What it is actually for
Safety stock does not cover demand. That is cycle stock, the one you rebuild at every replenishment. Safety stock covers something else: the gap between what you planned and what actually happens, during the one window in which you can no longer do anything about it — the replenishment lead time.
Two sources of variability add up there, and it is worth separating them because they are not addressed the same way:
- demand variability — customers order more than expected during the lead time;
- supply variability — the supplier delivers late, or incomplete. It is the one least often measured, for want of being tracked the way demand is.
How it is calculated
When only demand varies and lead time is reliable, the usual formula is:
Two things in that formula deserve a careful read, because they run against intuition.
The square root of lead time
Safety stock grows with the square root of lead time, not with lead time. Quadrupling the lead time does not quadruple the stock you need: it doubles it. Conversely — and this is the useful lever — dividing lead time by four only halves the stock.
The price of a service level
The z coefficient does not grow linearly. It is:
- 1.65 for a 95% service level;
- 2.05 for 98%;
- 2.33 for 99%;
- 3.09 for 99.9%.
In other words: moving from 95% to 99% service costs 41% more safety stock to remove four stockout cycles in a hundred. And aiming for 99.9% costs 87% more than 95%. The curve goes vertical well before reaching zero stockouts, which is never reachable at a finite cost.
These coefficients assume normally distributed demand. That is a working assumption, not a law: on a slow-moving portfolio, or one with infrequent orders, it understates the tail and therefore the stock required. The reasoning still holds; the coefficients have to be recomputed.
When the supplier is irregular too
The formula above assumes a reliable lead time. As soon as lead time varies too, both sources of variability have to be covered at once:
The second term is worth pausing on: it is proportional to the square of average demand. On a fast-moving item, even modest lead time irregularity can therefore weigh more than all of the demand variability. Comparing the two terms is exactly what settles, case by case, between making a supplier reliable and sharpening a forecast — and it is why this calculation should never be run without first measuring how regular deliveries actually are.
What it costs
The cost of holding stock is not limited to tied-up cash. It adds financing, warehousing, insurance, damage, obsolescence and write-downs. Depending on the sector, it is usually expressed as a far-from-negligible annual fraction of the value held — quite enough for oversized safety stock to show in the P&L.
That cost is visible, so it gets watched. The opposite cost — the stockout — almost never is: lost sale, emergency freight, production rescheduling, and sometimes a customer who does not come back. It is that imbalance in measurement, not a disagreement about theory, that makes inventory trade-offs lean the same way almost every time.
The levers that reduce it without adding risk
Safety stock is a symptom, not an isolated decision. Three levers act on its causes:
- Shorten lead time. A square-root effect, but a certain one, and it does not depend on any forecasting improvement.
- Make the supplier reliable. A steady three-week lead time needs less cover than a "one to three weeks" one. It is variability that costs, not duration.
- Improve the forecast. Reducing σD reduces stock proportionally — the most direct lever, and the slowest to obtain.
To which add pooling: merging two regional stocks into one central stock does not require the sum of the two covers, but appreciably less — variability does not strike everywhere at the same moment. The gain is paid for in delivery lead time to the customer, and the trade-off is framed in those terms.
Why the decision is hard in the room
Everything above is computable. The rule nevertheless holds up poorly against recent events: a few stockouts and the target goes up, a few quiet months and it comes back down. That is not incompetence — it is that the memory of an incident weighs more, at the moment of deciding, than a standard deviation.
The S&OP Game makes that behaviour visible: each player picks their cover every turn, and the end-of-game report overlays their stock, their backlog and the total cost of both. Everyone can therefore compare the rule they stated with the decisions they actually took — and that is precisely the subject of the debriefing.