Establishing the initial inventory: the count everything else is built on

01 / FIELD NOTE

Keep the decision tied to the operating context.

Every tracking system is a set of movements applied to a starting position. The system knows what has arrived, what has left and what has moved, and it derives what should be on hand from the figure it was given at the beginning. If that opening figure is wrong, the error does not sit still — it becomes the baseline against which every subsequent discrepancy is judged, and it is indistinguishable from a real one.

That is what makes the first count different in kind from the counts that follow. A later count corrects drift accumulated since the last one. The initial count has nothing to correct against, so its errors are invisible by construction: there is no prior state to reveal them, and the system will report confidently from a wrong starting point for as long as nobody checks.

The work is harder than a routine count for reasons that are practical rather than technical. The population being counted is usually the entire estate rather than a sample, it is being counted for the first time, the tags may not be applied yet, and the existing records are the very thing in doubt. That combination — full scope, no established process, unverified prior data — is what makes it fail rather than the counting technology.

The prior records deserve to be treated as evidence rather than as truth. A count that begins by loading the existing stock file and correcting it inherits every assumption in that file: items that were written off, receipts that were never made, transfers that were never posted. The count is the opportunity to establish what is actually present, which is a different exercise from validating a file.

Tags have to be applied before the count, and how that is done determines what the baseline can ever say. Applying tags at the granularity the business needs to manage means the first count establishes unit-level truth from the start; applying them coarser establishes a coarser truth that later reporting cannot refine. This is the last moment at which the level can be chosen cheaply.

Encoding errors belong in the baseline problem rather than beside it. A tag written with the wrong identity, or an identity associated with the wrong item, produces a record that is internally consistent and externally wrong. It will reconcile against itself perfectly and be discovered only when someone tries to pick the item or look it up. Verifying that a sample of encoded tags resolves to the right items is worth doing before the count rather than after.

Materials decide where the count will be hard, and it is worth knowing that in advance rather than discovering it. Metal shelving, metal products and liquid contents change how a tag responds, and a location where the radio performs poorly will produce an apparent shortage that is really a gap in coverage. Identifying those areas first means the count can plan a different method for them instead of recording a false result.

Phasing is how a large count gets finished. Counting by area, by building or by category reduces the peak effort, lets the process be corrected after the first area rather than after the whole estate, and keeps the operation running. The cost is that the phases are not simultaneous, so stock moves between them; the baseline is assembled from counts taken at different times, and the reconciliation has to account for that.

Reconciliation is the step where the work is either completed or quietly abandoned. Every line where the count and the prior record disagree has to be resolved to one of a small number of causes: the prior record was wrong, the item is genuinely missing, the item is present but elsewhere, or the count missed it. Each of those has a different correction, and the temptation to write them all to a single adjustment account is what makes the baseline worthless.

Freezing movement during the count is usually impossible and always disruptive, but ignoring movement is worse. Where goods must keep moving, the count needs a cut-off: movements after a stated moment are recorded against the old state and applied to the new one, so the two do not overlap. Without a cut-off, an item counted in one area and moved to another before it is counted again appears in both or neither.

The baseline is worth documenting as an artefact rather than as an event. What was counted, when, at what granularity, with what known gaps, and what was decided about each category of discrepancy. That record is what lets a discrepancy found a year later be traced back to a known limitation of the starting state instead of being treated as a new failure.

The reasonable expectation is not that the first count is exact. It is that its errors are known, bounded and recorded, so that later reconciliations can separate a genuine new problem from the residue of the starting state. A baseline with a documented margin is far more useful than one that was declared accurate without evidence — and considerably cheaper than a second attempt at the same count.