The 25x crash risk differential in your driving data
Independent validation of our AI has found that risk is concentrated in a small group of drivers. In fact, the highest-risk group is 25 times more likely to be involved in a crash than the lowest-risk group. Here, Anders Lindelöf explains exactly what that means.
Can you identify the group of drivers most likely to be involved in a car crash?
This is what our AI does – and it’s what was confirmed in our recent validation. When we compare the group of drivers our AI identifies as highest risk against the group identified as lowest risk, the highest-risk group is 25 times more likely to be involved in a crash of any severity.
How does the AI identify whether drivers are high or low risk?
Our AI analyzes driving behavior across entire trips – not just individual moments or events. It compares behavioral patterns against a database of billions of real-world driving profiles with known outcomes. By looking at the full picture rather than isolated events, it can identify which patterns tend to be associated with higher or lower crash risk, and rank drivers accordingly.
Why does the AI use patterns instead of events to measure risk?
Let’s use an analogy. It helps to think of archery. What reveals the most about an archer isn’t where their arrows land, but how consistent their shots are overall. If someone is consistently off-target (all shots to one side, for example) it’s easier to see why – and to correct it. It’s more difficult if someone’s shots are completely scattered – close to the bullseye one moment, then missing badly the next. Read more about visualizing driver risk
Driving patterns work in a similar way. One risky moment doesn’t tell us much, just as one arrow in archery doesn’t. But the more trips that take place, the more the patterns in behavior tell us – and the easier it becomes to see what is influencing risk. Read more about understanding driving patterns.
How significant is the 25x finding?
It’s significant because it means organizations can approach risk management in a different way. If risk was spread evenly, it would make sense to treat every driver the same. We know that it isn’t, which means organizations can target their attention on the drivers who carry a disproportionate share of the risk, rather than spreading their training resources thinly across everyone. Read more about the most effective method of driver training.
What does the 25x finding reveal about risk distribution?
It confirms something we’ve been seeing in our data for a long time: driver risk isn’t randomly or evenly distributed. Just as consistency reveals the most about an archer, it’s the drivers with consistent patterns who are easiest to identify. Once you see which group a driver’s behavior consistently falls into, you know exactly where to focus attention first.
Does 25x mean that higher-risk drivers will definitely crash?
No. The 25x figure describes a group, not an individual. It means that across a driving population, the highest-risk group experiences crashes at 25 times the rate of the lowest-risk group. It doesn’t guarantee that any single driver in that group will crash, just like it doesn’t guarantee that any single driver in the lowest-risk group won’t crash. The purpose of our AI isn’t to predict a crash for one person with certainty; it’s to measure how driver risk is distributed across a population. Discover how the AI is even more powerful at predicting the most serious crashes.
How is the 25x number useful if it doesn’t predict an individual crash with certainty?
Because prevention isn’t about focusing on a specific driver. It’s about knowing where to focus first. Medicine often works in the same way. It’s not about trying to pinpoint individual people who will definitely get ill, but about identifying the risk factors that place people within higher-risk groups.
How was the 25x finding calculated?
Drivers within a population were ranked by their Crash Probability Score, and the crash rate of the highest-ranked 10% was compared against the crash rate of the lowest-ranked 10%. That comparison is where the 25x figure comes from, and this was independently reviewed as part of our AI validation.
Does the 25x finding mean that organizations should pay attention only to the highest-risk drivers?
No, it’s about identifying the most effective place to start to help reduce risk, rather than saying to organizations that it’s the only place they should look. Reinforcing safe behavior in lower-risk drivers matters too. And because risk is dynamic, not fixed, our AI keeps predicting it continuously, rather than measuring it once and stopping there.
FAQs
In the context of Greater Than’s AI validation, what does the 25x finding mean?
The group of drivers identified as highest risk by our AI experiences crashes at 25 times the rate of those identified as lowest risk.
Does this apply to every driving population?
The 25x figure reflects the overall pattern found across our validation. The exact gap can vary between individual fleets, but the underlying principle that risk isn’t evenly spread holds consistently across vast amounts of data from many countries and environments.
Who validated this finding?
The 25x figure was confirmed through an independent technical validation of our AI, conducted by Anders Arpteg, Ph.D., an AI expert with no prior affiliation with Greater Than.
