ABC analysis for inventory: how to classify your SKUs, and what to do with the classes
ABC analysis ranks SKUs by how much money they move and splits them into three groups that get different amounts of attention. This post works one through on an 18-SKU catalog, shows how the choice of measure moves SKUs between classes, adds the XYZ layer for demand variability, and lists the cases where the classes mislead. The catalog and every number in it are invented so the arithmetic can be checked.
- Rank by a money measure, not by units. Annual consumption value (units × unit cost) ranks by inventory money flowing through. Gross margin contribution ranks by profit at risk. Revenue alone can promote high-price, thin-margin items.
- Sort, take a cumulative share, and cut. 80/95 (A to 80%, B to 95%) and 70/90 are conventions. Pick one and keep it stable.
- Each class is a set of choices you make: the review interval, the service level its safety stock targets, the count interval.
- XYZ adds how predictable each SKU is, using the coefficient of variation of monthly demand (standard deviation ÷ mean).
- Override the maths on purpose for new SKUs, cheap parts that a bestseller depends on, and seasonal SKUs measured on part of a year.
A catalog of a few hundred SKUs cannot all get equal attention. ABC analysis decides, once and in writing, which SKUs get the close watch. It is a policy view: it sets how much care each SKU gets over the next quarter. It does not tell you what to order today; for that, sort by days of supply against lead time, as in which products to reorder first.
Step 1 — Choose what you rank by
Ranking by units treats a $1 sleeve and a $120 grinder as the same problem. For a value-based ABC analysis, rank on a consistent money measure. There are three candidates:
- annual consumption value = annual units × unit cost. The inventory money moving through the SKU in a year; it fits questions of cash and carrying cost.
- gross margin contribution = annual units × (price − unit cost). The profit the SKU earns; it helps rank the profit potentially at risk from a stockout (the real cost also depends on substitution, backorders and lost customers).
- revenue = annual units × price. Easy to pull, but it mixes the two: a resold item with a high price and a thin margin ranks high on revenue while earning little.
Use the unit cost you use elsewhere (purchase cost, or landed cost if you track it). The example ranks by consumption value, then shows what changes under the other two.
Worked example (invented catalog): rank by consumption value
An invented kitchenware brand has 18 active SKUs with twelve months of sales. Multiply each SKU’s annual units by its unit cost, sort from largest to smallest, and keep a running total:
| Rank | SKU | Units / yr | Unit cost | Price | Consumption value | Share | Cumulative | Class |
|---|---|---|---|---|---|---|---|---|
| 1 | Cast-iron skillet 10in | 2,400 | $14.00 | $36.00 | $33,600 | 20.2% | 20.2% | A |
| 2 | Chef knife 8in | 1,500 | $18.00 | $49.00 | $27,000 | 16.2% | 36.5% | A |
| 3 | Espresso grinder (resold) | 180 | $120.00 | $149.00 | $21,600 | 13.0% | 49.5% | A |
| 4 | Stand mixer bowl | 300 | $62.00 | $95.00 | $18,600 | 11.2% | 60.7% | A |
| 5 | Cutting board, large | 2,100 | $7.50 | $28.00 | $15,750 | 9.5% | 70.1% | A |
| 6 | Dutch oven 5qt | 420 | $31.00 | $160.00 | $13,020 | 7.8% | 78.0% | A |
| 7 | Knife sharpener | 1,600 | $6.00 | $24.00 | $9,600 | 5.8% | 83.8% | B |
| 8 | Silicone spatula set | 3,000 | $2.40 | $14.00 | $7,200 | 4.3% | 88.1% | B |
| 9 | Linen apron | 900 | $6.50 | $29.00 | $5,850 | 3.5% | 91.6% | B |
| 10 | Seasoning oil 8oz | 1,800 | $1.60 | $7.50 | $2,880 | 1.7% | 93.3% | B |
| 11 | Skillet handle sleeve | 2,600 | $1.10 | $9.00 | $2,860 | 1.7% | 95.1% | C |
| 12 | Bench scraper | 700 | $2.80 | $13.00 | $1,960 | 1.2% | 96.2% | C |
| 13 | Knife roll bag | 160 | $11.00 | $51.00 | $1,760 | 1.1% | 97.3% | C |
| 14 | Oven mitt pair | 500 | $3.20 | $17.00 | $1,600 | 1.0% | 98.3% | C |
| 15 | Gift box sleeve (packaging) | 1,200 | $0.90 | not sold | $1,080 | 0.6% | 98.9% | C |
| 16 | Spice tin set | 240 | $4.50 | $15.00 | $1,080 | 0.6% | 99.6% | C |
| 17 | Mixer replacement whisk | 60 | $9.00 | $29.00 | $540 | 0.3% | 99.9% | C |
| 18 | Recipe card pack | 150 | $1.20 | $5.00 | $180 | 0.1% | 100.0% | C |
| Total | $166,160 |
Each share is the SKU’s value ÷ $166,160; the skillet’s is 33,600 ÷ 166,160 = 20.2%. The cumulative column adds the values down the list before dividing, so the chef knife row is (33,600 + 27,000) ÷ 166,160 = 36.5%. Shares and cumulative figures are rounded to one decimal, so the share column can differ from the cumulative steps by 0.1.
