Knowing Whether the Data Is Any Good

Location systems report confidently regardless of how well they are working. These four checks take an afternoon and are worth repeating quarterly.

Coverage

Proportion of working time with a valid fix, per vehicle.

Segment by area, which finds the dead zones.

A vehicle consistently below the fleet average usually has a hardware fault rather than a driver problem, and checking that first avoids an unfortunate conversation.

Track the trend. A fleet-wide drop usually follows a phone operating system update or a vendor change.

Geofence reliability

False entries: events at sites the vehicle passed but did not visit.

Missed entries: visits with no event, which the job record will reveal.

Flapping: repeated enter and exit within minutes, which means the fence is smaller than the position error.

Sample twenty visits against job records and count each category. Above a few percent, the fences need resizing.

Timestamp integrity

Difference between fix time and receipt time.

Large differences mean poor coverage and that any report ordering events by receipt is wrong.

Check that the system stores both, and that reports use the fix time.

This is the check nobody runs and it silently corrupts arrival-time analysis in areas with weak signal.

Attribution

Does the vehicle record match who was driving?

Vehicle swaps, shared vans, a driver taking a colleague's vehicle — all produce data attributed to the wrong person.

Compare against the rota for a sample week.

Where mismatches are common, individual-level use of the data is unsupportable, whatever the policy says, and that is worth knowing before anyone relies on it.

Reporting the four

On the same page as any analysis, so a reader knows what weight to apply.

State the accuracy assumption wherever a figure depends on position.

Name the known gaps.

A report that states its own limits is trusted; one that does not is tested once and then discounted.

Acting on them

Coverage low: check hardware first, then the dead zone map.

Geofences unreliable: resize, starting with the worst.

Timestamps diverging: fix the reporting to use fix time, then investigate coverage.

Attribution poor: stop using the data at individual level until it is fixed, and say so rather than quietly continuing.

Check attribution before anything individual

The check that decides whether the data can be used per person at all.

Compare vehicle records against the rota for a sample week.

Vehicle swaps, shared vans, someone taking a colleague's van all attribute data to the wrong person.

Where mismatches are common, individual-level use is unsupportable, whatever the policy permits.

Say so plainly rather than quietly continuing, because the alternative is a disciplinary meeting built on someone else's journeys.

Connect policy to configuration

The practical choices behind this note can be compared with online timesheets. Keep the written purpose in control of the setup, enable only the data needed and review the result with affected workers.

Independent reference

For an external point of reference, see the Office for National Statistics. Its statistical standards and publications offer a useful reminder to compare definitions before interpreting operational numbers.