Using Predictive Analytics Internally with Talent and Externally with Customers

Sue Lam's picture

This interview is the last in a series on workforce predictive analytics. APQC interviewed Steve VanWieren, Principal Statistician and Data Scientist at Ultimate Software.  In this role, he focuses on building a predictive analytics function, working with executives across all lines of Ultimate Software’s business, developing multiple statistical models, and actively speaking at HR and analytical conferences. In this article, he provides information on how Ultimate Software uses predictive analytics with its own talent in addition to how they use their software programs with customers. 

APQC: Could you tell me about your background and how you got into predictive analytics?

VanWieren: I received a Master of Statistics degree from University of Georgia, and I have been doing predictive analytics my whole career. I spent 15 years at Equifax, where I developed credit and marketing scores initially, but later managed the data quality of most of the consumer data in the United States. I entered into the HR predictive analytics arena because I felt I could pioneer something unique, trying to improve the way that companies address people management. I joined Ultimate Software just over three years ago, where I started up our predictive analytics function.

APQC: How do you define predictive analytics to an HR professional?

VanWieren: Predictive analytics is the blending of historical data to generate new data points or scores of some kind through an algorithm that provides an educated guess of a future event or behavior. I think that predictive analytics is useful when it generates a change in behavior once the new knowledge is made available.

APQC: Why do you think it’s important for today’s HR functions to leverage predictive analytics?

VanWieren: Predictive analytics can provide a great deal of value, even outperforming subjective opinion in comparison.  At Ultimate, we believe the blending of predictive analytics with manager instinct will provide the best outcome. It is by no means a replacement for how a manager evaluates his or her employees, but an additional resource to understanding employees.

APQC: What kinds of projects or questions can predictive analytics help HR answer?

VanWieren: I’ll use what we’ve done at Ultimate Software [as an example]. Three years ago, we created a metric called the UltiPro Retention PredictorTM. The Retention Predictor generates a score that identifies which employees are most likely to stay at their company for the next 12 months, so it’s a future-predicting model. We followed that up with solutions that focus on employee performance. One is called the UltiPro High Performance IndicatorTM, which is a data-driven way to identify who the high performers are today—displayed as a yes/no flag. If an employee is listed as a “no” or was not acknowledged as a high performer, we have a second tool called the UltiPro High Performer PredictorTM that reveals who is likely to become a high performer in the future. Now, we have a data-driven way to identify who the retention risks are, who the high performers are, and who are likely to become the high performers. In the future, we will begin to see a lot of predictors across various areas of human capital management including engagement, recruiting, and onboarding—pretty much anywhere human resources is involved in the business.

APQC: This is a software that was created by Ultimate Software, but is also used within the company? Can you give a case study example of how you’ve used it?

VanWieren: Correct. The UltiPro Retention Predictor is a logistic regression model used to predict whether a certain event will occur or not. It’s a probabilistic type of model. We tested more than 3,000 attributes and found 30 plus attributes that were most significant in the final algorithm. Once those are in place, we display those scores back to HR leaders and managers so that they can quickly identify not only the specific employees who are at risk, but also which parts of the organization are at risk—the parts of the organization that may have a disproportionate number of high performers. Anytime you start seeing data results like that, it leads to asking some questions. If you ask questions, it typically leads to action of some kind. The specific-use cases—and we use this internally ourselves—are used to help keep turnover low. Managers look out for someone doing something [that is indicative of turnover] and make adjustments as needed.

APQC: What do end-users and HR professionals need to know about data and analytics to leverage predictive analytics?            

VanWieren: Ultimate Software is a human capital management company so we develop and build software that can be used across all areas of human resources. For something like the UltiPro Retention Predictor, let’s use that as an example, any HR professional or business leader can review the results first to ensure the scores indicate what they’re supposed to. You can use historical data to prove that the models work. Once you’ve done that you can apply it to today’s population. When you do that, you can start to engage and more proactively support managers in the retention of top talent. 

APQC: What types of data go into that model? Do you look at hard metrics like cost, turnover or employee engagement and softer qualities?             

VanWieren: Any of those types of sources can be used depending on the tool you’re using and your objective for the information. [At Ultimate Software], we have focused so far on the employee work events. The examples I’ve cited—the UltiPro Retention Predictor, the UltiPro High Performer Indicator, and the UltiPro High Performer Predictor—are employee-focused scoring models, so they are based on things such as compensation history, job history, benefits, education, previous employment, licenses, etc.—anything that’s related to describing that person’s work history. Employee engagement and benchmarks and other sources of data can be tested as well to see if it’s valuable as an indicator.

