What ABC analysis actually told us about 10,000 SKUs
ABC analysis is the first technique anyone learns for inventory, and it is easy to run badly for years without noticing. Here is what changed when we ran it on the right variable.
The version everyone runs first
Rank every SKU by annual revenue, take the top 80% of cumulative revenue and call it A, the next 15% B, the rest C. It takes ten minutes in a pivot table and it produces a chart that looks like insight. We ran it that way for a year.
The output was not wrong, exactly. It was just answering a question nobody in the purchasing meeting was asking. Revenue rank tells you which products move money through the business. It says nothing about which products make money, and nothing at all about which products are money, sitting still on a shelf.
Two changes
The first was to rank on contribution margin instead of revenue. This sounds obvious and it reordered the list substantially — a handful of high-volume items that everyone treated as core turned out to contribute very little after discounts and supplier terms were accounted for.
The second was to stop treating the classification as one-dimensional. A single letter per product compresses two independent facts into one, and the useful signal lives in their combination. We cross-tabulated the margin class against turnover speed, which gives nine cells instead of three.
A slow product with a strong margin and a slow product with a thin margin need opposite decisions. One letter cannot tell you which one you are holding.
What the tail looked like
About 8% of the catalogue produced a little over half of contribution margin. That part was unremarkable. The part that changed behaviour was the bottom-right cell: slow-turning, thin-margin items that had accumulated because no individual SKU was ever large enough to trigger a conversation.
Roughly a fifth of stock value sat in items that turned over less than twice a year. Each one, on its own, looked like a rounding error. In aggregate they were a serious amount of working capital, concentrated in three categories that had never once appeared on a management agenda.
The mirror image was more interesting. Slow-turning items with strong margins had been quietly deprioritised in purchasing for two years, because the ordering heuristic looked at last month's volume. Some of those were the products the sales team most wanted to have in stock.
What I would tell myself two years earlier
- Pick the variable that matches the decision. If the decision is what to stock, revenue is the wrong axis.
- One dimension hides the case you care about. Two dimensions cost nothing extra to compute.
- Look at the tail in aggregate. Individually invisible items are how stock accumulates.
- Re-run it quarterly. A classification from a year ago describes a catalogue that no longer exists.
None of this is sophisticated. It is the same technique from the textbook, pointed at a variable that matters. Most of the analysis work I have found useful has had that shape.