How do unit-of-measure errors corrupt spend data?

In two ways, and the difference matters to finance. Most unit errors leave the money right and the unit cost wrong: the invoice amount is what was paid, so spend is correct, but quantity or price per unit is mislabelled, and every cost-per-unit, volume and supplier comparison built on it is off. The rarer kind moves the recorded spend itself, when a quantity and a price are multiplied in the wrong units. Either kind survives the same way. It is large on its own line, 12 times, 10 times, 1,000 times, and diluted in a total, where it looks like normal variation. The ratios it leaves in implied prices can find it.

What the usual answer says

The usual answer is that unit errors make prices and quantities incomparable, fragment purchase volumes and block invoice matching, and that standardising units and fixing conversion settings solves it. That is the right destination. It skips which errors change the money, why the errors survive, and how to find the ones already in the data.

When the spend itself moves

From Quarri's own work with a lumber and millwork manufacturer: a units labelling error had inflated recorded purchase spend by about a fifth.

That is the kind that changes the figure a board sees. It happens where the system computes spend from quantity and price rather than taking the invoice amount, so a quantity recorded in the wrong unit is multiplied through.

Lumber makes unit errors easy because one item has several correct quantities. The 2026 retail method-of-sale rules for lumber allow softwood to be represented by "the number of pieces", by dimensions, and by "either the length of individual pieces or the lineal footage", and define the board foot as "the volume of a board 1 ft long, 1 ft wide, and 1 in thick". In trade practice, softwood board feet are figured from nominal sizes. A record can pair any of those quantities with the wrong label.

Why line-level errors survive

A price per thousand board feet read as a price per board foot is a thousandfold error on that line. A quantity of 2x6 pieces recorded as board feet, or the reverse, is off by the item's pieces-to-board-feet factor; for a 12-foot 2x6 that is 12. A pack of 10 recorded as 1 is off by 10. Each is obvious on its own line. Mixed into a month of correct lines for the same supplier or category, the effect on the total can be a few percent, within normal variation, and nobody looks at the line.

Finding them by their fingerprints

Unit errors leave ratios. For every purchase line, compute the implied price per stocking unit and divide it by that item's typical price per unit. Most lines sit near 1. Then check the lines near 1,000, 12, 10, or the item's own pieces-to-board-feet factor first: those ratios are what unit and pack errors produce, though some will be real price changes. Plot the ratios on a log scale and the clusters stand out. The same test works on quantities received against quantities ordered, and on inventory value against quantity on hand.

Don't let the model do the conversion

AI reads invoices well, and it can identify units on a line. It should not compute the conversions itself. A November 2025 benchmark, ORCA, from authors at a calculator company, tested five leading AI systems of late 2025 on 500 real-world calculation tasks, with the models doing the arithmetic themselves. They achieved between 45% and 63% accuracy, "with errors mainly related to rounding (35%) and calculation mistakes (33%)". Conversions belong in code, driven by each item's stored dimensions and stocking unit. The model's job is to read the unit correctly, and flag lines where it can't.

How to prevent them

Give every item one stocking unit, with its dimensions recorded, so every other unit is derived. Store purchase prices with an explicit price unit, separate from the quantity unit. Take spend from the invoice amount, not from quantity times price. Hold lines whose implied price per stocking unit falls outside a band around the item's recent average until someone confirms them, and run the ratio scan monthly on everything already posted.

When it doesn't apply

Businesses buying only in one unit, such as pulpwood by the ton, have little room for this error. Businesses whose suppliers send structured electronic invoices using the buyer's item codes and units avoid most of it.

Quarri for finance and strategy teams is built for the people who close the month, explain the margin and answer the board.

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.