Predicting driver turnover: the model sends a message

user-gravatar Headshot
Updated Mar 2, 2017
Omnitracs ELD Driver Retention model uses HOS data to predict which drivers are most likely to be dissatisfied with their job due to fatigue and other factors.Omnitracs ELD Driver Retention model uses HOS data to predict which drivers are most likely to be dissatisfied with their job due to fatigue and other factors.

Predictive analytics are part of our daily routine. Many of our online activities are tracked and fed through computer models and algorithms that seek to learn our behaviors and make recommendations on what to read, watch or buy.

In the trucking industry, predictive analytics have been, and continue to be, applied to common challenges. One application is to identify the drivers who are most likely to leave.

Recently, a Chicago-based startup, Enlistics, created an online driver application management system. One of the differentiating features lets drivers optionally login to the app when applying for a job by using their Facebook profile, similar to how many consumer websites now work.

The Enlistics application uses the driver’s Facebook ID obtained from the login process to gather data on their social media activity. The application also collects drivers’ work history and runs background checks.

With these combined data sources, Enlistics says the model can predict the likelihood that a driver will stay for six months. This prediction is the only information fleets will see from the driver’s social media records, says the company’s founder, Austen Mance.

Enlistics is in the process of searching for carriers to work with and is offering a discounted rate, he says. The company previously launched a predictive employee turnover model for car dealerships. A common social media phrase that correlates with high turnover for car salespeople is “I’m sick of…,” he says. “That tends to show they get stressed easy.”

Using ELD data

One of the possible drawbacks of using social media to predict turnover is that the process is done without driver awareness. That is not the only source of data that drivers might not ever guess can be used for that purpose.

In this real-life example, Stay Metrics used its new predictive model to find a unique “predictive index” for one fleet (in green). This index shows the likelihood of drivers to leave based on responses to certain questions in Stay Metrics’ annual driver satisfaction survey. The red line shows what happens when the predictive index is applied to other Stay Metrics clients–nothing stands out, denoting that each fleet has unique predictors for turnover.In this real-life example, Stay Metrics used its new predictive model to find a unique “predictive index” for one fleet (in green). This index shows the likelihood of drivers to leave based on responses to certain questions in Stay Metrics’ annual driver satisfaction survey. The red line shows what happens when the predictive index is applied to other Stay Metrics clients–nothing stands out, denoting that each fleet has unique predictors for turnover.

Omnitracs Analytics, a division of Omnitracs, developed an ELD Driver Retention model that predicts driver turnover model by extrapolating more than 1,000 variables from hours-of-service (HOS) data from electronic logs.

The model uses HOS data that would indicate that a driver is dissatisfied with their jobs due to fatigue or variability of their work schedule, for example. Even so, it may not be easy for fleet managers to talk to drivers about predictive information that drivers are not being measured on and held accountable for.

The ELD Driver Retention model reports information through an online portal called the Driving Center. The reports give visibility to drivers with the highest turnover risk, the reasons why, and suggestions for how to remediate.

Managers use this information to proactively reach out to drivers and have a conversation, says Lauren Dominick, director of analytics and modeling at Omnitracs Analytics.

“The idea is to have a positive, non-transactional conversation,” she explains. A manager would begin the conversation by asking how the driver is doing. After listening, the manager could offer suggestions and help resolve problems caused by stress, fatigue, or inconsistent work schedules.

Increasing engagement

Some approaches to predicting turnover are transparent for the driver from start to finish.

Stay Metrics uses several methodologies that help motor carriers understand how drivers feel about their careers, and identify those with the most risk of leaving. Its methods include 7-day orientation and 45-day onboarding interviews, an annual Driver Satisfaction survey, exit interviews, and a privately branded online rewards, recognition and driver engagement platform.

“We know that behavior follows engagement and surveying drivers both increases their engagement but also gives us important insights into controllable causes of turnover,” says Tim Judge, director of research for Stay Metrics, and the Joseph A. Alutto Chair in Leadership Effectiveness at the Ohio State University’s Fisher College of Business.

Recently, Stay Metrics announced the next generation of its predictive driver turnover model. Developed by Dr. Judge, the model extrapolates data from its full product suite, which has been integrated into a single database, and provides insights — specific to each motor carrier — on why drivers leave their companies.

How data is collected for predictive models might just be as important as the information about who will leave, and why. Some models are very efficient at collecting data without any driver involvement, while others give drivers a chance to provide direct feedback and be engaged in the process.