APQC: What characteristics do good predictive analytics functions and analysts/data scientists have?

VanWieren: Definitely for anybody in an analytic role, they have to have curiosity. There needs to be data to back up any claim that’s made and typically that data leads to additional questions. Those questions create a circle and you start asking more and more questions and you analyze the data more, which leads you to a better finish line. But there also needs to be good business acumen in place. You have to be thoughtful with the actions that you are taking in order to facilitate a good outcome. Most good analysts and data scientists are good story tellers and can recognize when a bad action is taking place. Being able to tell that story the right way is also important because it helps to sell the business case to get people onboard with what needs to happen next once the data is in hand.

APQC: Do you see any common misconceptions when working with HR and predictive analytics?

VanWieren: The most common thing that I see is when a predictor shows the likelihood that an employee will do one thing, but they actually do the opposite, all of the sudden the credibility of analytics as a whole gets questioned. Nothing can be further from the truth. There’s no way that any predictive analytic will have every single point [of data] on a human being—sometimes we are going to miss it. Interestingly enough to me, people don’t hold their gut feeling to the same tough standard. Predictors will always outperform gut feeling when the two are looked at independently, but when used together, the best overall outcome can be achieved.

APQC: What are three tips that you could give to HR functions that are starting up their predictive analytics work?

VanWieren: First, to start small. Choose something with a black and white outcome like turnover—something that either happens or doesn’t happen. Second, don’t overthink it. Certainly when you’re trying to figure out what actions to take or how you would use this, start with the most obvious and simple one and measure the effectiveness of that action, and then optimize from there. Third—and this is the one I have to stress frequently—do not give up. Chances are you are going to make some mistakes if you’ve never done this before and the point is to optimize. Measure your effectiveness and move on from there. If you keep trying over and over again, eventually you realize that you don’t have to be a data scientist [to do this work]. You just have to be able to solve problems. At Ultimate Software, we are developing and building software solutions around the customers so they don’t have to [be data scientists].

APQC: What predictions do you have for the future of predictive workforce analytics? What capabilities might become available ten years from now?

VanWieren: We are just at the infancy of predictive workforce analytics. As I mentioned to you about my career path and getting here, I was surprised that more data wasn’t being used in this area. But I think so many things that HR organizations are responsible for are difficult to measure. They are very gray in nature. Ten years from now, you’ll see some standards that exist. It will be easier to quantify behavior, probably in ways that we can’t even think about today. I certainly think that the ways we recruit and hire people, onboard them, develop them, manage them, train them—anything where you are measuring engagement or culture—will become more data driven. Because these things are becoming more data driven, I think what tends to happen is there is a fear that machines and math are going to replace the human side of it and it couldn’t be further from the truth. These tools are meant to provide additional value towards all of those things. It will always be a blending of data and intuition to provide the best answer.

Ultimate Software has been named on Fortune's 100 Best Companies list for four years. You can follow Ultimate Software on Twitter. You can also hear Steve speak about predictive analytics in talent retention at the Predictive Analytics World for Workforce conference in San Francisco on March 31st. PAW Workforce is offering friends of APQC a 10% discount. To register, click here and enter APQCWF15 to take advantage of this great offer. 

Click here to read about using predictive analytics in business travel burnout.

Click here to read about using predictive analytics as a strategic HR solution.

Follow me on Twitter!

 

1 Comment

Anonymous's picture

Cool stuff. We at Joberate have been working on this for a while so it's great to see/hear that others are also making headlines as it helps the whole space. The one point that is made is so spot on, is that when a predictive analytic suggests something that goes against conventional group think, the analytics come under fire. For example, if you listen to the marketing hype machine of Jobvite, LinkedIn, Careerbuilder, Glassdoor, and countless others, you'll come to believe the anywhere between 50%-80% of your employees would consider opportunities elsewhere, while we used our technology to track actual behavior, not self-reported opinion, and by tracking the actual activities/behavior of 32,000+ employees across the entire Fortune 100, the data shows that only 12.1% of the workforce is actually exhibiting job seeking behaviors, a massive gap between self-reported surveys and actual activities. So again, thanks for this...it's a good piece, and if needed for the future we'd love to contribute input as you see fit.