Supply Chain
Safety Stock: The Formula and When It Lies to You
The standard formula is sound and its assumptions are usually false. Knowing which assumption your business breaks tells you which direction the number is wrong in.
Safety stock exists to absorb two kinds of surprise: demand that is higher than forecast, and supply that arrives later than promised. The textbook formula handles both, and it is a good starting point provided you know what it assumes.
The formula
The two terms under the root are the two risks, combined the way independent variabilities combine. The first is demand risk over the lead time; the second is lead-time risk at the average demand rate. It is worth calculating them separately once, because the larger term tells you where to spend your effort — and for most businesses it is the lead-time term, which is a supplier conversation rather than a forecasting one.
Where the assumptions break
• Normally distributed demand — false for spare parts and slow movers, where demand is intermittent and the formula badly understates cover.
• Independent demand and lead time — false in a shortage, when everybody orders more at once and lead times stretch together.
• Stationary averages — false for anything seasonal or trending; a twelve-month average is wrong in both directions across the year.
• Reliable lead-time data — often the promised time, not the delivered time.
The correlation between demand spikes and lead-time stretch is the reason formula-driven stock fails in exactly the month you needed it.
Use the measured lead time
The single most common error is feeding the formula the supplier's quoted lead time. Use the actual receipt dates from your own history. The variance is the point of the exercise, and a quoted time has none — it is a constant, which silently deletes half the formula.
Service level is a commercial decision
Z is not a technical parameter. Moving from 95% to 99% roughly doubles safety stock for the same variability, and that money buys four percentage points of availability. Whether that is worth it depends on the margin, the cost of a stockout, and whether the customer waits or leaves. Set it per item class, never once for the whole catalogue.
What to do about the slow movers
For intermittent demand, drop the normal assumption and model the demand distribution directly, or accept a simple policy: hold one or two units of a cheap critical part and review on consumption. A crude rule that acknowledges lumpy demand beats an elegant formula built on a distribution the item does not follow.
Recalculate quarterly, not annually, and always after a supplier change. Safety stock is a snapshot of variability, and variability is the thing that moves.
Splitting the two terms under the root and looking at which dominates was the most actionable thing here. Ours was 80% lead-time variance, which meant the fix was a supplier conversation, not a better forecast.
Feeding the formula the quoted lead time is such a common error. A constant has no variance, so half the formula silently evaluates to zero and the output looks reassuringly small.
The correlation caveat is the one that bit us in 2021. Demand spiked and lead times stretched simultaneously across every supplier, which is precisely the scenario the independence assumption excludes.