Skip to content
Above the Threshold

All notes  /  Buying

Choosing a System

Products differ in what they measure, how they score and what they let you change. The questions that surface each.

Buying · Procedure

Every product in this market detects harsh events and produces a score. The differences are underneath.

The measurement questions

Is speed read from the vehicle or derived from position?

Is the accelerometer in the device or the vehicle, and how is mounting specified?

How is calibration performed, and can it be re-run?

Which parameters come from the vehicle bus rather than being inferred?

A vendor who answers these vaguely is selling an inference engine, and the outputs will be contestable.

The scoring questions

Show me the formula. Weights, exposure basis, normalisation.

Can the weights be changed?

Can thresholds be set per vehicle type?

Is there a minimum exposure below which no score is produced?

If the model is proprietary, the score cannot be explained to the person it describes, which is the position you do not want in a grievance.

The video questions

Triggered or continuous, and can continuous be disabled?

Is live viewing possible, and can it be turned off at account level?

Is audio recorded, and can it be disabled?

Is viewing logged, and can drivers see when their footage was watched?

Retention per camera direction, separately configurable?

The trial

On your hardest route, with real drivers who know it is a trial.

A month of silent logging, then look at the distribution.

Check a sample of speeding events against the actual limits, which finds the map database errors.

Review twenty clips and count how many are false triggers.

Deliberately drive a known defensive manoeuvre and see what the system makes of it.

Red flags

Scoring presented as the product rather than as one output.

A league table as a headline feature.

Continuous cab recording as the default.

Gamification modules.

"Proprietary risk model" as an answer to how the score is built.

The contract

Data export in a documented format, including the raw events rather than only the scores.

Notification before changes to the scoring model, which otherwise moves under you and makes trends meaningless.

Retention terms matching yours, with deletion confirmed on termination.

Ask for the formula in writing

The request that sorts this market fastest.

Weights, exposure basis, normalisation, and whether the model changes over time.

A refusal is an answer: a score that cannot be explained cannot be defended to the driver, to a representative or to a tribunal.

Attach the response to the contract, along with notification before any change to the model.

Reproduce the workflow

For another way to make this requirement testable, consult workforce monitoring feature comparison. Reproduce the case with real roles, failures and recovery steps.

Independent reference

For a thematic point of reference, see TRL. Consult the source directly because technical, operational and legal details change.