Safety stock covers variability, not average demand. Cycle stock covers the average — that’s what your reorder quantity handles. Safety stock exists purely to absorb the difference between what you expected and what happened, in both demand and supply.
This distinction matters because it tells you what drives the number. It isn’t how much you use. It’s how unpredictable your usage and your resupply are. A component you consume in huge volumes with metronomic regularity needs very little safety stock. A component you use rarely but unpredictably, from a supplier whose lead time swings between two and seven weeks, needs a lot — and the flat two-weeks-of-demand rule gets both of these badly wrong.
The Standard Formula
The commonly used version accounts for variability in both demand and lead time. It combines a service factor — derived from the service level you’re targeting — with the standard deviation of demand over lead time and the standard deviation of lead time itself multiplied by average demand. The two variability sources are combined under a square root rather than added, because they’re independent and rarely both go wrong at once.
The mechanics matter less than the three inputs: your target service level, your demand variability, and your lead time variability. If you don’t know the last two as actual measured numbers, no formula will help you.
Where It Goes Wrong in Practice
Using forecast demand instead of measured variability. The calculation needs the standard deviation of actual consumption, not a planning number.
Ignoring lead time variability entirely. For most manufacturers this is the larger of the two sources, and it’s the one most often left out because supplier lead times are recorded as a single fixed value rather than a distribution.
Applying one service level to everything. A ninety-five percent target on a critical component that stops your line is sensible. The same target on a low-value item with three alternative suppliers is expensive and pointless.
Setting it once. Demand patterns shift, suppliers improve or deteriorate, product mix changes. A safety stock level calculated eighteen months ago is describing a business that no longer exists.
Treating seasonal demand as variability. If you sell three times as much in October every year, that’s a pattern, not noise. Holding safety stock against a predictable swing means carrying inventory all year for a problem that occurs once.
A Better Approach
Segment your items, set service levels according to what a stockout actually costs, measure both variability sources from real transaction history, and recalculate on a schedule. Where demand is genuinely seasonal or trending, use a model that separates the pattern from the noise — otherwise you’ll systematically over-hold.
ticktick.ai calculates safety stock per item from live consumption and supplier lead time data, and updates the levels as those distributions change.
