Cameras capture events. Sensors detect harsh braking. Algorithms generate scores. All of it is useful. None of it, on its own, changes how a driver drives.

I have spent years studying why fleet safety programs plateau. The answer is not that fleets lack data. Most are drowning in it. The answer is that data, on its own, does not change how a driver thinks. And until you change how a driver thinks, you have not changed how a driver drives.

Here is the tension worth sitting with. Over the past decade, fleets have spent heavily on visibility, yet the industry’s safety numbers have not followed. Between 2014 and 2022, as GPS telematics and video adoption climbed toward record levels, the fatal-crash involvement rate for large trucks rose roughly 31 percent, and the number of people killed in crashes involving large trucks rose about 52 percent (FMCSA and NHTSA). Buying the technology has not, by itself, made fleets safer. The fleets that did get safer used the technology for something more.

The Evidence Is Not Ambiguous

Every serious study reaches a version of the same conclusion: monitoring alone produces meaningful but limited results, and adding structured behavioral coaching roughly doubles the outcome.

One of the largest video telematics providers published data showing that fleets using cameras and telematics without consistent coaching saw a 37 percent reduction in collisions, while fleets adding consistent coaching to the identical technology saw a 73 percent reduction.

Putting aside the need to take these self-reported numbers with a grain of salt, this does show that even these technology companies’ own numbers validate this fact.  The technology was the same in both groups. The only variable was whether anyone used the data to train and coach.

Peer-reviewed research confirms the more modest baseline. A 2025 AAA Foundation randomized controlled trial found telematics feedback reduced speeding 11–13 percent and hard braking 16–21 percent, translating to roughly a 4–5 percent reduction in crash-claim frequency. A real effect, and an order of magnitude below vendor marketing. Which makes the coaching gap—the distance between 37 and 73—the most important number in the industry.

Peer-reviewed research confirms the more modest baseline but tells the same story. The NETS Strength in Numbers benchmark, aggregating safety performance from dozens of large corporate fleets, confirms the pattern:

• 3–4 percentage points lower collision involvement for fleets with closed-loop corrective-action follow-through
• About 3 percentage points lower for fleets using in-person behind-the-wheel training versus remote or digital-only
• 20–30 percent collision-rate reduction for fleets combining telematics with active coaching programs

This makes the coaching gap—the distance between 37 and 73—the most important number when considering how to get the most value out of your technology investment.

 Why “More Coaching” Is Not the Answer Either

At this point, the obvious conclusion is: add coaching. That is directionally right. But the kind of coaching matters enormously.

Most fleet coaching today works like this. A manager gets an alert. They sit down with the driver and say, “You had 14 harsh-braking events this month. That needs to come down.” The driver nods. They try harder for a week and improve. But the events come back.

The problem is that the coaching is not connected to anything the driver was ever taught. The driver hears a number, not a principle. They do not learn that harsh braking is usually the downstream consequence of failing to scan the traffic environment far enough ahead. That connection—from the event the technology captured, to the decision that caused it, to a principle the driver learned in training—is the through-line that separates programs that plateau from programs that keep improving.

The newest version of this problem wears a more sophisticated face. Telematics and camera vendors are now rolling out AI-generated coaching: automated feedback videos, AI avatars that deliver personalized weekly debriefs, in-cab audio prompts, and machine-written talking points for managers. It is a real advance in efficiency. But it is not a solution to the underlying problem, because it starts in exactly the same place event-based systems always have: with a flagged event and a number on a dashboard.

AI coaching can tell a driver to reduce hard braking or improve a following-distance score faster and more cheaply than a person can. What it cannot do, any better than a rushed manager can, is connect that correction to a driving principle the driver was actually trained in. The delivery got cheaper. The mechanism of behavior change did not change.

A 2025 study from the University of California, San Diego, funded by one of the video telematics vendors itself, concluded that manager-led, personalized coaching remains the most effective method for durable behavior change, and warned that automated coaching without real human interaction can reduce driver engagement and trust. Automating a disconnected conversation makes it faster. It does not make it work.

The fleets that break through the technology ceiling share a common characteristic: they operate from a named, structured behavioral framework. Not a set of rules. Not a list of violations. A defined set of driving decisions that drivers learn, coaches reference, scorecards measure, and corrective action reinforces—the same language at every stage.

At Smith System, we have spent more than 70 years building and validating that framework, the Smith5Keys. But the principle is bigger than any one provider. Any fleet that wants to break through the monitoring ceiling has to answer one question: does your program have a shared behavioral language that connects training, coaching, scoring, and corrective action? If the answer is no, the evidence says you are leaving half the outcome on the table.

What Happens When the Loop Actually Closes

A national food and beverage retailer operating 22 distribution centers with more than 2,200 drivers had telematics in place before it connected a behavioral framework to the program. After integrating behind-the-wheel training, behavioral scoring, and coaching referencing the same driving decisions, preventable accidents dropped 38 percent over two years. Average indemnity cost per incident fell from about $43,000 to $1,134.

The same hardware was generating the same signals. What changed was that those signals began flowing through a framework that translated them into specific, coachable decisions.

The most durable evidence comes from a state transportation department that adopted the same framework through a peer-led, train-the-trainer model. Annual incidents fell roughly 30 percent in the first year and held there as the workforce grew. Then the program was paused during the pandemic, and incidents climbed back to baseline. When training resumed, they dropped again—and the fleet’s best year on record came several years into the program, not at the start.

There is no novelty effect that lasts six years, disappears when you stop, and returns when you restart. That is what a sustained behavioral system looks like: gains compound while you maintain them, and reverse when you do not.

Five Questions for Your Own Program

Regardless of who built your program or what technology you run, these questions will tell you whether you are capturing the full outcome or leaving results on the table:

1. Can your coaching language trace back to something the driver was trained in? Coaching that does not reference training is a conversation about numbers. Coaching that reinforces training is a conversation about principles.
2. Does your scorecard tell you which decision to coach, or just which driver to worry about? If it flags a high-risk driver but cannot tell you whether the problem is scanning, space management, or situational awareness, you have a ranking tool, not a coaching tool.
3. When a driver completes corrective action, do you verify that the targeted behavior actually changed? Completion rates are an administrative metric, not a safety metric. An open-loop system assumes training works without ever checking.
4. Can you identify where risk is concentrating across your fleet by behavior—not just by driver? A systemic following-distance problem needs a different response than a speeding problem. Treat the root cause at scale, not symptoms one driver at a time.
5. Has your collision-rate improvement continued past the first year, or has it plateaued? Every monitoring program tends to have a strong first year. If your improvement curve has flattened, you have likely reached the ceiling of what observation alone can deliver.

The Other Half

The fleet safety industry has spent the last decade investing in visibility. But visibility was half of the problem. The other half—turning what the technology sees into behavior change that compounds over time—requires a behavioral framework that drivers learn, coaches reference, scorecards measure, and programs reinforce.

The safest fleets are not the ones with the most advanced hardware. They are the ones that built the bridge between what the technology sees and what the driver does. The evidence on that point is not ambiguous, and it is not mixed. The first half of the investment has been made. The question for every fleet safety leader is whether the second half is in place, or if it is still waiting to be built.

Editor’s Note: This contributed article reflects the author’s opinion on the role of technology, training, and coaching in fleet safety and does not necessarily reflect the perspectives of Automotive Fleet.