What changes if you rank by revenue or margin
Run the same sort and the same 80/95 cut on the other two measures (the invented prices are in the table above; the totals are $516,470 of revenue and $351,390 of gross margin on the 17 SKUs sold on their own. The gift box sleeve is packaging with no sale price, so it counts as zero on both and stays in C). Twelve SKUs keep their class. Six move:
| SKU | Consumption value | Revenue | Gross margin |
|---|---|---|---|
| Espresso grinder (resold) | $21,600 · A | $26,820 · B | $5,220 · C |
| Stand mixer bowl | $18,600 · A | $28,500 · A | $9,900 · B |
| Knife sharpener | $9,600 · B | $38,400 · A | $28,800 · A |
| Silicone spatula set | $7,200 · B | $42,000 · A | $34,800 · A |
| Skillet handle sleeve | $2,860 · C | $23,400 · B | $20,540 · A |
| Bench scraper | $1,960 · C | $9,100 · C | $7,140 · B |
The grinder is the clearest case: $120 cost, $149 price. It ties up the third-largest pool of inventory money, which is why it is A by consumption value, but only three of the SKUs sold on their own earn less margin, which is why it is C by margin. The handle sleeve runs the other way: $1.10 cost, $9 price, so it barely registers as inventory money but earns the seventh-largest margin.
Neither ranking is wrong; a cost measure watches where your cash sits, a margin measure watches where the profit at risk sits. One approach is to classify on one measure and flag any SKU that is two classes apart on the other, as the grinder is, for a manual decision.
The measure decides the classes. Write down which one you used, and why, before anyone argues about where a SKU landed.
Step 2 — Pick the cut-offs, and treat them as a choice
The example uses A up to 80% of cumulative value, B up to 95%, C the rest. A SKU takes the class of the band its cumulative figure falls in, so the Dutch oven at 78.0% is A and the sharpener at 83.8% is B. Under that rule:
- A: 6 SKUs (33% of the catalog), 78.0% of consumption value.
- B: 4 SKUs (22%), 15.4%.
- C: 8 SKUs (44%), 6.7%.
That is not “20% of items, 80% of value”, and it does not need to be. 80/95 (also written 80/15/5) and 70/90 (70/20/10) are conventions, not laws of inventory. Under a 70/90 split the same catalog gives 4 A SKUs, 4 B and 10 C: the cutting board and the Dutch oven drop to B, the apron and the seasoning oil drop to C. Nothing about those SKUs changed; the line moved.
Whichever split you pick, keep it stable, so a SKU changes class because its sales changed, not because the rule did, and look at the rows next to each line. The handle sleeve sits at 95.1%, 0.1 points into C, and $20 of consumption value separates it from the seasoning oil above it. A SKU that close to a line belongs in whichever class your judgement says, and the next sections give a reason to move this one.
Step 3 — Decide what each class changes
A class is only useful if it changes what you do. Here are three levers and the choices this invented brand made; yours depend on your team’s hours, your cash and what a stockout costs you.
Review frequency
The interval at which someone checks the SKU’s forecast, open orders and stock against plan. This brand chose weekly for A, every two weeks for B, monthly for C. The daily reorder list still catches any SKU about to run short; the class sets how frequently its forecast and settings get a second look.
