Measurement

Inventory turnover: which number belongs on top?

Most write-ups of inventory turnover spend their time on the denominator — which average, over which window. The denominator matters. The numerator can break the ratio even more dramatically. On one SKU, one year, one ledger, the same stock turns 4.0 times, 3.0 times or 10.0 times depending on two choices nobody writes down. The numbers are an invented worked example, built so the arithmetic can be checked end to end.

Sell-through asks whether a delivery went. Turnover asks something colder and more useful to a bank balance: how many times over did you replace the money sitting on that shelf? One turn a year means every dollar of stock came back once. Ten turns means it came back ten times and did ten jobs instead of one.

Which is why it matters that the ratio is quoted more casually than almost any other number in the business. Two people can pull turnover for the same SKU, from the same ledger, for the same twelve months, and hand you 3.0 and 10.0. Neither has made an arithmetic mistake.

The formula, and the word doing all the work

The formula is not in dispute:

The word doing all the work is cost, and it is doing it in both halves. Cost of goods sold is what the units cost you. Average inventory, in that formula, is also what the units cost you. The ratio is a comparison of one pool of money against another pool of the same kind of money, which is the only reason the answer means anything.

Break that and the ratio still computes. It just stops describing anything.

Step 1 — Put the same basis on both halves

One SKU. One calendar year. It cost $12.00 landed and sold for $30.00, and 4,800 units went out the door.

Cost over cost:

Now the substitution that gets made constantly, usually because the two numbers come out of two different systems and both look like “the sales number”:

Same SKU, same year, same shelf. The second figure is not 6 turns better. It is the first figure multiplied by the relationship between your selling price and your cost — which is to say, by your margin, which has nothing to do with how fast anything moved.

⭐ The inflation is exactly predictable, and that is how you catch it

Putting revenue over inventory-at-cost multiplies the true ratio by:

At a 60% gross margin that is 1 ÷ 0.40 = 2.5. And 4.0 × 2.5 = 10.0, which is the wrong answer we just produced. The factor does not depend on how you averaged the inventory: further down we compute the same SKU a second way and get 3.0 turns, and the mixed-basis version of that is 7.5 — again exactly 2.5×.

This gives you a check you can run in your head. If your reported turnover is suspiciously close to your true turnover multiplied by your margin multiple, someone has put revenue over cost. The higher your margin, the more flattering the error: at 75% margin it is 4×.

⚠ One honest caveat on that identity: it is exact here because every unit in this example carries the same $12.00 cost, so cost of goods sold is precisely units × cost. In a real ledger with FIFO layers, landed costs that moved mid-year, or a margin that varies by channel, the relationship holds as a close approximation rather than an equality. It is reliable enough to catch the error and not precise enough to back out the true figure from the wrong one.

⛔ But sales on top is not automatically wrong

It would be tidy to end the section with “never put sales on top,” and it would be false.

Retail-method businesses — a lot of traditional retail, and some tools built for it — carry inventory valued at retail rather than at cost. In that world, dividing sales at retail by average inventory at retail is cost-basis-consistent in exactly the way the standard formula is. Both halves are on the same basis. The answer is comparable to itself over time, which is most of what you want from a ratio.

So the rule is not about which number goes on top. The rule is that the two halves have to be the same kind of money. The reason the sales-over-cost version is the common break is not that people prefer revenue; it is that the revenue figure and the inventory figure are usually pulled from two different places, and neither one announces its basis.

⚠ And “the same kind of money” has a third reading worth naming, because a cost-or-retail binary quietly excludes it: units sold over average units on hand is also internally consistent, and on our SKU it gives 4,800 ÷ 1,200 = 4.0 — the same answer as the cost version, because one unit carries one cost here. It is common in apparel and anywhere a category is priced uniformly. What it gives up is money: it cannot be aggregated across SKUs of different value, and it will not notice that your landed cost rose 20% this year. Useful for a buyer looking at one category, weaker for anyone asking where the cash went.

Ask of any turnover figure you are handed: is the inventory in this denominator valued at what we paid or at what we charge? If the answer is “I’d have to check,” the ratio is not yet a number.

Step 2 — Average of what, over what

This is the argument most articles have, and unlike the one above it does not have a single right answer. It has a defensible answer and a lazy one, and the gap between them is where intra-period movement lives — seasonality being a common reason it matters.

The textbook average is opening plus closing, over two. Our SKU opened the year at 2,400 units and closed it at 800:

But the stock did not travel in a straight line between those two points. Here are the twelve month-end counts:

Those twelve add to 14,400 and average to 1,200 units — $14,400 at cost, which is the denominator we used at the start:

Three turns or four turns, same year. The two-point average happened to catch this SKU on January’s restock and December’s run-down, and read a third more stock than the business actually carried. It is not that the two-point method is wrong; it is that it samples twice, and stock that moves within the period — a seasonal SKU is a common case — is exactly the thing two samples misrepresent.

