What You Need to Know about Driver Scoring/Why You Should Care
From a fleet safety perspective, you are making decisions based on scores, right?You are providing feedback to drivers based on this score. As an insurer, you may be predicting and pricing exposure to risk with these scores. You would also hope that your insured’s use of a loss prevention program actually delivers the desired end result, namely, an improvement in loss history. You may even financially participate in your insured’s use of this technology if it aids your underwriting and fleet loss histories. These driver scores ought to be as accurate as possible, right?And yet there is a pervasive lack of sophistication when it comes to driver scoring. Lots of “good enough” or “better than nothing” scoring, but very few cases yielding truly accurate driver scoring.
This paper explains how you can evaluate the scoring; specifically why typical driver scorecards are flawed; and also provides a guide to clarify what questions you ought to be asking your vendor to determine if scores are as accurate as possible. You simply can’t draw good scoring conclusions from seriously flawed science.
Driver Scoring is Focused on Interpreting the Risk of Vehicle Momentum Changes
There is a tremendous range of sophistication in the calculation of driver scorecards from telematics data. Driver scoring is predominantly driven by the interpretation of changes in vehicle momentum as measured most typically by accelerometer data. Most currently available solutions fall far short in applying the best science to the scores. The end result of this is data that is seriously compromised in its determination of risky driving behaviors and events. Each potential compensation, poorly applied (or not applied at all), will lessen the accuracy of your scores.
What is Wrong with G-Forces?
While g-force (or Delta V, Delta T) evaluations are the key to driving risk identification, a g-force, by itself, is a poor indicator of risk. G-Forces do effectively measure momentum changes, but those momentum changes only have real meaning when taken into the context of the entire driving situation. You don’t really know if you have a real event by looking just at the g-force. You have no idea of the magnitude of the event from a g-force reading. A high g-force may actually represent relatively low risk, and visa versa. A focus on g-forces is a vestigial remains of the days when driver scoring was new and data transfer was much more expensive.
What is Wrong With the Typical Solution?
The typical solution, out of expediency, will measure exception-based events. The device in the vehicle captures “events” that exceed pre-determined and arbitrary thresholds. This means something along the lines of “I exceeded a pre-determined g-force and this therefore qualifies as a hard-braking (for example) event”. The tally of the number of such infractions thus goes into the calculation of some overall driving score. While being better than nothing, this approach is only a very crude proxy for the determination of driving risk. While it may have some limited value to fleets (it is better than nothing), the commercial insurer is looking for greater, more accurate identification of driving risk. With an “event-based” system (real time determination at the device level), this accuracy is impossible to attain. A sophisticated scoring system should take advantage of the opportunity to draw conclusions from much bigger data sets. Bigger data post-processed yields the greatest insight into behaviors since all appropriate compensations can then be applied in parallel.
How Does Poor Science Negatively Impact Driver Scoring?
- Are your driver scores a function of the challenge of the driven route? A more challenging route has the opportunity for more driving infractions, thus worse driver scores. Good science, looking at second by second momentum changes and actual driving events, can determine the impact that the route has on driver scores. Drivers should not be punished for driving well in more challenging environments yet insurers want to know their total exposure to risk (which includes the impact of the route).
- Do your driver scores fail to differentiate between the magnitude of different events? “I know Joe had a hard-braking event last week but did it just exceed a pre-determined g-force threshold or was it a slam on the brakes precursor to an accident?” How do you differentiate this event from any other braking event? Both the frequency and severity of driving events should be taken into consideration.
- Do small vehicles score worse than larger vehicles? Without appropriate compensation they do. You cannot effectively score driving behavior without the weight of the vehicle taken into the calculation. Larger vehicles score better because they can’t make the momentum changes that smaller vehicles are capable of. Driver scoring is a function of vehicle momentum change analysis.
- Can you change behavior and notice the difference in your score (immediately)? Is scoring sophisticated enough to immediately “react” to changed behavior?
- Can you accurately compare all drivers so you can determine which ones are targets for improvement? Are you accurately segmenting where risk actually lies across your entire driver base?
- Is there a “back end” to your solution? Meaning, once you have your score, do you know how to interpret it, provide appropriate intervention/feedback to drivers, and track resultant changes in behavior
- Can you measure the underlying behavior tendencies that yield exposure to eventual actual driving events? (Second by second analysis of all “accelerometer noise” in addition to identification of events
- Is an “event” determined by more than just one variable exceeding a pre-determined threshold? To really understand an event, you should include consideration of speed at time of g-force infraction, weight of vehicle, duration of event, known risk of combination of behaviors, severity of event, impact of the driven route on event determination, etc.
- The simplest question can be the most meaningful. Ask your potential vendor “How is a risky driving event determined?”
Science, applied correctly, can provide very meaningful identification of risky driving behavior and be the basis for an effective driver improvement program as well as be the basis for delivering more insight in regards to underwriting. While UBI solutions have been mainstream for a long time, required accuracy of driver scoring has been a gating factor in generating driver scores that can truly give commercial insurers superior insight into the risk exposure of driving behavior. The pending expansion in telematics data interpretation in the commercial marketplace will be based on delivering more accurate driver risk scoring. The “good enough” scoring of so many driver dashboards and the simpler focus on UBI scoring simply will not get the job done in the commercial arena.