How do you calculate safety stock?

When demand and lead time both vary: safety stock = z × √(average lead time × demand variance + average demand² × lead-time variance). Pick z from your target service level (1.65 for 95%), use consistent time units, and add the result to expected demand during lead time to get the reorder point.

What is the z-score in the safety stock formula?

It is the standard-normal factor for your target cycle service level, the probability of not stocking out during a replenishment cycle. Common values: 1.28 for 90%, 1.65 for 95%, 2.05 for 98%, 2.33 for 99%, 3.09 for 99.9%. Higher service costs disproportionately more buffer.

What is the difference between safety stock and reorder point?

Safety stock is the buffer quantity itself. The reorder point is the on-hand-plus-on-order level that triggers a new order: average demand times average lead time, plus safety stock. The first term covers the expected wait; safety stock covers variability during it.

What service level should safety stock target?

Set it per item by the cost of a stockout, not plant-wide. Items that stop a production line or lose a customer justify 99% or higher; slow movers with tolerant demand can run lower. Because the z-score climbs steeply past 95%, a uniform high target wastes cash on items that never needed it.

Does safety stock work for slow-moving or intermittent demand?

The standard formula works poorly there, intermittent demand is not normally distributed, so the z-table misestimates risk. For slow movers and spares, plan from criticality and lead time (can you wait out the replenishment?) and treat any formula output as a rough starting point to be reviewed.

How often should safety stock be recalculated?

Quarterly is a reasonable default, plus immediately when something structural changes: a new supplier, a lead-time shift, a demand trend or seasonality change. A safety stock set once and never revisited drifts into either stockouts or dead cash, usually both, on different items.

Why do stockouts still happen with safety stock in place?

Common causes: the on-hand record was wrong (an accuracy problem, not a buffer problem), demand and lead time moved together during a shortage, the inputs were stale, or the buffer was sized on polluted history. Fix record accuracy with cycle counting first, no formula survives a wrong on-hand number.

Safety Stock: The Formula and the Judgment

The z-score math, a worked example, and the assumptions the formula quietly makes.

Safety stock is extra inventory held beyond expected demand to absorb variability in demand and supplier lead time. The standard formula is safety stock = z × σ_dLT where z is the service-level factor (1.65 for 95 percent) and σ_dLT is the standard deviation of demand over the replenishment lead time. It is a statistical buffer, not a fudge factor, and it only works when the inputs are honest.

This post gives the formula, works a full hypothetical example, shows the service-level trade-off curve, and then does the part most explainers skip: where the formula breaks, and what judgment has to fill in.

What Is Safety Stock For?

Safety stock exists to keep a target service level despite two kinds of variation: demand that arrives faster than forecast, and replenishment that arrives later than promised. Cycle stock, the part that drains and refills every order cycle, handles the averages. Safety stock handles the deviations.

Normal cycles drain to the top of the safety band. The band exists for the bad weeks: the late truck and the demand spike both dip into it without stopping the line.

The reorder point ties the two together: reorder point = (average demand × average lead time) + safety stock. When on-hand plus on-order falls to that level, you order. The first term covers the expected wait; the safety stock covers the bad weeks. This logic applies wherever replenishment is triggered by a stock level, finished goods in a make-to-stock operation, raw materials, and maintenance spares alike.

What Is the Safety Stock Formula?

When both demand and lead time vary independently, the combined form is:

Safety stock = z × √(L̄ × σ_d² + d̄² × σ_L²)

z the service-level factor from the standard normal distribution: 1.28 for 90 percent, 1.65 for 95 percent, 2.33 for 99 percent, 3.09 for 99.9 percent.
d̄ and σ_d average demand per period and its standard deviation (same period units as lead time).
L̄ and σ_L average lead time and its standard deviation.

If only demand varies (reliable lead times), it collapses to z × σ_d × √L̄. If only lead time varies, it collapses to z × d̄ × σ_L.

