Most carriers know their CSA percentiles roughly the way they know their credit score: a number somebody told them once, vaguely bad news if it goes up, and no clear idea who is looking at it.
That vagueness is about to get expensive, because FMCSA is in the middle of rewriting the math that produces those numbers. The agency detailed a revised Safety Measurement System methodology in a Federal Register notice in November 2024 and has been phasing it toward production since. Same inspection data, different arithmetic — which means a percentile can move without anything changing in your yard.
So it is worth being precise about what these numbers do.
SMS is not a safety rating
This is the confusion that causes the most bad decisions, so get it straight first.
Your safety rating — Satisfactory, Conditional, Unsatisfactory, or none at all — comes from a compliance review: an investigator looking at your operation. It is a formal finding with a formal process behind it.
Your CSA percentiles come from the Safety Measurement System, which sorts roadside inspection results and crash records into compliance categories and ranks you against other carriers. No human decides it. It is a triage tool, built so FMCSA can point finite investigators at the carriers whose data looks worst.
A high percentile does not revoke anything. It raises your odds of getting attention. Those are different problems with different fixes.
What it genuinely affects
Three things, and only one of them is the government.
Enforcement attention. Percentiles are what prioritize you for interventions — a warning letter, a targeted roadside focus, an investigation. If you are above the threshold in a category, you have moved up a queue.
Whether brokers will book you. This is the one that hits the bank account, and it has nothing to do with FMCSA. Percentiles in the public categories are visible on FMCSA's SMS site, and every carrier-vetting service scrapes them. A broker's system flags you before a human ever reads your name. Some shippers write a percentile ceiling into their carrier agreements.
What insurance costs. Underwriters treat the data as a loss predictor, and they are not obliged to explain how much weight they give it. It sits alongside your loss runs at renewal.
Notice the pattern: two of the three consequences are commercial, arrive without notice, and have no appeals process. The regulator will at least send you a letter.
What it does not affect
It does not, on its own, pull your authority. It does not fine you — the ticket at the scale fines you, and the violation then also lands in the data. It does not determine your safety rating. And it is not a verdict on whether your drivers are safe; it is a percentile against peers, which means the number can move because other carriers changed, not because you did.
That last point matters more under the revised methodology. Part of what is being reworked is how carriers are grouped for comparison — so a small fleet may find itself measured against a different peer set than before, with no change in its own record.
Read your own data before someone else does
You have access to everything a broker sees, and more of it than they get.
- Pull your SMS profile. Look at it by category, not as one blob. A bad Vehicle Maintenance number and a bad Unsafe Driving number are completely different diseases with completely different treatments.
- Run the Prioritization Preview. FMCSA published a preview tool that shows your own data scored under the revised methodology next to your current profile. Compare the two yourself. If a category is going to look worse, you want to find that out before a customer does.
- Check every inspection against your own records. Wrong USDOT number, a violation attributed to the wrong carrier, a citation that was later dismissed — these happen. They are correctable through DataQs, and only if you notice.
- File DataQs challenges quickly. Old requests are harder to substantiate, and the data is doing damage the whole time it sits there.
The unglamorous part actually works
There is no trick here. The system weights recent events more heavily than old ones, so the mechanism that fixes a bad percentile is a run of clean inspections while the old ones age out. That means the boring stuff: pre-trips that are actually performed, lights and tires and brakes caught in your yard instead of at a scale, defects closed out with a record of the repair, and drivers who know that an inspection report is a document with consequences rather than a receipt.
Vehicle maintenance violations are where most small fleets bleed, and they are the most controllable category on the list. A tire or a lamp found in your own shop is a repair. Found at a scale, it is a repair plus a data point sitting in your record where brokers and underwriters can read it.
Our guide to DVIR and pre-trip inspections covers what actually has to be recorded, and what FMCSA registration requires from you now covers the other compliance record a broker can check on you.
Where the software fits
The connection between a defect and a percentile is a paperwork trail, and paperwork trails are where this falls apart — a driver reports a problem, nobody closes it out, and the same item shows up at a scale three weeks later.
In TFS Fleet, pre-trip and walk-around inspections are submitted with photo evidence and go through an approve-or-reject workflow, work orders carry their own parts and labor lines, and dispatch can be gated on a truck that is overdue for a walk-around — so an open defect is visible rather than filed. See Maintenance and Inspections and Analytics for the reporting side.
The short version
CSA percentiles are a triage score, not a verdict — but the people who use them commercially treat them like a verdict, and they do it silently. The math behind them is being rebuilt right now, which is reason enough to look at your own numbers this month instead of finding out from a broker's rejection. Pull the profile, run the preview, challenge what is wrong, and fix what is yours.