Thin stock. Sold-out SKUs. Products that move once a quarter. Most replenishment systems handle these with manual overrides and reference SKUs. Stockmind™ measures them.
Replenishment software is largely built around a comfortable assumption: products are stocked deep, sell steadily, and are rarely unavailable. That describes some businesses well. It does not describe a catalogue of device accessories, spare parts, or fashion variants — where the majority of SKUs are stocked one or two units deep and tied to product generations that turn over fast.
Here's the actual distribution from one live catalogue of 226,000 SKUs:
For a business like this, thin and intermittent stock isn't an edge case to be handled separately. It is the catalogue. A tool that treats it as noise is treating the business as noise.
Thin stock is also where Phantom Slow Movers™ breed: a product holding one unit that sells on day three spends the next eighty-seven days unable to sell anything, and then gets judged on the silence.
Nearly every replenishment engine derives demand the same way: units sold divided by days elapsed. That single formula carries a hidden assumption — that the product was on the shelf the whole time.
For deep-stocked fast movers, the assumption mostly holds. For thin stock it collapses. A product holding one unit that sells on day three spends the remaining eighty-seven days unable to sell anything, and the arithmetic dutifully reports it as a slow mover. The tool then declines to reorder it — so it stays empty, sells nothing more, and the next cycle confirms the verdict.
The products this misreads most severely are the ones that sell out fastest. The maths has a built-in bias against your winners.
Statisticians have a name for this: censored demand. When a product is unavailable you don't observe zero demand — you observe nothing, and those are different things. The literature goes back decades. What's striking is how rarely replenishment software does anything about it.
It isn't solved by reaching for a more sophisticated forecaster either. Croston's method, the standard tool for sporadic demand, works by measuring the gaps between sales — so a forty-day stockout reads as a forty-day gap in demand, and the product gets reclassified as slow-moving precisely because it sold out. A Poisson model fitted to the same data estimates a rate from observations that include the empty days. Neither is malfunctioning; both are being handed data that misrepresents what happened, and neither has any way to tell.
The industry has answers for this. It's worth looking closely at what they actually are.
The buyer inspects a suggestion, recognises it's wrong, and types in the right number. This works — it's also the buyer performing, by hand, the calculation the software was purchased to perform. Every override is a small admission that the model couldn't read the product.
When a product's own history is unusable, borrow a similar product's. But a strap for a newly launched watch and one for the model it replaced are not interchangeable demand signals — one is ramping, the other decaying. Reference SKUs import a demand shape from a product living a different life.
Set a floor per product and top up to it. Simple and predictable, but it isn't demand-driven at all — it's a static rule, maintained by hand, that drifts out of date the moment demand shifts.
None of these are unreasonable. They're sensible responses to a real limitation. But notice what they have in common: each one asks a human to supply the judgement the system couldn't derive — and each one has to be maintained forever.
Overrides scale fine across five thousand SKUs. The arithmetic changes at two hundred thousand.
On the catalogue above, measuring actual availability revealed that 82% of selling products had understated velocity — the fastest movers by as much as six times. Correcting that by hand would mean roughly forty thousand overrides, revisited every buying cycle as stock positions change. That isn't a tedious process; it's an impossible one. Which means, in practice, that most of the tail never gets corrected at all — the errors simply go unnoticed, one lost sale at a time.
A tool that needs manual correction on 82% of your catalogue isn't giving you replenishment intelligence. It's giving you homework.
Rather than asking you to compensate for what it can't measure, Stockmind measures it. It samples your warehouse's historical stock balances across the period — daily by default, placing availability within about half a day — and establishes, for every product, how many days it was genuinely available to sell, then calculates velocity against that.
No reference product. No override. No maintained rule. The product's own history, read correctly.
| The problem | Conventional answer | Stockmind |
|---|---|---|
| Product sold out mid-period | Manual override, or accept the understated figure | Measured: velocity reflects its ShelfDays™ — the days it was actually on the shelf |
| Single-unit stock that sells immediately | Minimum stock rule, maintained by hand | Treated as the fast mover it is, with a floor so one sale can't cause an over-order |
| Newly launched product | Reference SKU borrowed from a predecessor | Measured from launch — absent from earlier snapshots, so never penalised for a short history |
| Demand rising or fading | Flat average treats both identically | Recent rate weighted against the full period, with confidence scaled to volume |
| Erratic vs. predictable demand | One safety-stock figure for everything | Buffers scale to each product's own variability — and a stockout isn't mistaken for volatility |
Every under-ordered fast mover is a stockout you chose without realising it. The sale doesn't get delayed — it goes to someone else, and the demand signal disappears along with it, quietly confirming the tool's mistaken verdict.
This is a measurable claim, not a marketing one. Connect your warehouse system and Stockmind will tell you what proportion of your selling products currently have understated velocity, and which specific SKUs are worst affected.
It takes about half an hour to run. Whatever number comes back is the size of the problem you're currently solving by hand — or not solving at all.
From there, a weekly briefing ranks what to do about it, and an optional AI summary reads it back as prose — including which ranges are worth widening with new products. The figures it quotes are checked against the source data, and the replenishment maths itself never involves a language model.
We'll measure your catalogue and show you exactly which products are being under-ordered, and by how much.
Request an availability audit →