Ops Knowledge Garden

Fleet Safety

Living noteLast tended 5 min read

Fleet Safety

During the first year of this project, the rate moved from 1.0 to 0.7 accidents per 1,000 trips. This is the method we used and what I think it taught me.

The Problem With How EMS Handles Accidents

The reactive pattern I saw in EMS was familiar: something happens, someone fills out a form, maybe there is a conversation, maybe there is retraining, and then everyone moves on and waits for the next one.

The form is not useless. It is usually doing the job it was designed to do: notify the organization and preserve a specific set of facts. The problem is expecting that same form to carry the entire investigation. A record can be accurate and still omit the conditions needed for operational learning.

We needed a better diagnosis than “the operator made an error.” That description may be accurate and still leave the conditions around the event unexplored.

The Framework: DMAIC

We used DMAIC — Define, Measure, Analyze, Improve, Control — borrowed from Lean Six Sigma. You don’t need a Six Sigma certification to apply it. You need discipline and honest data.

Define: What exactly are we trying to reduce? We got more specific than “accidents” in general. Backing incidents, intersection incidents, and low-speed lot incidents can involve different conditions. Breaking the total apart gave us a better place to investigate.

Measure: What did we actually have, and how were we capturing it? We implemented fleet telematics—GPS, accelerometer data, hard-braking events, speed profiles. It added objective signals alongside incident reports and witness accounts. It also showed driving events that had not become reported collisions, giving us more to examine than the outcome count alone.

Analyze: Where were incidents clustering? What conditions preceded them? Time of day, shift length, call volume, specific units, specific locations. We ran fishbone analysis on the high-frequency incident types. The contributing conditions were almost never “operator error” alone—they were layered.

Improve: Targeted interventions based on the contributing conditions we could support. This included:

  • A behavioral coaching program for high-incident operators (not punitive — coaching)
  • Deployment videos for consistent backing and spotting procedures
  • Spotter protocol enforcement at specific high-risk locations
  • FTO integration so new providers learned safe habits before bad ones formed

Control: How do you hold the gains? Monthly review of telematics data, leading indicators tracked alongside lagging ones, and building fleet safety into regular supervisor conversations rather than treating it as a separate safety-department function.

The Telematics Layer

Telematics changed what was possible. Before it, we relied heavily on incident reports and witness accounts. Those records remained important, but telematics added a different view: hard-braking events, speeding patterns, and backing behavior across the fleet in near-real time.

The surveillance concern is real. The distinction is what the organization does with the data. Used primarily as a disciplinary feed, telematics teaches people to fear the measurement. Used as one source in a transparent coaching and safety process, it can make behavior visible that no supervisor could observe across every unit and shift.

The key is what you do with the data. Raw data without structured review and coaching is just noise. We built a review cadence — supervisors were looking at their units’ data, not just waiting for incidents to come to them.

What I think mattered

Backing incidents were the highest-frequency category. We did more than issue another reminder about spotters. We reinforced the behavior at the crew level and used telematics to make backing patterns available for review.

My strongest recollection is that the coaching conversations changed the quality of the work. Sitting with an operator and looking at their own data as a shared safety problem produced a different conversation than beginning with a disciplinary write-up. I cannot isolate how much of the measured reduction came from coaching, spotter practice, training, telematics, or other changes happening at the same time.

The 30% change was measured: the rate moved from 1.0 to 0.7 accidents per 1,000 trips during the first year. The process is documented well enough for another team to adapt and test. The result is not a promise that another fleet—or even the same fleet in another year—will produce the same number.

The Deeper Point

An accident can involve conditions beyond the individual operator: equipment, fatigue, deployment, training, local layout, and the way a policy meets the actual work. Those conditions do not erase individual decisions. They give the investigation more places to look and the organization more possible controls than discipline alone.

This project is where the sprinkler-system idea became measurable for me. It is also part of the lineage of chatIR, which is my attempt to make this kind of structured incident learning easier to repeat without pretending the software can conduct the human investigation.

The Vehicle Accident Investigation Form and Incident Findings Report are the paper companions to this method.

Graph View

Backlinks