FMCSA Crash Rates Show Where AI Safety Must Start

A single metric in the latest FMCSA crash data should make every shipper, broker, and fleet leader sit up: some midsize carriers (250–500 drivers) are running year-to-date crash rates above 11% per driver, while the highest large-fleet rates cluster closer to 5–6%. That gap isn’t a rounding error. It’s a signal.

And it lands at the worst possible time of year to ignore signals. Mid-December freight brings peak retail pressure, tight delivery windows, seasonal weather volatility, and a lot of “just get it there” decision-making. When safety becomes a tradeoff against service, the industry pays twice—first in claims and downtime, then again in litigation, insurance renewals, and shipper confidence.

This post is part of our AI in Supply Chain & Procurement series, where the through-line is simple: risk has become a procurement problem. Carrier safety is no longer a compliance checkbox—it’s a sourcing decision. FMCSA crash records show where risk concentrates, but AI is what turns that data into actions procurement and operations teams can actually use.

What FMCSA crash data actually tells you (and what it doesn’t)

FMCSA crash data is an outcome signal, not a blame signal. The Motor Carrier Management Information System (MCMIS) collects state-reported crash records based on police accident reports for commercial motor vehicles involving fatalities, injuries, or towaways. That makes it useful because it’s grounded in real incidents—not just paperwork violations.

But you can’t treat it like a perfectly clean scoreboard.

The data is valuable because it’s outcome-based

Inspection violations can be “managed.” Crash outcomes are harder to hand-wave away. Even though fault and preventability aren’t embedded in the headline numbers, crash frequency is still a practical proxy for exposure to disruption and liability—two things procurement teams care about.

The data has integrity limits you need to design around

FMCSA data comes with known caveats:

• Underreporting can be substantial (the source analysis notes national under-reporting can reach 30–40%, varying by state).
• Normalization can be distorted because driver counts are pulled from carriers’ biennial MCS-150 updates. If a carrier’s driver count is outdated (or underreported), their “crashes per driver” rate can look artificially high.
• Culpability isn’t represented. These crashes reflect involvement, not fault.

Here’s my take: none of those caveats make the data unusable. They just mean you shouldn’t use it as a single gate that approves or bans carriers. You use it as a risk feature—one of several—inside a smarter evaluation model.

The most actionable insight: midsize fleets show higher crash rates

Midsize fleets (250–500 drivers) show materially higher crash rates per driver than large fleets (>500 drivers). The analysis summarized from FMCSA records found:

• Large-carrier “top” crash rates hover around 5–6% year-to-date.
• Midsize-carrier “top” crash rates jump above 7%, with several above 9%, and at least one near 11.87%.

That doesn’t mean “midsize equals unsafe.” It means midsize fleets are more likely to have safety capability gaps—especially gaps that scale poorly.

Why this size band is vulnerable

A lot of fleets grow into the 250–500 range faster than they grow safety infrastructure. That creates predictable failure modes:

• Training programs that worked at 80 drivers break at 300.
• Safety coaching becomes reactive (after an incident) instead of preventive.
• Maintenance scheduling gets squeezed by utilization pressure.
• ELD and compliance monitoring become “handled by a few people,” not a system.

Large fleets often have deeper resources—dedicated safety analytics, standardized onboarding, stronger OEM safety tech penetration, and more disciplined maintenance controls.

From an AI in supply chain risk management angle, this size pattern is gold: it suggests segmentation and benchmarking should be part of any carrier safety score you build. Comparing a 300-driver carrier to a 4,000-driver carrier without context can produce the wrong incentives.

Why public safety scoring still leaves procurement teams exposed

The industry’s biggest blind spot is that the most outcome-relevant metric isn’t fully visible to the market. The FMCSA’s CSA program uses the Safety Measurement System (SMS) to prioritize interventions across multiple BASICs. The Crash Indicator BASIC, which reflects crash frequency and severity, has historically not been publicly displayed.

At the same time, crash preventability can be reviewed internally via the Crash Preventability Determination Program, where carriers submit evidence to remove non-preventable crashes from the internal measure.

The practical problem for shippers and brokers: you’re often left to vet carriers with tools that overweight what’s easiest to measure (inspection violations) and underweight what you actually pay for (crashes, claims, downtime, and lawsuits).

Procurement teams are now operating in a liability environment that punishes “we didn’t know” reasoning. If you source capacity without a defensible safety methodology, you’re carrying reputational and legal risk you can’t spreadsheet away later.