⚠ Neither figure here is the “true” one in any deep sense — twelve month-end snapshots are themselves twelve samples, and a SKU that spikes and drains mid-month would fool them too. The honest statement is this: month-end balances are a common practical improvement over two endpoints; more frequent snapshots can be more faithful where inventory moves sharply within the month. Whichever you pick has to stay picked. A turnover series that silently changes its averaging method halfway through is worse than either method consistently applied.

What that does to days of inventory

Days of inventory is turnover restated: for an annualized turnover rate, 365 divided by the turns (a shorter period uses the days in that period). It is not a second opinion, and it inherits every choice made above.

Thirty-six days or a hundred and twenty-two. That is the same shelf, the same year and the same 4,800 units, and the range is wide enough to justify opposite decisions about whether this SKU is tying up too much cash.

Ratios that answer a different question

Three measures sit near turnover and are regularly swapped for it. Two are genuinely different questions. One is the same fact wearing different units, and it is worth being clear about which is which.

Days of inventory is not a different measure. It is 365 ÷ annual turns, as above. If someone offers it as a cross-check on turnover, there is nothing to cross-check: given the same turns figure it cannot disagree. (If two reports do disagree, the usual culprit is a 360-day convention on one of them, or two different windows — which is a discrepancy in the inputs, not a second opinion on the stock.)

GMROI — gross margin dollars divided by average inventory at cost — asks how much margin each dollar of stock earned, which turnover genuinely cannot answer. Our SKU produced $86,400 of gross margin on $14,400 of average stock:

Six dollars of margin per dollar tied up. That is the number that lets a slow high-margin SKU defend itself against a fast thin one, and turnover on its own will always make the fast one look better.

Weeks of cover — units on hand divided by forecast weekly demand — is the forward-looking sibling. Turnover is entirely historical: it can only tell you what already happened to stock you already bought. Cover is the one that answers “do I need to order,” and it depends on a forecast, which turnover never touches.

Sell-through is a fourth relative and gets its own treatment, because its argument runs the other way — the fight there is about the denominator, not the numerator.

Step 3 — Read it per SKU, because the aggregate hides the money

Business-level turnover is total cost of goods sold over total average inventory, which works out to an inventory-value-weighted average of the SKU turns — and a weighted average conceals exactly the thing you are looking for. Three SKUs:

The business turns 3.08 times, which is unremarkable and would not start a conversation. Inside it, SKU B turns once every two years.

Now look at where the cash actually is. B holds $18,000 of the $37,200 of average stock — 48.4% of the money — and generates $9,000 of $114,600 of cost of sales, which is 7.9% of COGS. Nearly half the working capital is parked in the SKU doing the least work, and the combined figure is arithmetically incapable of showing you that.

⭐ The aggregate is not a lie. It is a correctly computed number answering a question nobody usefully asks. Cash is not tied up by a business; it is tied up by line items, and turnover only becomes actionable at the level where you can actually place or cancel an order.

A turn is a speed, not a score

There is no turnover figure you should be hitting, and the benchmarks circulating for your category are mostly comparing businesses with different margins, different lead times and different averaging methods.

High turns are not free. Pursuing higher turns by ordering smaller lots more frequently can increase PO, freight or stockout costs. A SKU turning 1.5 times with a 70% margin and a twelve-week lead time may be entirely rational. Turnover measures speed; whether that speed is right depends on what it costs you to go faster and what you lose by being empty.

What turnover is genuinely good at is change. The same SKU, measured the same way, on a falling trend across four comparable quarters — that is a real signal, and it does not require you to know what good looks like for anyone else.

What the number actually decides

• How much cash you can free without losing sales. Rank SKUs by average inventory at cost, not by turns. The slow ones at the top of that list are the money.
• Order cadence. A rising turn means COGS increased relative to average inventory; a falling turn means the opposite. Then inspect receipts, demand and average inventory to find out whether order size or cadence caused the change.
• When to stop defending a SKU. Low turns plus a low GMROI is a different conversation from low turns plus a high one. The first is a candidate to investigate for exit; the second may just be a SKU you are financing on purpose.
• What it should not decide: Turnover should not make a purchasing decision on its own. Promotions, stockouts and bulk receipts are examples of why — each can distort one half of the ratio without touching the other — not the only periods when judgment is required.

What to do this week

1. Find out what basis your inventory value is on. Cost or retail. One question, and until it is answered no turnover figure in your reporting means anything.
2. Check the numerator against it. If inventory is at cost, the top of the ratio is cost of goods sold, not revenue. If your reported turns are roughly your real turns times your margin multiple, you have found the problem.
3. Pick an averaging method and write it down. Month-end average over the window if you can get it; opening-and-closing if you cannot. The choice matters less than never changing it silently.
4. Recompute your ten largest SKUs by inventory value, individually. Not the business. The list you want is sorted by cash tied up, and it will not look like the list sorted by turns.
5. Put GMROI next to it for anything slow before deciding a low-turn SKU is a problem.

Questions people ask