Harsh braking, mobile usage, distracted driving, speeding, rolling stops and following distance are all driver behaviors that can increase the risk of crashing a vehicle. If a driver in a vehicle equipped with a Samsara camera looks away from the road, the AI detects that and will give an audible alert like, “Put down phone.”
Safety systems have focused on alerting drivers in the cab, in real time when they trigger an individual safety event, but those isolated behaviors don’t paint the full picture of risk, said Samsara Head of Safety Product Arpan Podduturi. Instead, the riskiest driver profiles are identified when individual behaviors compound.
Mobile phone usage is only 1.7 times more likely to cause a crash. Mobile use combined with harsh braking and night driving, however, is 5.4 times more likely to result in an accident.
That’s according to Samsara’s recently released Compounding Risk Report that determined the top 10% of a fleet’s riskiest drivers account for nearly half of all crashes, and the top 30% of a fleet’s riskiest drivers account for 76% of crashes.
"The reason the top 10% accounts for so many crashes is precisely that those drivers tend to exhibit combinations of behavioral patterns — not single events," Podduturi said in the report. “Even modest gains in coaching the top decile can translate into meaningful reductions in crashes at scale."
Coaching priority
These statistics were mined from Samsara data collected between July 1 and Dec. 15, 2025, using the company’s proprietary, patent-pending Risk Model, which informs Samsara's AI-powered Coaching Priority framework that surfaces and prioritizes coaching opportunities across fleets. The Risk Model evaluates approximately 50 factors across four categories:
- Driving behavioral patterns: speeding tendencies, mobile usage, harsh events, etc.
- Exposure: how much time a driver spends on the road, including miles driven and driving concentration
- Context: the environment in which driving occurs, including road type, urban or rural setting, nighttime conditions, weather, etc.
- Developmental factors: driver tenure, coaching history, etc.
Prioritizing patterns
Alerting drivers to their risky behavior in the moment with in-cab alerts has helped improve safety over the years, but that is reactive. Podduturi said the next step is proactive. That means getting better at recognizing the behavioral patterns that precede a crash and surfacing them early enough to potentially change that outcome.
“Drivers involved in crashes usually aren’t having a sudden bad day. They are driving in identifiable patterns over time, and those patterns can be visible weeks in advance,” Podduturi said in his report. “Identifying which drivers are most at risk is valuable. Identifying them early — before an incident occurs — is what makes that identification actionable.”
Safety leaders can use Coaching Priority to identify the behavioral patterns most associated with crashes and address those behaviors in the top 10% of riskiest drivers to potentially prevent up to 47% of accidents, according to the report.
In a shift away from individual events, the Risk Model reads tendencies — a consistent pattern of driving behavior across trips and over time, often in conditions that amplify impact. Single behaviors carry risk, but none on its own is a strong basis for prioritization, which is the highest-leverage move a fleet can make, Podduturi said.
This means a driver doesn’t get flagged for braking hard once. Instead, they get flagged for a pattern of how they drive.
Perfecting prioritization
According to the report, the Risk Model prioritizes a driver involved in a crash three out of four times in advance of the accident, compared to a driver who is not involved in a crash.
Podduturi said that prioritization will become even more precise as data depth grows, further increasing Samsara’s ability to see how driving behaviors interact and how context amplifies risk. Future iterations of the Risk Model will incorporate more granular behavioral signals, sharper context modeling and improved interpretability, he added.
“Samsara’s Risk Model is not designed to predict whether a specific driver will crash on a specific day. Crashes are rare and depend on many factors outside any model's view,” Podduturi noted in the report. “The model is designed to rank drivers by their relative risk based on the behavioral patterns they exhibit — and to do so consistently enough, far enough in advance, that fleets can act on the information.
“Prioritization, not prediction, is where pattern-based risk identification creates value.”