How AI turns FMCSA crash records into preventive action

AI doesn’t “predict crashes” like a fortune teller—it predicts risk conditions that correlate with crashes, and it tells teams what to do next. When you blend FMCSA crash history with operational and behavioral signals, you can build a working system that reduces both incident probability and supply chain disruption.

1) Build a carrier risk model that’s useful, not just “accurate”

A procurement-friendly carrier risk score should be:

• Comparable within peer groups (size band, operation type, geography)
• Fresh (updated monthly or weekly, not annually)
• Explainable (you can justify decisions to stakeholders)

Practical features an AI model can ingest:

• Crash involvement rates (per driver, per power unit, or per estimated miles)
• Inspection patterns (frequency and severity, but not as the only driver)
• Operating geography and seasonality (winter corridors matter in December)
• Cargo type and stop density (regional multi-stop vs longhaul)
• Claims and incident narratives (when available internally)
• Maintenance and telematics signals (hard braking, following distance, speed variance)

A key design point: don’t treat “crash rate per driver” as a verdict. Treat it as a trigger for deeper review.

2) Use AI for “next-best action,” not just scoring

Scores are passive. Actions reduce crashes. AI can recommend operational interventions that fit the shipper-broker-carrier reality:

• Route risk adjustments (avoid high-risk weather windows and terrain)
• Appointment scheduling changes (reduce night driving or fatigue-prone legs)
• Load assignment rules (match high-sensitivity loads to lower-risk carriers)
• Targeted safety coaching (driver-level, lane-level, or terminal-level)

If you only rank carriers, you’ll end up with a brittle network. If you pair ranking with interventions, you improve the whole ecosystem.

3) Make preventability part of your internal truth

One critique in the source analysis is spot-on: outcome metrics matter, but preventability matters too. AI can help here by:

• Classifying incident narratives into preventability categories
• Flagging crashes likely to be non-preventable vs behavior-linked
• Reducing noise from “wrong place, wrong time” events

For procurement, that means fewer false alarms and more confidence when you must defend a carrier choice.

A practical playbook for shippers and brokers (you can start in January)

You don’t need a multi-year transformation to get safer outcomes. Here’s a procurement-and-operations playbook that fits real timelines.

Step 1: Segment your carrier base by operation type and size

Start with three slices:

• Large fleets (>500 drivers)
• Midsize fleets (250–500 drivers)
• Small fleets (below 250; evaluate differently rather than ignoring)

Then segment by operation: longhaul, regional, last-mile, hazmat, tank, passenger, waste, etc. Crash exposure is not uniform.

Step 2: Create a “Safety + Service” sourcing tier system

Build tiers that combine safety indicators with operational performance:

1. Core carriers: preferred lanes, predictable freight
2. Growth carriers: good service, emerging safety maturity—support with interventions
3. Spot carriers: limited use, higher monitoring and constraints

This is where AI in supply chain procurement becomes practical: it’s not about banning half your capacity. It’s about aligning freight to risk tolerance.

Step 3: Add leading indicators (telematics) to lagging indicators (crashes)

Crash data is lagging—by the time it moves, something already happened. If you have access to telematics from carriers (even partial), prioritize:

• Hard braking rate
• Speeding over threshold
• Lane departure events
• Hours-of-service risk patterns (where available)

Then create simple rules:

• If crash risk score rises and hard braking rises, require corrective plan.
• If crash risk rises but leading indicators are stable, investigate data quality (driver count updates, reporting anomalies).

Step 4: Operationalize governance (this is the missing piece)

Set a cadence:

• Weekly: exception review for high-risk lanes and weather windows
• Monthly: carrier tier review and corrective actions
• Quarterly: sourcing adjustments and network redesign

AI can surface the exceptions. Humans still need to run the governance.

People also ask: “Can AI predict the next trucking crash?”

AI can’t predict a specific crash on a specific day, but it can predict which combinations of carriers, lanes, schedules, and behaviors are trending toward higher crash probability. That’s the difference between prediction as a headline and prediction as a tool.

If your goal is fewer incidents and fewer disruptions, that tool is enough.

Where this goes next for AI in Supply Chain & Procurement

FMCSA crash rates are a mirror. They show where the system is strained—especially among midsize fleets—and they underline why procurement can’t separate cost, service, and safety anymore. Carrier selection is risk selection.

If you’re building a more resilient supply chain in 2026, here’s the next step I’d take: treat safety as a measurable supplier risk dimension, then let AI help you monitor it continuously—lane by lane, carrier by carrier, week by week.

The forward-looking question worth asking internally is simple: If a carrier’s risk profile started deteriorating this month, would we see it before the next incident—or after?