in ,


Shedding New Light on Hidden High-risk Fleet Drivers

Fleet managers dedicated to minimizing accidents among their drivers have several options for assessing and managing risk. Until now, that has primarily involved tracking and responding to recent driver history.
An ongoing study of a new approach employing predictive analytics methodology is showing promise for more accurate and effective fleet driver risk assessment. In fact, it reveals that the standard model for assessing fleet driver risk actually understates the risk some drivers pose, making them “hidden” high-risk drivers. Used as a supplement to current best practices, predictive analytics may help fleets that have used systems that aim at changing driver behavior to make further gains in preventing accidents, protecting drivers’ lives and health, and reducing their net cost of accidents by millions of dollars.


For the past 20 years, fleet managers have employed various driver risk-assessment programs in an attempt to quantify and change risky driving behavior. Those programs reflect studies dating back to the 1970s, which found relationships between the combination of a driver’s history of motor vehicle violations and accidents over a two-year period and the likelihood of being involved in an accident in the third year.
Although fleets have different scales and formulas for assessing driver risk, their programs to prevent fleet accidents have these features in common:

this post is proudly sponsored by:
  • Documenting driver behavior by collecting driver Motor Vehicle Report (MVR) and accident history
  • Intervening with targeted training to encourage better driving behavior
  • Establishing consequences for continued high-risk driving behavior
  • Maintaining a record of fleet driver evaluations and interventions as protection against potential enterprise liability for negligent entrustment

Fleets using these methods typically sort drivers into three or four ascending levels of risk based on their performance over a rolling 36-month period. They assign point values for each type of accident and traffic violation, and each drivers’ point value places them into one of the appropriate risk levels.
Fleets intervene with drivers—holding them accountable—when new data pushes them into a higher risk level. Consequences can range from being required to take additional on-line or behind-the-wheel training to being assigned a less desirable vehicle, losing some driving privileges, being denied a raise or bonus, all the way to termination.
This behavior-based approach to preventing accidents has had enormous success. For example, fleets using CEI DriverCare™ accident-prevention programs have experienced, on average, a 15-percent reduction in accidents after three years and a 25-percent reduction after five years. Over the longer term, fleets have prevented thousands of accidents and cut their accident rates by as much as 35 percent.
As impressive as these results have been, the fleet industry has been hoping that the science of predictive analytics—a method that mines data to predict future events—could be applied to their drivers and achieve better results.


A practical predictive analytics model for assessing fleet driver risk has been the focus of a five-year coordinated development effort between CEI and Dr. Feng Guo, a professor at the Virginia Tech Transportation Institute.
A key difference between CEI’s predictive models and the current, standard method of assessing driver risk is the incorporation of driver demographic data that are precluded from fleet safety policies—like gender and age—the industry in which the fleet operates, and whether or not the driver is a manager. Unlike the standard industry practice, CEI’s predictive model takes into account five years of driver history instead of three, and overlays it with national aggregate accident frequency statistics for each demographic variable.
Traditional vs predictive risk assessment chart
The predictive model uses that data to calculate the probability each fleet driver faces for an accident over the following 12 months. The numbers can then be used to sort drivers into groups with similar probabilities, project those groups’ accident rates, and compare them both to one another and the fleet’s overall projected and actual accident rates.
Six-month interim findings from a beta test of CEI’s predictive analytics model reveal surprising differences in driver risk assessment rankings and a high correlation between predicted and actual accident rates.
The last column in Figure 1—the trending accident rate—is based on the actual numbers of accidents each group of drivers had after six months. Perhaps the most revealing results are the significant increase in the number of drivers in the two highest risk categories, and the disparity between the accident rates for those groups on the one hand, and the rates for the safer groups and the fleet as a whole.
While the standard risk assessment method identified a total of 221 drivers at elevated risk (87 in Risk Level 3, plus 134 in Risk Level Two), the predictive model found 1,020 (162 rated “Highest” and 858 rated “High”), a nearly five-fold increase. Six months after the predictive model was run, the fleet’s overall trending accident rate was close to the predicted rate of 26 percent, but the rate for the “Highest” risk group was nearly 47 percent, and more than 35 percent for the “High” group. Meanwhile, the trending rate for the “Safest” drivers was less than 19 percent.


As this example demonstrates, predicting driver accident potential reveals that many more drivers may benefit from additional training and coaching than the standard approach to risk assessment indentifies. And, because the model identifies those drivers independently of any recent accident or traffic conviction, it provides additional opportunity to change driver behavior and prevent accidents.
None of this means the standard approach to monitoring driving behavior is obsolete. In addition to tracking accidents and motor vehicle violations, DriverCare also takes into account telematics data, traffic camera violations, and public complaints, and fleets can still benefit from holding drivers accountable when new events elevate them to higher risk levels. But, predictive analytics holds promise for an unprecedented level of accuracy in uncovering the high-risk drivers who may go unidentified by the reigning approach to fleet driver risk assessment.


Brian Kinniry is senior director of strategic services at The CEI Group, Inc., a fleet driver management company that provides technology-enhanced accident prevention and accident management to automotive fleets. Find out more about CEI, visit


Did you enjoy this article?
Subscribe to the FREE Digital Edition of Modern WorkTruck Solutions magazine.

Maxwell Engine Start Module