By: Eric Siegel, Founder, Predictive Analytics World

In anticipation of his upcoming conference presentation, Risk Management Algorithms – Paving the Way to Value Creation at Predictive Analytics World Financial in New York City, October 23-27, 2016, we asked Frank Fiorille, Sr. Director of Risk Management at Frank Fiorille imagePaychex, Inc., a few questions about his work in predictive analytics.

Q: In your work with predictive analytics, what behavior or outcome do your models predict?

A: We have a portfolio of models that predict both external client behavior as well as internal employee behavior.  Our external client portfolio includes (but is not limited to) models predicting upsell opportunities, client retention, discounting, segmentation and likelihood to default on payments.  Our internal portfolio consists of models related to employee churn.  As Paychex has over 600,000 clients and generates hundreds of millions of transactions yearly it is not economical to contact every client/prospect or screen every payroll Paychex processes.  This is where our models excel as they provide resource efficiency increases as well as enhance the impact of any actions taken.

Q: How does predictive analytics deliver value at your organization – what is one specific way in which it actively drives decisions or operations?

A: One of our most impactful model ensembles predicts client losses over time. Due in large part to the success of the model, Paychex created a dedicated client retention team solely working off the results of the model. The team proactively reaches out to clients the model has identified as likely to leave Paychex, often engaging them in correcting previously unknown service issues, pricing concerns, etc.

Q: Can you describe a quantitative result, such as the predictive lift of your model or the ROI of an analytics initiative?

A: Attributing quantitative value to predictive modeling varies depending on the model, how it is deployed, and how the value will be interpreted. In order to deduce value from the client retention model mentioned above, we compare the loss rates of clients across model scores and retention team activity. In one particular segment, losses were cut by 50% when they had proactively been called by the retention team.

Q: What surprising discovery or insight have you unearthed in your data?

A1: When clients switched IT platforms, it was expected that more errors in the transition lead to higher churn rates. We investigated if this expectation was backed empirically and found that the effect was inverted; clients with more errors were less likely to leave. This finding echoed similar results where anecdotal beliefs were refuted by empirical analysis.

A2: A natural experiment arose in a segment of our data where we were able to analyze how different discount rates affected client retention rates. We investigated to see if clients that automatically received extended discounts versus lower discounts had a higher risk of leaving Paychex. We found no significant correlation, a disappointing but common result in most cases but not in this case.


Don't miss Frank's conference presentation, Risk Management Algorithms – Paving the Way to Value Creation at Predictive Analytics World Financial NYC, on Tuesday, October 25, 2016 at 2:40 to 3:25pm. Click here to register to attend.

By: Eric Siegel, Founder, Predictive Analytics World