Safety Stock: What It Is and How to Calculate It

Safety stock is the deliberate buffer you hold beyond expected demand, sized for the weeks that refuse to be average: the demand spike, the late container, both at once. It is not padding and it is not fear; it is a calculated answer to variability. This guide gives you the standard formula, two worked examples, and the honest shortcut for teams without a demand planner.

Why average-based planning fails

Suppose you sell an average of 70 units a week and your supplier delivers in two weeks. Expected demand during the resupply window is 140 units, so you reorder when stock hits 140 and everything works, on average. But "average" means half your weeks are busier. The first 90-unit fortnight, or the first shipment that lands four days late, and the shelf goes empty. The cost of that emptiness is the subject of our stockouts guide; safety stock is the insurance you buy against it, priced in held inventory.

The standard formula

safety stock = Z × σdLT

Z is your service-level factor and σ_dLT is the standard deviation of demand over the lead time. In the common case where demand varies but the lead time is fairly stable:

safety stock = Z × σd × √L

where σ_d is the standard deviation of daily demand and L is the lead time in days. The Z value sets how often you accept running out:

Service level	Z	Meaning
90%	1.28	Out of stock in roughly 1 in 10 replenishment cycles
95%	1.65	The common default for A items
98%	2.05	For SKUs where a stockout is a fired customer
99.9%	3.09	Hospital-grade; expensive to hold

Worked example

A coffee roaster's 250 g espresso bags sell 10 a day on average, with a daily standard deviation of 4 units. The supplier lead time is 10 days, and the owner wants a 95 percent service level.

safety stock = 1.65 × 4 × √10 = 1.65 × 4 × 3.16 ≈ 21 units

Twenty-one bags ride along as the buffer. The reorder point becomes expected lead-time demand (100) plus safety stock (21) = 121 units.

When lead time varies too

If your supplier's "10 days" ranges from 8 to 16, variability compounds and the fuller formula applies:

safety stock = Z × √(L × σd² + d̄² × σL²)

where d̄ is average daily demand and σ_L is the standard deviation of the lead time in days. In the example above, if σ_L is 2 days: √(10 × 16 + 100 × 4) = √560 ≈ 23.7, so 1.65 × 23.7 ≈ 39 units. Nearly double. Late suppliers, not spiky demand, are what quietly inflate safety stock, which is why tracking quoted-versus-actual lead time per supplier pays for itself.

The honest shortcut

Most small teams do not have clean daily standard deviations, and pretending otherwise produces false-precision buffers. A defensible heuristic: hold half a lead time of demand as safety stock. The roaster above: half of 10 days at 10 a day = 50... which is generous; tighten toward the formula as your data improves. The heuristic's virtue is that it scales with the two things that matter (demand rate and lead time) and it can be computed for a thousand SKUs with no statistics at all. It is the screening default our own stock scan uses on pasted rows, stated plainly in the results.

Three mistakes to avoid

• One blanket buffer for the whole catalog. A items deserve computed buffers; C items can run leaner. Averaging across the catalog buys too much of the cheap risk and too little of the expensive one.
• Setting it once. Demand and lead times drift; a buffer sized in January lies by July. Recompute on a schedule, or use software that recomputes continuously.
• Confusing safety stock with dead stock. A buffer on a dying SKU is not safety, it is dead stock with a job title. When the sales rate collapses, the buffer should too.