Why the textbook safety stock formula quietly fails your best sellers
Open any planning textbook and you'll find the same safety stock formula: Z × σd × √L — a service factor, times the variation in daily demand, times the square root of the lead time. It's in every ERP, every planning tool, every consultant's spreadsheet. And it has one assumption baked so deep that most people who use it daily have never noticed it's there.
The hidden assumption
The formula assumes your demand follows a normal distribution — the smooth, symmetrical bell curve from statistics class. Under that assumption, the maths is elegant: pick a service level (say 95%), look up the matching Z score (1.64), multiply through, and out comes a buffer that will protect you against exactly 95% of demand outcomes.
The problem: real wholesale demand almost never looks like a bell curve.
Watch the actual sales of a real product for a year. It sells 30 units a day for three weeks, then a national account places a pallet order and one day shows 900. It goes quiet for a fortnight in August. It doubles ahead of the season and halves after it. Plot that on a chart and you don't get a bell — you get a spiky, lopsided mess with a long tail of extreme days.
The bell curve says a day of demand ten times your average is essentially impossible. Your order history says it happens every few months — and it's called your biggest customer.
Who gets hurt: exactly the wrong products
Here's the cruel part. The formula's error isn't random — it's systematic, and it runs in the worst possible direction:
- Your spiky best sellers get under-protected. Products with lumpy, big-order-driven demand have far more extreme days than a bell curve allows. The formula sizes their buffer for a politeness that doesn't exist — so the buffer runs out precisely when the big order lands. These are usually your highest-revenue lines, so every miss is expensive.
- Your boring, stable products get over-protected. Steady sellers with genuinely smooth demand get buffers they'll never use. That's cash sitting on a shelf doing nothing — often for years.
Add it up across a catalogue of thousands of products and you get the classic mid-market symptom: too much stock overall, and still stocking out of the products that matter. Most businesses respond by nudging the service level up across the board — which mostly buys more protection for the products that didn't need it.
The second failure: stockout-contaminated history
There's a subtler trap too. The σ in the formula is measured from your sales history. But if you were out of stock for three weeks last year, those weeks show zero sales — not because demand was zero, but because you had nothing to sell. Zeros drag the measured variability down, which shrinks the buffer, which causes the next stockout, which adds more zeros. The formula quietly learns from its own failures and repeats them.
What the giants do instead
The world's biggest supply chains stopped trusting the bell curve years ago. The approach they use is conceptually simple, even though the computation is heavy:
- Forecast every product, then measure how wrong those forecasts actually were for that specific product — not in theory, in its real history.
- Resample those real misses thousands of times to build a picture of what "being wrong over a lead time" genuinely looks like for that product — spikes, droughts, lopsidedness and all.
- Size the buffer from that picture, at your chosen service level. No bell curve assumed anywhere.
The result: spiky products get the bigger buffers their real behaviour demands, stable products stop hoarding cash, and the protection you pay for is the protection you actually get. Until recently the computing power to do this per product, across a big catalogue, was an enterprise-platform luxury. It isn't any more — it's what Optimal Chain does for every product, from a spreadsheet export.
What you can do this week (even in Excel)
- Try the formula on your own numbers with our free safety stock calculator — then read its "where this breaks down" section with your spikiest best seller in mind.
- Find your five spikiest high-revenue products (biggest single-day sale ÷ average daily sale). If that ratio is over 5, the bell curve is not your friend on those lines.
- Strip stockout weeks from your history before measuring variability — zeros you couldn't sell through aren't demand data.
- Stop using one service level for everything. 97.5–99% for critical best sellers, 95% for the middle, 90% or less for the tail is a far better starting point than a blanket 95%.
See it on your own data
The fastest way to know how much this matters for your business is to look at your own catalogue. Send us a sales export — any system, Excel included — and we'll come back with a free blueprint: a forecast for every product, buffers sized from each product's real track record, and the difference against what the textbook formula is telling you today. No card, no sales calls. Plans from £199/month if you like what you see — the full price list is published.