Service level, and so safety stock
The service level is the share of replenishment cycles you aim to get through without running out. In the common safety stock formula it enters as a z-value, so it scales the buffer directly:
safety stock = z × standard deviation of monthly demand × √lead time in months
In plain words: the buffer grows with how much demand swings, with the square root of the wait, and with how sure you want to be. That form assumes demand that is roughly bell-shaped and a fixed lead time; the safety stock guide covers the versions for variable lead times. This brand targets 98% for A (z = 2.054), 95% for B (z = 1.645) and 90% for C (z = 1.282; the safety stock guide rounds these to 2.05, 1.65 and 1.28). For the skillet, with a monthly standard deviation of 12.42 units (worked out in Step 4) and an invented two-month lead time, √2 = 1.414, so 12.42 × 1.414 = 17.56 and:
- at 98%: 2.054 × 17.56 = 36.1 units
- at 95%: 1.645 × 17.56 = 28.9 units
- at 90%: 1.282 × 17.56 = 22.5 units
Moving a SKU from 95% to 98% raises its safety stock by 2.054 ÷ 1.645 = 1.249 times, about 25%. That is the trade the classes let you make on purpose: assign more buffer to the SKUs your policy treats as higher priority. A class built on consumption value says where the inventory money is, not where a stockout hurts, so if service level is meant to follow stockout cost, let margin or criticality move the class, or override it, as the next sections do.
Count frequency
The interval at which you cycle count the SKU. This brand counts A monthly, B quarterly and C twice a year: 6 × 12 + 4 × 4 + 8 × 2 = 104 counts a year, against 18 × 12 = 216 if it counted everything monthly. A cheap, easily miscounted item can justify more counts than its class suggests.
The C class is also where the question of whether to carry a SKU at all comes up. That is its own decision, covered in when to stop reordering a SKU.
Step 4 — Add the XYZ layer: how predictable is each SKU?
ABC says nothing about how steady demand is, which matters as much for safety stock. XYZ classification ranks SKUs by the coefficient of variation (CV) of their monthly demand:
CV = standard deviation of monthly units ÷ mean monthly units
In plain words: the typical swing as a share of the typical month.
Worked example (invented monthly units): the skillet, step by step
The skillet sold these units over twelve months: 190, 205, 180, 210, 195, 200, 185, 215, 190, 205, 200, 225.
- Mean: 2,400 ÷ 12 = 200.
- Deviations from the mean: −10, 5, −20, 10, −5, 0, −15, 15, −10, 5, 0, 25.
- Squared: 100, 25, 400, 100, 25, 0, 225, 225, 100, 25, 0, 625. Sum = 1,850.
- Variance: 1,850 ÷ 12 = 154.17. (This divides by 12, the population form. Dividing by 11, the sample form, gives a slightly larger figure; pick one and use it for every SKU.)
- Standard deviation: √154.17 = 12.42.
- CV: 12.42 ÷ 200 = 0.062.
The same steps for three more SKUs (monthly units are in the calculation script):
| SKU | ABC | Mean / month | Std dev | CV | XYZ |
|---|---|---|---|---|---|
| Cast-iron skillet | A | 200.00 | 12.42 | 0.06 | X |
| Skillet handle sleeve | C | 216.67 | 17.24 | 0.08 | X |
| Cutting board, large | A | 175.00 | 50.41 | 0.29 | Y |
| Espresso grinder (resold) | A | 15.00 | 16.83 | 1.12 | Z |
The cut-offs here are one convention, as arbitrary as 80/95: X below 0.25, Y from 0.25 to 0.75, Z above 0.75. The grinder sold zero units in five of the twelve months and 45 in its best, so its standard deviation is larger than its mean.
What the combined class suggests
- AX (the skillet): high value, steady. The buffer formula works well here, and tight, frequent ordering keeps cash low without much risk.
- AY (the cutting board): high value, moderate swings. Its CV of 0.29 comes mostly from a summer rise and a November and December peak, which are predictable; a seasonal forecast can explain swings that a flat CV calls noise.
- AZ (the grinder): high value, erratic. The bell-curve buffer formula fits lumpy demand with zero months poorly, and a 98% target on a $120 item ties up a lot of cash. Options include holding a small fixed quantity, ordering against confirmed demand, or accepting a longer delivery promise. That is a judgement for the owner of the SKU, not a formula output.
