Safety stock, shown honestly: the formula, the assumptions, and when it lies
The safety-stock formula is one line of algebra resting on three assumptions your demand quietly violates. Here is the whole thing, in the open.
Safety stock is the buffer you hold to survive the gap between what you forecast and what actually shows up. The textbook formula is one line:
safety stock = z × σ_LT
where z is the service-factor for your target service level, and σ_LT is the standard deviation of demand over the replenishment lead time. Clean. It also rests on three assumptions your real demand rarely honours.
Assumption 1 — demand is normally distributed
For fast, steady movers this is roughly fine. For intermittent or lumpy demand — the long tail of most SME catalogues — it is simply false. A normal curve around a low mean will happily suggest negative demand, and it badly under-buffers the spikes. This is why SangamIQ routes intermittent series to a different bench entirely, judged by MASE, and never lets a seasonal model near them.
Assumption 2 — lead time is constant
The formula above uses σ of demand only. The moment your supplier's lead time varies — and in India, between monsoon, transporter strikes and GST checkposts, it does — you need the combined variance of demand and lead time. Ignore it and your buffer is systematically too small exactly when it matters.
Assumption 3 — the parameters are known
σ is estimated from history. If you have six months of history, your σ is a guess with a wide confidence interval, and a safety stock built on it is a guess squared. This is why we refuse to compute certain numbers below nine months of history rather than hand you false precision.
What we do instead
We still compute safety stock — but every number carries a rung badge saying whether it was measured, computed, estimated or a default, and the "why" trace shows you which assumptions were in play. A buffer you can interrogate beats a buffer you have to trust.
Every number in SangamIQ shows its work — see our live accuracy page, or start free and put your own CSV through it in ten minutes.