Inventory turnover from continuous data

01 / FIELD NOTE

Keep the decision tied to the operating context.

Turnover is calculated as cost of goods sold divided by average inventory over a period. The formula is simple enough that the difficulty is easy to miss: both terms are supposed to describe what happened across the period, and both are usually assembled from a small number of measurements taken at points inside it.

Average inventory is the weaker of the two. Where the record is maintained by periodic counting, the average is computed from the values observed at those counts, which means the periods when stock was highest and lowest are represented only if a count happened to fall on them. A seasonal business that counts in a quiet month reports a healthier average than it holds, and the error is systematic rather than random.

Cost of goods sold carries its own distortion. It comes from the financial system, where it is derived from purchases and adjustments rather than from the movement of individual units. Shrinkage is the clearest case: stock that left without a sale is not in cost of goods sold, so it inflates the average inventory in the denominator instead of appearing anywhere as a loss, and the ratio reads worse than the trading performance warrants.

Continuous capture changes the character of the numerator rather than its definition. When every movement is recorded as it happens, average inventory can be computed from the actual holding across the period rather than interpolated between counts, and the two extreme weeks stop being invisible. That is a change in measurement quality, and it is the reason the ratio becomes usable as a decision input rather than as a period-end score.

It also changes what can be said about the relationship between turnover and the individual items. A single ratio for a business is the sum of a fast-moving majority and a slow-moving minority, and the two need opposite responses. Knowing the ratio has not improved does not tell you whether to buy less or to clear more; knowing which categories moved and which stopped requires a record that holds the movement history of each.

Shrinkage becomes visible as a separate quantity rather than hidden inside an average. If the system knows which units were received and which were sold, then units that left neither way are identifiable as a difference rather than inferred from a count variance. That does not recover the goods, but it separates a loss from a forecasting error, which are managed by different people.

The timing of the feedback is what makes the difference operational. A ratio computed at period end describes a position that has already been acted on — the purchases were made, the stock is held, the markdown decision is due. A measure computed continuously can be read while the position is still forming, which is the window in which changing a purchase order is cheaper than clearing the result.

Accuracy of the underlying record sets the ceiling on all of this. A continuous data stream built on an inaccurate count is an inaccurate stream delivered faster, and a low count confidence produces the opposite error from the one expected: businesses buy more to cover uncertainty, so an unreliable record tends to raise stock rather than lower it. Improving the measure’s frequency without improving its accuracy moves the decision point earlier without making it better.

The measures worth reading alongside turnover are the ones that explain it rather than restate it. The age profile of what is on hand, the share of stock present but unavailable, the proportion recorded in the wrong location, and the time between a movement and its posting. Each of those is a leading indicator, and unlike a ratio none of them can improve without the operation changing.

There is a limit to what the record can establish. Cost of goods sold depends on what the business paid, which is a financial fact rather than a physical one, and no amount of counting determines it. Average inventory depends on what was held, which counting can establish. A turnover programme is therefore a collaboration between the movement record and the accounts, and the useful question is which of the two is currently the source of the error.

The practical sequence is to improve the inventory term first, because it is the one the operation controls. Making the holding figure continuous and trustworthy turns the ratio into a measure that responds to decisions, and only then is it worth reading trends from it or setting targets against it. A target set on an estimate produces pressure to improve the estimate rather than the operation.

What this does not do is reduce stock by itself. Turnover improves when purchasing, assortment, lead times and record quality move together, and a better record moves one of those. Its contribution is that it makes the position legible early and makes the effect of a decision visible afterwards — which is the difference between managing the ratio and reporting it.