Resources · Planning
ABC and XYZ classification: where your attention and your stock should go
ABC ranks your parts by how much money they represent. XYZ ranks them by how unpredictable their demand is. Put the two together and you get nine boxes, each with a sensible way to buy, so you stop giving every part the same amount of care.
The problem it solves
A buying team has a limited number of hours a week and a limited amount of cash. If you review ten thousand parts with the same effort, the few that matter get too little and the many that do not get far too much. Classification is just a way to decide where the effort goes.
ABC: how much is it worth?
Rank the parts by annual spend (or by volume, if you do not have costs), highest first. Add up the running share of the total. The parts that make up the first 80% are A, the next 15% are B and the rest are C. Here is a small example, with annual spend in thousands.
| Part | Spend | Running share | Class |
|---|---|---|---|
| P1 | 400 | 40% | A |
| P2 | 250 | 65% | A |
| P3 | 120 | 77% | A |
| P4 | 60 | 83% | B |
| P5 | 50 | 88% | B |
| P6 | 40 | 92% | B |
| P7 | 30 | 95% | B |
| P8 | 25 | 97.5% | C |
| P9 | 15 | 99% | C |
| P10 | 10 | 100% | C |
Three parts out of ten carry 77% of the spend. That lopsidedness is normal; it is the reason the method works. The cut-offs are conventions, not laws. Some teams put the part that crosses 80% into A, others do not. Pick a rule, write it down and apply it the same way each time.
XYZ: how predictable is it?
For each part, take twelve months of demand and work out the coefficient of variation: the standard deviation divided by the average. It is a scale-free way of asking how bumpy the demand is.
Common cut-offs are X below 0.5, Y from 0.5 to 1.0 and Z above 1.0.
| Part | Twelve months of demand | Average | CV | Class |
|---|---|---|---|---|
| Steady | 95, 105, 100, 98, 102, 100, 97, 103, 100, 99, 101, 100 | 100 | 0.03 | X |
| Lumpy | 0, 0, 300, 0, 0, 0, 200, 0, 0, 0, 100, 0 | 50 | 1.91 | Z |
Both parts have a sensible average. Only one of them can be planned from it.
The nine boxes, and what to do with each
| X (steady) | Y (variable) | Z (erratic) | |
|---|---|---|---|
| A (high value) | Tight cover, frequent small orders, close supplier relationship | Review often, keep a measured buffer | Buy against orders or firm forecasts. Do not hold large stock of something expensive that may not move |
| B | Standard reorder rules, regular review | Reorder point with a buffer you have calculated | Reorder point, watch for excess |
| C (low value) | Order in bulk, less often; accept some extra stock | Simple min-max rules | Simple rules; check once a quarter for dead stock |
This is a starting policy, not a rule. A cheap part that stops your line is not really a C, and a flag for such parts (critical, hard to source) is often added on top. The buffer for Y and Z parts should come from a calculation like the one in how to set safety stock, with the caveat there that the method assumes a reasonably bell-shaped demand, which Z parts often do not have.
Finding dead stock
The most practical use of this is on the C parts. Add a third question: has this part moved at all in the last twelve months? A part with no consumption and stock on the shelf is not a C, it is cash tied up for no reason. These are rarely noticed individually because each is small. Added together, they are often worth looking at. Flag them, decide whether there is any real chance of use, and either return, sell or write them off. Then stop reordering them.
Reclassify regularly. A part that was an X last year can be a Z this year if a customer left. A classification that nobody refreshes is a polite way of keeping last year’s assumptions.
What people get wrong
- Classifying once and never again.
- Using spend alone when risk matters: a low-value part with one supplier and a long lead time deserves attention that its value does not earn.
- Calculating CV on too little history. Three months is not twelve.
- Mixing seasonal demand into “variable.” Seasonality is predictable, so it is a different problem from randomness.
- Treating the boxes as a replacement for judgment. They are a way of deciding where to start looking.
In the software
Procurement Control Tower ranks parts by A, B and C on either volume or spend, using the 80% and 95% cut-offs described above, and flags parts with healthy stock but more than nine months of cover as excess. It does not calculate the XYZ variability class, so that half of the matrix is a manual step for now. See what it covers.