Covenant Transportation Group, Inc. (CTG) is a $650 million freight transportation and logistics services company headquartered in Chattanooga, Tennessee. Three of its business units carry loads for shippers, but these unit’s business models were not exactly driver friendly. CTG does not have many dedicated opportunities in its largest Business unit leading to few regular routes. Two business units have operations that request drivers to stay on the road for at least 21 straight days. Another unit has an expediting service, which leads to deadline pressures. They also have their own Commercial Driver License Class A (big truck) training operations. Turnover for new drivers, who don’t fully understand the demands of this profession, is very high.
So it is perhaps not surprising that in 2012 CTG was slightly above than the industry norm in terms of driver turnover. According to Doug Schrier, the Vice President of Continuous Improvement and Program Management at CTG, that just was not good enough. Even though in many ways CTG’s operations are more demanding than other carrier groups, and the way they calculated turnover was more stringent than some other carriers – they include drivers they have let go for performance problems in their calculation – the company wanted to perform significantly better on this key performance indicator than the industry as a whole.
CTG started a predictive analytics project in 2012; “we learned a lot” Mr. Schrier said. “But while we began to understand the power of predictive analytics, we also realized our in-house capabilities were immature. We wanted to bring these capabilities inside and improve our maturity.” In 2014 they hired an expert, Chris Orban, who had been doing predictive analytics for over a decade.
The next decision involved which analytics vendor to use. They had several bids. Mr. Schrier said, “we chose SAP HANA because they had a good modeling tool.” In particular, they believed that SAP’s ability to rapidly develop analytic models was superior to the other suppliers.
The HANA project started in March of 2014, they had developed their first model by August. The main bottleneck was that in addition to deploying a new tool, they needed to develop a mechanism to deliver the task recommendations to managers.
“Each model is a bit different,” Mr. Schrier explained. “A model for a professional driver who has been on the job for 6 months is different from a model for a driver that has been on the job for five years. The model for someone who is home every other week is different than that for a driver who stays out for 21 days.”
There are a number of predictive factors considered. A few of the more powerful predictive factors include how frequently drivers run the same hours during the course of the day; how fully utilized the driver is; and the variability of driver pay over multiple weeks. “If a driver can’t predict what they will make next week, that adds stress.”
The company is also looking for inflection points. Changes in behavior can indicate an inflection point. For example, if a driver didn’t routinely take advances and then begins, this indicates a change in the drivers financial behaviors which could indicate that a driver is under stress. Stress leads to an increase likelihood of turnover. There are 100 to 200 attributes that have predictive power in the models, but thousands that are considered. “The power of HANA, is that a new factor can be considered and then computed that quickly.”
Safety issues are also a predictive factor for turnover; “the data off trucks is amazing” Mr. Schrier said. “Near misses and other critical events may indicate that a driver is not focused on their current task.” Safety issues lead to coaching by safety personnel and other management. “If the manager handles the coaching poorly, the driver is more likely to leave. It’s just like any other interaction, but potentially more stressful because of the adverse nature of the discussion. But safety factors are key. If a driver is not coachable on safety, we want them to leave anyway.”
The company also brings in external data. One example of this is that the CTG searches job boards for potential new hires. They sometimes discover that a driver they employ is looking for work.
When a model predicts that a driver may be thinking about leaving, there is an intervention event between the manager and the driver. “These are relatively simple,” Mr. Schrier explained. “We ask the manager to have a ‘what’s up, Jack’ conversation.” Day to day operations can be so busy, that many managers rarely have these types of conversations unless they are prompted to. “We encourage the fleet manager to spend 5 to 10 minutes on a call with the driver and build rapport. The manager needs to “see where they are at” and how explore how the company could better assist them. For example, if the driver has a sick child, they may need to spend more time at home with that child. Another driver may be experiencing financial pressure, advances or more predictable schedules might help that driver.
What are the results of this program? Despite trucking operations that are more challenging than many of their peers, CTG’s job turnover is 5 to 8 percent lower than when starting the program.