Reading a Vendor Claim

Marketing in this field leans on safety outcomes, which are the hardest claims to verify and the most emotive to dispute.

"Reduces collisions by N percent"

Almost always a before-and-after comparison at a customer fleet, with no control group.

Which cannot distinguish the programme from regression to the mean, a fleet renewal, a route change or a quiet year.

Ask: was there a comparison group, over what period, and what else changed at that fleet?

A vendor who has a controlled study will say so immediately.

"AI detects distraction and fatigue"

Ask for the false positive rate, on vehicles like yours, with drivers wearing glasses, masks and high-visibility headwear.

And ask what happens at night and in low sun.

Request a trial where you count the errors yourself, because this is the claim that diverges most from the demonstration.

"Risk score validated against collision data"

Ask whose collision data, in which operation type.

A model validated on long-haul trucking does not transfer to urban multi-drop, and the weights encode the original context.

And ask whether it is recalibrated on your data or applied as delivered.

"Fully compliant with privacy regulation"

Compliance is a property of your deployment and your justification.

Ask instead: can continuous recording be disabled, can live viewing be turned off at account level, is viewing logged, can retention be set per camera direction?

Those four answers tell you whether a defensible configuration is even available.

"Improves driver engagement"

Usually gamification.

Ask what happened to near-miss reporting at reference customers, which nobody will have measured and which is the question worth planting.

"Proprietary risk model"

The single most important red flag.

A score that cannot be explained cannot be defended to the driver, to a representative or to a tribunal.

Ask for the formula in writing. A refusal is an answer.

The general test

Any safety claim without a comparison group, a named operation type or a number you can check is a statement of intent.

Put the questions in writing and attach the answers to the contract, because written answers are more careful and the carefulness is itself informative.

Ask about the control group

The one question that separates evidence from marketing.

Was there a comparison group, over what period, and what else changed at that fleet?

Without one, a before-and-after figure cannot distinguish the programme from regression to the mean, a fleet renewal or a quiet year.

A vendor with a controlled study says so in the first minute.

Turn the boundary into a test

For a concrete product reference, review this workflow page can help turn this boundary into a test. Verify current behaviour with representative roles and record what reaches the destination.

Independent reference

For a thematic point of reference, see the European Agency for Safety and Health at Work. Use the current material as an outside check before turning the principle into a system rule.