytics
If moving toward prescriptive analytics sounds complex, that’s because it is. But the task is not insurmountable, organizations simply need to get started on the first step in the process and it’s not adopting a new fancy technology or hiring the most brilliant minds to develop custom algorithms.
The most practical place to start is to simply take stock of the data the organization collects about people. HR needs to create a strategy for organizing people data and combine that with the many other data sources needed for prescriptive analytics. Consider this is a whole-company challenge, meaning that there is a need to create a continuously updated source of data.
But perhaps the most important part of moving toward prescriptive analytics is understanding the need for balance. While it’s necessary to get the aforementioned buy-in from leadership to use prescriptive analytics in making business decisions, it’s important to understand where it will inevitably fall short.
“Human behavior is notoriously difficult to predict, and an action that is prescribed may be highly successful in one situation but not in another,” Johnson said. “Initially I think prescriptive analytics will be most valuable and useful in transactional scenarios. For example, training experiences may be based on performance on a certain internal training program. Attrition trends can identify the need to hire specific skills in a handful of company locations. The selection of certain employee benefits options may signal a move either toward retirement or women going on maternity leave, which then leads to staffing recommendations. Other decisions or actions, such as promotion decisions or salary change recommendations, may be challenging to drive with predictive analytics, as these decisions may also be influenced by variables that are not easy to measure and record.”
At the end of the day HR is a people business and people are complicated, they don’t always fit neatly into an algorithm and the human side of HR will always be something that humans have to own. While HR teams are recognizing the value of data intelligence, very few companies have conquered the predictive analytics part, to say nothing of prescriptive analytics. It will be a long time before this is common in organizations across the enterprise data ecosystem.
“People are complex and difficult to predict,” Johnson said. “While we can do our best to create prescriptive models we need to be willing to accept some element of error in the prescriptions. The data we have on people can only predict so much.”
Photo Courtesy of Stock Photo Secrets