The camera cannot change your CSA score

Good morning.

Last Tuesday I said I would ask what AI driver coaching actually changes — the score, the premium, or just the volume of video somebody still has to review.

The answer turns on one sentence in the federal regulations, and it is not the one the pitch decks quote.

What is being bought is video-based safety: cameras that watch the driver, score the driving, and queue clips for coaching. The pitch is almost always the same — fewer incidents, better CSA score.

Half of that is plausible. The other half is not how the score works.

A hard-braking alert is neither. It cannot enter your score, because there is no door for it.

That does not make the cameras worthless. It makes the most common reason for buying one wrong.

INDICATIVE FIGURES. NOT A SUBSTITUTE FOR PAID INTELLIGENCE.

Sources: FreightWaves, Overdrive, Transport Topics, CDLLife.

What the camera can touch, and what it can’t

Your CSA score is assembled from two inputs: roadside inspection violations and DOT-recordable crashes. That is what Appendix B describes, and it is why the only events that move your score are events a government employee recorded.

An in-cab camera produces neither. It produces a vendor’s score — proprietary number, proprietary scale, computed by the company selling you the camera. Useful internally. Invisible to FMCSA.

So the chain from camera to CSA is one link long and slow: the camera changes behaviour, the behaviour prevents a crash or a violation, and the absence of that event stops making your score worse. You are not improving a score. You are declining to damage it, then waiting for the old damage to age out.

The benefit that arrives fastest is the one vendors mention second: footage after a crash. When a four-wheeler brakes in front of your truck and the police report says otherwise, the video is the whole case. That is a litigation asset, worth buying on its own terms.

The camera generates events, and every event is a decision somebody makes and a conversation somebody has. A system that finds more problems than you have hours to address does not reduce risk — it documents it, in a timestamped record a plaintiff’s lawyer can subpoena.

Keller sells advisory services, so read the remedy with that in mind. The diagnosis still holds.

AI driver coaching — the category, not a vendor

Buy it for the right reason and it earns out. Buy it for the reason on the slide and you have bought a very expensive way to feel monitored.

Buy it for exoneration. Footage that survives a disputed crash is the clearest return in the category, it arrives the first time you need it, and it does not depend on anyone changing behaviour.

Do not buy it to move a CSA score. There is no mechanism. The score reads inspections and crashes; the camera writes to a vendor dashboard.

Price the review labour. Multiply that first answer by the minutes it takes to watch a clip and have the conversation. That number, not the subscription, is the real cost.

Unreviewed events do not disappear. They sit in a system of record showing you were told and did nothing — the risk that gets sold to you as a benefit. And get the premium answer in writing: vendor case studies are not underwriting.

Registry note: FMCSA removed five more devices from the registered ELD list. Same check as Issue #15 — device name, model number, software version and ELD identifier, all four.