- CX (the sleeve): cheap and steady. A generous buffer costs little here.
Where this breaks
New SKUs with no history
A SKU launched six weeks ago has no annual figure to rank. One heuristic: give it a provisional class from its launch forecast, or from the class of the SKU it replaces, mark it provisional, and rank it on its own sales once it has a full quarter. The forecast itself is the harder part; see forecasting a new SKU with no sales history.
C by value, critical in practice
Three SKUs in the example show how value can understate importance:
- The handle sleeve ships with the skillet, the number one SKU. If the brand bundles it, a sleeve stockout can stop skillet orders, so the sleeve inherits the skillet’s risk.
- The gift box sleeve is packaging: it is never sold alone, so it has no revenue or margin at all, and ranks last on both. If gift orders need it, running out stops those orders.
- The replacement whisk sells 60 a year, but a customer who cannot get one may stop buying the bowls.
One way to handle this: classify components and accessories at the class of the item they enable, and keep a written override list with reasons rather than quietly bending the cut-offs.
Seasonal SKUs classified on part of a year
Classify on twelve months, or the result depends on which months you happened to include. If the brand had annualised the cutting board’s first six months (840 units, doubled to 1,680), its consumption value would read $12,600 instead of $15,750, 20% low, because the November and December peak is missing. In this catalog it would still be A, but its cumulative figure would rise from 70.1% to 77.6%, 2.4 points from the line. A SKU that sits nearer the cut-off can lose a class the same way, and get a lower service level going into its peak.
When to reclassify
One heuristic: re-run the analysis quarterly on a trailing twelve months, and move a SKU only when it has landed in the new class at two reviews in a row. That stops boundary SKUs like the sleeve from flipping every quarter and dragging their safety stock and count schedule with them. Keep the list of changes; a SKU that keeps sliding from A toward C is a signal worth reading next to its turnover.
What to do this week
- Pull twelve months of units for every active SKU, with unit cost and price.
- Choose your measure: consumption value for cash and policy, gross margin for stockout cost. Write the choice down.
- Sort, compute the cumulative share, and cut at 80/95 or 70/90. Note every SKU within a point or two of a line.
- Compute the CV of monthly demand for at least your A SKUs, and flag the Z ones for a manual stocking decision.
- Write the override list: new SKUs, components, kit parts, accessories to a top seller.
- Set review frequency, service level and count frequency per class, and put the next reclassification on the calendar three months out.
Questions people ask
How do you do an ABC analysis of inventory?
Multiply each SKU’s annual units by a money measure (unit cost for consumption value, or price minus cost for gross margin), sort from largest to smallest, and keep a running cumulative share of the total. Cut at the percentages you choose, such as A up to 80%, B up to 95% and C the rest.
Should ABC analysis use revenue or cost?
It depends on the question. Units times cost ranks by the inventory money a SKU moves, which suits cash and carrying-cost decisions. Units times margin ranks by profit at risk, which suits stockout decisions. Revenue alone can rank a high-price, thin-margin item above more profitable ones.
Is 80/15/5 the right split for ABC classes?
It is a convention, the same cut as 80/95. 70/90 (70/20/10) is another. The split changes which SKUs land in each class, so pick one, keep it stable between reviews, and look by hand at the SKUs that sit close to each cut-off.
What is ABC XYZ analysis?
ABC ranks SKUs by value; XYZ ranks them by how predictable their demand is, using the coefficient of variation of monthly units (standard deviation divided by mean). Combined, they separate steady high-value SKUs from erratic ones, which need different stocking decisions.
How do I calculate the coefficient of variation for demand?
Take twelve months of units, find the mean, subtract it from each month, square the differences, average them and take the square root to get the standard deviation. Divide that by the mean. Use the same form, dividing by 12 or by 11, for every SKU.
When should I redo ABC classification?
A quarterly re-run on the trailing twelve months is one reasonable cadence. One heuristic to stop SKUs near a cut-off from flipping: move a SKU only after it lands in the new class at two reviews in a row.
Is ABC analysis the same as deciding what to reorder?
No. ABC sets policy: how closely each SKU is reviewed, counted and buffered. What to order today comes from comparing each SKU’s days of supply with its lead time, whatever its class.
