A sudden brake at highway speed sends a spike through a telematics dashboard before the driver’s foot has fully come off the pedal: the accelerometer reading gets smoothed by a filter, compared against a g-force threshold the vendor picked for you, and filed as a red row in a weekly harsh event report that somebody will eventually turn into a coaching conversation. Most coaching programs go wrong somewhere between that row appearing and the conversation happening. I have been reading these reports on my own driving for a few years now, and I still can’t always tell which rows describe a habit of mine and which describe the world happening at me.
This post is about how I read those reports now, and what I think a fleet lead should do with them before sitting down with a driver.
Where my own data confused me
My car came with a connected app that keeps a driving summary, and after long trips I open it on my phone. I skip the overall score, because a composite number squeezed onto a phone card tells me nothing about what to change. The harsh brake count broken out by day is more useful, since one bad stretch of road can drag a whole week into the red while the daily view at least lets me find the day that did it. Then I try to remember what actually happened on that day. There is more on that kind of after-the-fact reconstruction in a post on using trip replay to work out what went wrong on a delivery, and the habit carries over. Sometimes the memory comes back right away, and sometimes nothing comes back at all.
When it comes back, it is usually something outside my control, like a rideshare car stopping dead to collect a passenger or a cardboard box sliding out of a pickup bed on a rural two-lane road.
The brakes I can’t remember worry me more. A harsh brake I have no memory of probably didn’t feel unusual at the time, which suggests it is something I do on autopilot, and that is the kind worth working on. The data can’t point it out for me, though, because a brake caused by following too closely and a brake caused by a deer look identical in the log. I was sitting in the seat for every one of those events and still struggle to sort them, so a safety lead reading hundreds of rows a week from other people’s driving has a much harder job than the dashboard makes it look.
Start with the map, then the names
A typical harsh event report gives you a driver ID, a timestamp, a location, a severity bucket, sometimes a video clip, and a number for how hard the deceleration was. The common mistake is to treat each row as a finished verdict, as if the platform had already interpreted it and your job was just to deliver the news.
Group events by location before you group them by driver. If several unrelated drivers keep braking hard at the same intersection, the intersection is the thing to look at, and coaching each of them separately will cost you trust you spent months building. A blind merge or a badly timed light is cheaper to report or route around than a round of near-identical conversations with everyone who happens to drive through it.
Then divide by distance or driving hours before you rank anybody. A courier working dense city blocks will log far more braking per mile than someone running long highway stretches, and ranking them on raw counts mostly measures their routes.
Telling a habit from bad luck
Once the numbers are normalized, the question is whether a driver’s higher rate is a habit or a run of bad luck. Virginia Tech’s transportation institute has studied coaching programs for years, and the point I took away from their writeup on hard braking is that monitoring by itself does little unless somebody follows up with the driver afterward.
A single harsh event, even a severe one, is usually situational. Bringing a one-off to a coaching meeting tells the driver they are being watched, and a defensive driver spends the rest of the meeting explaining instead of listening.
There are two signs I look for before I’d call something a pattern. The first is repetition across different conditions, where the same driver brakes hard on different routes and in different traffic, so the road changes and the driver is the part that stays the same. Harsh acceleration and harsh braking showing up close together is the second, and it usually means following too closely: the driver speeds up hard to close a gap, the gap closes faster than they expected, and they brake hard, and the report records that single habit in its braking column and again in its acceleration column. When neither sign is there, I’d file the event and move on. If both are there, and the weekly rate has been climbing for over a month on a similar route mix, that is worth a conversation.
Comparing before and after without fooling yourself
Most before-and-after comparisons I have seen are close to meaningless, and I’ve made the same mistake reading my own numbers. Someone pulls a driver’s raw event count for the month before a coaching talk, pulls it again for the month after, and celebrates or despairs depending on the difference. That difference is mostly noise from changed routes, weather, holidays and the plain fact that an unusually bad week tends to be followed by a more normal one whether anyone intervened or not. A comparison that means something uses a rate instead of a count, over several weeks on each side of the talk and on a similar mix of routes.
Here is what that can look like with made-up but realistic numbers. Say a driver averages about 2.5 harsh brakes per 100 miles before a talk about braking late on downhill grades, and on a similar route mix afterward the rate drops to under 1, with most of the remaining events bunched around a construction zone that was added to the route halfway through. That is a real improvement. The construction zone probably can’t be coached at all, and pushing the driver further on it would undo some of the goodwill the first talk earned.
Why two vendors disagree about the same brake
This part surprised me the first time I pulled raw data. Accelerometer data is extremely noisy, and every pothole and speed bump shows up as a spike, so most platforms run the stream through a low-pass filter (often a moving average over a fraction of a second) before deciding anything was a harsh event. The g-force figure the dashboard shows you is the peak of that smoothed signal, which can be noticeably lower than the actual peak during the brake. A short, sharp stab at the pedal gets averaged down more than a long, firm stop of similar strength, so two brakes that felt very different from the driver’s seat can land in the same severity bucket. Vendors also pick different window sizes and different thresholds, and I have rarely found either written down anywhere a customer would look, which means comparing harsh event counts across two telematics products is comparing two different measurements of the same driving. Once I understood that, I stopped being surprised when the app in my car flagged brakes that a friend’s plug-in tracker, sitting in the same car on the same road, ignored completely.
The conversation with the driver
If you walk in and read the report out to the driver, you have lost them before you finish. Drivers remember the moments you’re asking about, and sometimes the brake in question was exactly the right response to something a row in a spreadsheet can’t show.
Start with the trend instead of a single incident. Something like “your harsh brake rate has been going up over the last few weeks and I want to understand what changed” is about the numbers moving, and it gives the driver room to explain. Ask before you explain anything, because often the driver already knows what’s going on (a new stop with no parking, a tighter schedule since the route was reshuffled) and has been thinking about it longer than you have. If there is video, show one clip and watch it together, asking what they saw in the seconds before the brake. A reel of someone’s worst moments feels like an attack no matter how you meant it. Finish with one specific change and a date to look at the numbers again.
Then actually look at them on that date. Drivers notice whether the follow-up happens, and if it doesn’t, the next coaching talk carries less weight because everyone has learned it was for show. Records from a program that really follows up can also help outside of safety, which comes up in the post about fleet data and insurance negotiations, though how much an insurer will give you for them varies a lot.
What I still can’t answer
I still open that app after long drives, and I still can’t sort every event into my fault or the road’s. For a single brake, I’ve mostly stopped trying. I look at whether the weekly rate is rising or falling, and whether the bad days line up with roads I drive every week or with odd one-off trips. At fleet scale the same approach applies, except getting it wrong costs you drivers instead of a bit of my pride, and when the trend is unclear it is fine to tell the driver you aren’t sure yet and want to check again in a couple of weeks.
The part I keep going back and forth on is the filter itself. Every dashboard I’ve seen hides the window size and the threshold, so the line between “harsh” and “normal” is a decision somebody at the vendor made and never wrote down where a fleet lead could find it. If you run a coaching program, would you want to know where your vendor drew that line before you held anyone to it?