By Mark Maaskant, road safety expert
Providing insight into driving behaviour is harder than you think. That is, if you want to do it as thorough and accurate as we do. Using data to interpret driving behaviour is not new; you’ve probably heard of a ‘Pay How You Drive’, driving behaviour insurance. A telematics device installed in cars is used for these insurances to track driving behaviour. But the data coming solely from these devices is not enough to really assess driving behaviour. In this blog I will tell you how we enrich and interpret data to get a clear view on driving behaviour and actionable opportunities for road safety.
Reliable data and behaviour scores
Usually, a telematics device measures driving behaviour, such as acceleration, deceleration and speed, with standardised values. The disadvantage of using standardised values is that this does not take in account variables depending on the type of vehicle, type of road or the weather. Another disadvantage is the accuracy of the map data used to determine speed limits, which is 90%. Although this number sounds high, it impacts the accuracy of the driving score and with that the credibility. An unjustified bad score annoys drivers, which makes them less likely to believe other scores or follow any improvement instructions.
I test a lot of systems myself. I drive an electric car that brakes on the electric motor (regenerative braking). If I have driven a 40-kilometer ride without touching my brake pedal (one pedal driving, very economical!) and I get feedback that I have braked too hard, I can no longer take the data seriously.
Besides the accuracy of all data, the telematics devices also don’t interpret behaviour that is not necessarily outside safe parameters but does pose a risk. For example, a driver who does not drive faster than 100 km/h on the highway but fluctuates between 90 and 98 km/h may be distracted by a phone while driving. How do you deal with that?
Agnostic data platform
Before I dive into driving behaviour data and scores, I want to make it very clear: one of the core values at gribb is that our data is used to make people (even) better, not to exclude people. Our data enables companies to contribute to safer traffic, a safe working environment for the driver and to improve the drivers ability! And when it comes to privacy: our end users are people who drive a vehicle professionally, so the employer simply wants the vehicles to be driven safely. It already keeps track of where the vehicle is driving and who is behind the wheel. We interpret and enrich this data.
Back to that driving score. What data points will you use to determine if someone is driving safely? How do you subsequently interpret that data and how do you avoid only measuring incidents and assessing a driver based on that?
At gribb we work with an agnostic data platform. This means that our platform can communicate with multiple telematics suppliers or OEMs (car manufacturers). Each of those data streams run differently, in different values and in different intervals, hence agnostic, we adapt to the TSP (Telematics Service Provider) and not the other way around.
We calculate a driving score based on the available data and the interval of the available data. We assume that the customers who come to us already have telematics in their vehicles. We do not sell telematics devices. These already existing telematics, mainly determine the quality of the driving score that we can provide.
What data do we use
With most TSPs we can read X,Y,Z sensor data, driver ID and GPS position. With X,Y,Z data you can partially determine how fast the driver accelerates, brakes or goes through the bends. But this data differs per vehicle type, these values are very different for a passenger car than for a truck with a trailer. In other words: if you are going to use this data (and yes, we do) you must consider other values per vehicle type and also the nature of the labor. A driver who largely drives on the highway has a completely different driving behaviour than a courier who makes more than 100 stops per day in mostly urban areas.
It is also important in which interval this data comes in, 1 x per second (High Frequency), 1 x every 10 seconds or 1 x every 30 seconds?
With the lat/long data or the GPS position, we can see where the vehicle is driving. We overlay this with map data. This way we can tell whether a car is on the highway or, for example, within a urban area. A truck has a lower speed limit on the highway, so we use the vehicle registration data to verify whether the vehicle we track is a truck, a delivery van or a passenger car.
GPS position is not 100% accurate, so you run the risk of projecting a vehicle onto a parallel road instead of a main road with a different speed limit. We use calculations to rule this out. for example, if we see that you were driving from roundabout to roundabout, then you were driving on the main road and not on the parallel road, where there were no roundabouts.
The map data that we use to determine the speed limits per road type is provided by a combination of data suppliers. We are even able to include the speed limit on matrix signs when determining the maximum speed and we can read warning signs in traffic.
We do this in various European countries where traffic rules and road types are different. Those G-forces, for example, not only vary per vehicle type, but are also very different when driving in the mountains in Austria or on B-roads in England. It is really a complicated game where you continuously encounter new challenges, insights and customer wishes.
Multiple drivers per vehicle
To make things even more difficult, our customers often have multiple drivers per vehicle. If we want to provide a driver with feedback on their driving behaviour at the end of the shift, we need to know which driver was driving the vehicle. This is relatively easy for trucks and taxis because we have the tachograph and driver’s card for this. It becomes more interesting for the driver of a delivery van without a driver ID. To achieve the right identification of each driver, our developers do very clever things to connect the hand scanner or mobile phone with the employer’s app to the vehicle. In short, data quality, data frequency, data interpretation are extremely important to arrive at a good driving score. However, we are not there yet….
You cannot assess whether a driver can improve based on just a few trips. A particular driving score is also influenced by the environment, the schedule or the condition of the vehicle. Sometimes the cause of a low driving behaviour score does not lie with the driver. You must look at the whole picture, which is why we have our own data analysts who, based on historical data, see when the driving behaviour of a good driver starts to deviate. What’s the reason behind that? Are there issues with planning or the vehicle, is the driver not feeling well? And how can we use all this data to predict and prevent issues? For example, Ramadan; if your drivers are awake before sunrise and are not allowed to eat and drink during the day, you can take this into account and plan the shift accordingly. Around 3 pm these drivers will get extra tired; do you accept the increased risk or act in advance and create understanding for your drivers.
OBD and OEM data
We are most happy with OBD and OEM data. With OBD data we can measure derived driving behaviour, for example based on a steering angle or accelerator pedal sensor. We can see when there is too little distance between vehicles, whether seatbelts are being worn, whether a driver is maneuvering backwards or whether they are driving in heavy rain without adjusting their speed.
Currently, the data comes from aftermarket telematics suppliers. In the coming years, this will shift to the data provided by car manufacturers (OEMs). We are already talking with manufacturers to unlock and interpret this data. We also do this with bicycles, cargo bikes and LEVs.
As you can read, we take data linking, data collection, data interpretation, data enrichment and data analysis seriously. So, it’s cool that large parties in transport, logistics, passenger transport, insurers and intermediaries are now using our data to help themselves and their customers with damage prevention. In 1993 we started skid courses to make traffic safer and now we work with data and marketing.