Spot check
Spot check explained: what a small unannounced sample can honestly tell you
A spot check is an unannounced inspection of a small number of items drawn from a larger population, carried out to test whether the routine process is working. It is triggered by suspicion or by a sampling plan, not by an event and not by a date the person being checked knows about.
The distinction from every other inspection is the unit of interest. A damage inspection is about one asset. A spot check is about a whole population, and the assets you actually look at are only a means of getting at it.
What is a spot check for?
Two purposes, and they need different designs. Confusing them is where most spot check programmes go wrong.
- Deterrence. Everyone knows a check can happen and nobody knows when, so the routine gets followed even when nobody is watching. This needs unpredictability. It does not need a large sample or a statistically clean one.
- Measurement. Estimating the real rate of a defect, a shortcut, or a compliance failure across the population. This needs proper random selection and enough units to say anything.
A check designed for deterrence should never be reported as a measurement. Three vans looked at because they happened to be in the yard on Thursday is a legitimate deterrent. It is not a fleet compliance rate, and the moment it appears in a board pack as one, it starts driving decisions it cannot support.
The sampling problem, stated honestly
A spot check tells you about the sample. It only tells you about the population if the sample was chosen properly, and in practice it rarely is.
Consider how units actually get picked. The vehicles parked nearest the office. The job files whose paperwork was easiest to pull. The sites on the way home from another meeting. The technician the supervisor already has doubts about. Each is a convenience or judgement sample, biased in a direction you cannot measure. The vehicles nearest the office may be the ones nobody uses, and therefore the cleanest. The technician you already doubt represents nobody but himself.
Bias is worse than a small sample. A small random sample gives you a wide range of uncertainty that you can at least state. A biased sample gives you a precise number pointing at the wrong population, and it feels like a measurement while it is doing it.
The fix is unglamorous. Write down the sampling frame, which is the list you are drawing from, and make sure it is the whole population rather than the reachable part of it. Draw the units with a random number before you know which ones come up. Record the sample size and the frame alongside the result. Where this matters commercially, formal acceptance sampling standards such as ISO 2859-1 already link a sample size and an acceptance number to a stated quality level, and there is no reason to invent your own scheme.
Spot check example: reading a published rate correctly
The Commercial Vehicle Safety Alliance runs International Roadcheck, a three day unannounced inspection blitz across Canada, Mexico, and the United States. In 2025 enforcement personnel conducted 56,178 inspections between 13 and 15 May, and CVSA reported a vehicle out of service rate of 18.1 per cent and a driver out of service rate of 5.9 per cent.
Now read it carefully. Inspectors chose which vehicles to pull in, using training and judgement about which ones looked worth stopping. So 18.1 per cent is the out of service rate among vehicles selected by experienced inspectors looking for reasons to select them. It is not the out of service rate of the North American truck fleet, and CVSA does not present it as one. The figure is useful for year on year movement and for deterrence. Quoting it as the state of the industry overstates the problem.
When a spot check is the right tool
- Verifying that a routine check is being done at the inspection frequency the schedule claims, rather than signed for.
- Testing contractor work quality between formal milestones, alongside proof of work evidence.
- Following up a specific signal: a pattern in the data, a complaint, a run of failures from one source.
- Sampling submitted paperwork against the physical reality it describes.
How it differs from the scheduled checks
- Versus the annual inspection. The annual is known in advance and can be prepared for. That predictability is exactly what the spot check exists to defeat.
- Versus a quality assurance inspection. QA inspection is built into the process at defined points. A spot check sits outside the process and audits it.
- Versus an audit. An audit examines the system and its records end to end. A spot check tests one narrow claim the system makes, quickly, and leans on the audit trail for the rest.
Two practical cautions. A finding in a spot check is a reason to look harder, never a verdict on the population, and one fault in three units checked is not a 33 per cent failure rate. Second, the value drops to nothing if the checks become predictable, which they do the moment they settle into the same week each quarter or the same three depots. Where sites are spread out, the cost of getting a person there is usually what makes checks predictable in the first place. Venta Capture, a product of VentaVid, is used for that pattern: an unannounced request goes out by link, whoever is on site captures the specific views asked for through a guided flow on their own phone, and the submission comes back as a structured case. It decides nothing, and it does not make the sampling sound. Selecting the sites properly is still your job. What it removes is the travel cost that quietly shrank the sample to whatever was nearby. What to do when a check turns up a real problem sits with risk assessment.
