Samsara Compounding Risk Report Finds 10% of Drivers Account for ~50% of Crashes

Samsara’s new Compounding Risk Report uses its Risk Model to show crash risk is heavily concentrated among a small subset of drivers and behaviors.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

New analysis from Samsara’s patent-pending Risk Model shows leaders where to prioritize coaching efforts to increase driver safety

SAN FRANCISCO--(BUSINESS WIRE)-- Samsara Inc. (“Samsara”) (NYSE: IOT), the pioneer of the Connected Operations® Platform, today released its Compounding Risk Report, new research based on Samsara’s patent-pending Risk Model that shows the top 10% of risk-ranked drivers account for 47% of crashes. The report evaluates approximately 50 factors spanning driving behavior, exposure, context and driver development. By analyzing how these factors interact over time, the model gives safety leaders a way to prioritize recurring patterns instead of reacting to isolated events.

“By concentrating their most intensive coaching on just 10% of drivers, managers can reach the group associated with nearly half of crashes,” said Arpan Podduturi, Head of Safety Product at Samsara. “From there, a strong safety culture and self-coaching can reinforce safer decisions across the broader workforce — giving managers a way to focus their limited time where it can have the greatest impact.”

The research shows that individual behaviors carry risk, but combinations provide a stronger signal. Drivers exhibiting mobile use alone are 2.7x more likely than the overall driver population to fall into the highest risk tier. Add harsh braking and that figure rises to 4.5x. When mobile use, distraction, and harsh braking appear together, it reaches 5.4x.

Key findings include:

• Risk concentrates among a small share of drivers: The top 10% of risk-ranked drivers account for 47% of crashes, while the top 30% account for 76%.
• Risk compounds when behaviors appear together: Mobile use, distraction and harsh braking in combination are associated with a risk concentration twice as high as any of those behaviors measured alone.
• Behavioral patterns persist over time: When comparing a driver who is subsequently involved in a crash with one who is not, the model prioritizes the crash-involved driver approximately three out of four times. This holds across both next-day and seven-day evaluation windows.
• The most consequential risks are coachable: Separate causal analysis identified aggressive driving, distracted driving and speeding-related patterns as the most consistent and credible behavioral contributors to crash risk. Conditions such as night driving, freezing temperatures and urban exposure can amplify risk but are not coaching targets on their own.

Samsara’s Risk Model is designed to rank drivers by relative risk, not predict whether a specific driver will crash on a specific day. That distinction allows safety leaders to use the model as a prioritization tool — focusing limited coaching resources on the behavioral patterns most associated with elevated risk.

The model powers Coaching Priority, Samsara’s AI-powered feature for surfacing and prioritizing coaching opportunities across a fleet. Coaching Priority evaluates approximately 50 risk factors and consolidates these signals into a single view that shows managers where attention may have the greatest impact, while automated self-coaching helps reinforce safer habits across the broader workforce.

“Coaching Priority shows us exactly where to focus — down to the specific locations, behaviors, and drivers driving risk. That clarity empowers us to take targeted action and elevate safety across our entire fleet,” said Tom Karnowski, Vice President, Environmental, Health, and Safety at USIC.

The Compounding Risk Report is based on aggregated data used to train and evaluate Samsara’s Risk Model from July 1 through December 15, 2025. The dataset spans drivers and fleets across multiple industries and regions. Combined-profile findings describe statistical associations rather than causation. Statements identifying certain behaviors as credible contributors to risk are based on separate causal analyses using double machine learning with calibration and causal forests for policy ranking.

The full Compounding Risk Report, including detailed methodology and recommendations for applying the research to fleet safety programs, is available here.

Combined-profile findings reflect statistical associations and do not predict individual crash outcomes. See the full methodology in the report.

About Samsara

Samsara (NYSE: IOT) is the pioneer of the Connected Operations® Platform, which is an open platform that connects the people, devices, and systems of some of the world’s most complex operations, allowing them to develop actionable insights and improve their operations. With tens of thousands of customers across North America and Europe, Samsara is a proud technology partner to the people who keep our global economy running, including the world’s leading organizations across industries in transportation, construction, wholesale and retail trade, field services, logistics, manufacturing, utilities and energy, government, healthcare and education, food and beverage, and others. The company’s mission is to increase the safety, efficiency, and sustainability of the operations that power the global economy.

Samsara is a registered trademark of Samsara Inc. All other brand names, product names, or trademarks belong to their respective holders.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260908845426/en/

Media Inquiries Contact
media@samsara.com

Source: Samsara

Key Terms

double machine learning technical

A statistical method that combines flexible machine-learning models with econometric techniques to estimate causal effects from observational data while guarding against bias from overfitting. It does this by first using machine learning to predict nuisance parts of the data, then “orthogonalizing” the main effect and repeating the process with cross-fitting so errors in the machine learning step have little impact on the final estimate. Investors care because it helps measure the likely impact of actions, policies, or signals in messy, high-dimensional data the way a neutral referee separates play from noise to judge the real effect.

causal forests technical

An ensemble machine‑learning method that builds many decision trees to estimate causal effects and how those effects vary across individuals or groups. Causal forests aim to separate correlation from causation by comparing similar units and averaging over many tree-based partitions, so they reveal which subgroups respond differently to an action, policy, or treatment. For investors, that matters because the technique helps interpret whether observed relationships reflect real causal impacts and where effects are strongest, similar to running many small, tailored experiments to find who benefits most.