A worked example (hypothetical)

The numbers below are illustrative, not from a real plant. A component with:

Input	Value
Average demand (d̄)	100 units/day
Demand std. dev. (σ_d)	20 units/day
Average lead time (L̄)	10 days
Lead time std. dev. (σ_L)	2 days
Target service level	95% → z = 1.65

Step by step:

Demand-variability term: L̄ × σ_d² = 10 × 20² = 4,000.
Lead-time-variability term: d̄² × σ_L² = 100² × 2² = 40,000.
Combine: √(4,000 + 40,000) = √44,000 ≈ 210 units. Note what dominates: the supplier's 2-day wobble contributes ten times more variance than the daily demand noise.
Apply the service factor: 1.65 × 210 ≈ 346 units of safety stock.
Set the reorder point: (100 × 10) + 346 = 1,346 units.

That step-3 observation is the most useful thing the formula teaches: for many items, lead-time reliability is the cheapest safety stock reduction available. Cutting σ_L from 2 days to 1 drops the buffer from ~346 to ~244 units, a 30 percent inventory reduction from a supplier conversation rather than a reorder-point edit.

How Much Does a Higher Service Level Cost?

Safety stock scales linearly with z, but z scales brutally with service level. Moving from 90 to 95 percent raises the buffer about 29 percent (1.28 → 1.65). Moving from 95 to 99.9 nearly doubles it (1.65 → 3.09), and by definition, that last increment defends against events that almost never happen.

Safety stock scales with the z-score, and z climbs steeply past 95 percent. The last nine is the expensive one, choose service levels per item, not plant-wide.

The practical consequence: set service levels per item, not plant-wide. A $2 fastener that stops a line earns 99 percent or more. A slow-moving finished good with a tolerant customer does not. This is the same criticality logic used in spare parts inventory management where the downtime cost of a stockout, not the part price, sets the target.

Where Does the Formula Break? The Judgment Part

The formula is honest math built on assumptions that are frequently false on real data. Know them before you trust the output:

It assumes roughly normal, independent variation. Real demand is often lumpy (three quiet weeks, then one big order), seasonal, or trending, and intermittent demand for slow movers is nothing like a bell curve. For those items the z-table quietly lies; treat the output as a starting point, not a verdict.
It assumes demand and lead time do not move together. In shortages they do: everyone orders more, and lead times stretch at the same time. The formula underestimates exactly when you need it most.
Forecast error, not raw demand variation, is the right σ_d if you plan from a forecast. Measuring variability around a good forecast can cut the buffer; ignoring forecast bias inflates every downstream number.
Garbage inputs, garbage buffer. Lead-time statistics computed from purchase-order dates that nobody maintains, or demand history polluted by stockouts (you cannot ship what you did not have), will mis-size every item. And a perfectly sized buffer is useless when the on-hand record itself is wrong, which is why cycle counting is a prerequisite for statistical inventory control, not an alternative to it.
It is per-item math in a multi-item world. Shared capacity, shelf life, storage limits, and cash constraints all live outside the formula.

The working posture: recalculate quarterly (or when a supplier or demand pattern changes), review the items where the formula and the planner disagree, and record why overrides were made. The plants that get this right treat safety stock as a living number driven by live data. For context on how much capital is at stake nationally, the U.S. Census Bureau's Manufacturing and Trade Inventories and Sales series put total business inventories near $2.7 trillion at the end of March 2026, with an inventories-to-sales ratio of 1.32, buffers are a material share of American balance sheets, which is why sizing them with statistics instead of fear pays.

How Do You Keep Safety Stock Honest Over Time?

Three habits separate a living buffer from a fossilized one. First, watch actual dips: if an item never touches its safety band across many cycles, the buffer is too big; if it dips every cycle, the buffer is too small or an input is stale. Second, trend your suppliers' real lead times against promises, σ_L is a measured number, not a contract term. Third, connect the data: when receipts, usage, and stockouts are captured live rather than batch-entered, the statistics feeding the formula stop being archaeology. That inventory-signal problem, spotting gaps and shortages early from live floor data, is one of the things Harmony's platform is built to do (see the inventory and shortage intelligence module). The same live-record discipline covers storeroom spares, where MRO inventory stockouts convert directly into downtime hours.