The plan is always a full count once a year, and the outcome is always that it happens twice and then stops, at which point nobody knows how accurate the register is.
Key takeaways
A full annual count is planned every year and rarely completed.
Twenty assets a month gets done and covers a substantial register annually.
Weight the sample: high value, high mobility and never-verified first.
Publish the accuracy figure; it is the only measure of whether this works.
Not found is a state with a count, not an immediate write-off.
Why the full count fails
Fig 1. Three approaches to verification over a year. The full count starts well and stops; sampling continues and covers more.
The full count fails for the obvious reason: it is a day’s work for several people, it is nobody’s priority, and it is scheduled for a month that turns out to be busy. Two years running it slips and then it stops being scheduled.
Twenty assets a month is twenty minutes with a phone and it survives being busy, which is the property that matters.
Choosing the sample
Fig 2. How the monthly sample is chosen. Deliberate weighting finds more errors than a random sample of the same size.
Compute
Management
Analytics
Front-end & mobile
What the check involves
Go to the recorded location, find the label, scan it, and answer one question: is it here and does it look like what the record says? If yes, tap and move on. If it is somewhere else, scan it wherever it is and the location updates. If it cannot be found, record that.
The whole interaction is a scan and a tap, and if it involves anything more it will stop happening within two months.
The accuracy figure
Fig 3. Three quarters of sampling results. The trend is what matters, and the register’s credibility depends on this figure being published rather than assumed.
Publishing it does two things. It tells anybody relying on the register how much to rely on it, and it turns register maintenance from an act of faith into something with a measurable outcome that improves.
The improvement in that chart came from the label printing moving into purchasing and the replacement-purchase disposal question, both of which are small changes whose effect is only visible because the accuracy was being measured.
Not found is not gone
An asset that cannot be found on one attempt is usually somewhere else, being used by somebody, or out for repair. Marking it not found and re-sampling it the following month resolves most of them.
After a stated number of attempts — three is reasonable — it moves to presumed disposed, which requires a person to confirm and which produces the accounting event. That is a slow enough process to avoid writing off things that were merely on somebody’s desk.