Why programs stall
The failure is almost always operational rather than technical. The system works. The workflow around it doesn't.
- Alert volume exceeds review capacity, so reviewers triage by skipping.
- False positives train reviewers to distrust the feed, and eventually to ignore it.
- The gap between event and conversation stretches to weeks, by which point the driver doesn't remember the moment.
- Coaching gets delivered as discipline, so drivers optimize for not being flagged rather than for driving better.
- Nobody owns the workflow, so it happens when someone has a slow afternoon.
If your reviewers are behind, adding coaching sessions won't help. Cut alert volume until the queue is clearable, then rebuild.
1. Tune the triggers before you tune the drivers
A program that flags every hard brake in city traffic is flagging normal driving. Reviewers learn this within a week and start clearing events without watching them, which is the same as having no program while paying for one.
Start narrow. Pick the two or three behaviors most correlated with the crashes you have, usually following distance, distraction, and speed relative to conditions, and set thresholds strict enough that a flag means something. Widen only when the queue is consistently clear.
2. Close the loop in under 48 hours
This is the highest-leverage variable in the whole program. A driver shown a clip from yesterday can reconstruct the moment: the traffic, the weather, what they were thinking. A driver shown a clip from three weeks ago is being asked to defend a stranger.
Fast feedback also changes the emotional register. Same-week coaching reads as attentive. Month-old coaching reads as a case being built.
Recency does more work than eloquence. A mediocre conversation two days later beats a great one a month later.
3. Coach the pattern, not the event
One hard brake is noise. Three hard-brake events in the same interchange at the same time of day is information, and it might not even be a driver problem. It might be a routing problem.
Framing matters here. "You had a hard brake on Tuesday" invites a defence. "You've had four following-distance events this month, all in stop-and-go, here's the pattern" invites a conversation, and gives the driver something they can change.
- Group events by behavior type before reviewing, rather than by chronology.
- Look for route and time-of-day clustering before assuming it's the driver.
- Compare the driver against their own past trend rather than the fleet leaderboard.
- Bring one clear example to the conversation, not a dossier.
4. Make the program visibly two-sided
Drivers accept a system that also works for them. So the exoneration wins have to be as visible as the flags, and ideally more visible. The first time footage clears a driver of a claim, tell the whole fleet.
It also means recognizing improvement rather than only escalating failure. A driver who cut their event rate in half has done exactly what the program asked, and if the only communication they ever get is a flag, they'll conclude the program has one direction.
- Publish exonerations. Every one.
- Recognize improvement trends, not just top-of-leaderboard performance. The driver who went from worst to average did more work than the one who was always good.
- Be explicit and written about what triggers a recording and who can view it.
- Never repurpose footage for something outside the stated policy. One breach ends the trust permanently.
Who should do the coaching
Whoever has an existing relationship with the driver. A safety manager the driver has met twice will get a defensive conversation. A dispatcher or driver manager they talk to daily will get an honest one.
That argues for distributing coaching rather than centralizing it, with the safety team owning the standard, the queue, and the escalation path rather than every conversation. It scales better and it lands better.
Metrics that tell you it's working
Activity metrics like events flagged and sessions held measure effort. These measure results.
- Repeat-event rate. Of drivers coached on a behavior, what share repeat it within 60 days? This is the core measure.
- Time from event to conversation, tracked as a median rather than an average.
- Fleet-wide events per million miles, trended monthly.
- Distribution shift. Is the tail of high-event drivers shrinking, or is the average being carried by drivers who were already safe?
- Preventable incident rate, the outcome the whole program exists to move.
