Safety stock formula: three methods and when to use each
Updated September 15, 2026
Safety stock is the inventory you keep so that normal ups and downs in sales and deliveries don't cause a stockout. Too little and you lose sales; too much and cash sits on shelves.
Method 1: max minus average
Safety stock = (max daily sales × max lead time) − (average daily sales × average lead time)
Easy and needs no statistics. It assumes your worst sales day and your slowest delivery happen together, so it usually over-stocks. Reasonable for a handful of critical SKUs.
Method 2: service level (demand variability)
Safety stock = z × σd × √L
- z: the z-score for your target service level: 1.28 for 90%, 1.64 for 95%, 2.05 for 98%, 2.33 for 99%.
- σd: standard deviation of daily sales (on in-stock days).
- L: lead time in days.
Example: 95% service, daily sales vary by ±4 units, 16-day lead time → 1.64 × 4 × 4 = 26 units.
Method 3: demand and lead-time variability
Safety stock = z × √(L × σd² + d² × σL²)
Adds σL, the standard deviation of the supplier's lead time, and d, average daily sales. Use this when a supplier's deliveries vary: a supplier who takes 30 ± 6 days needs far more buffer than one who takes 30 ± 1.
This is the method Reorder Radar uses. It measures σL from the expected and received dates on your purchase orders, and falls back to 10% of the lead time until three orders have been received.
Try all three methods with your own numbers in the safety stock calculator.
What service level to pick
| Service level | z | Good for |
|---|---|---|
| 90% | 1.28 | Slow movers, cheap to be out of |
| 95% | 1.64 | Most products (a sensible default) |
| 98–99% | 2.05–2.33 | Best sellers, products that drive repeat orders |
Moving from 95% to 99% raises safety stock by about 40%. Spend that on SKUs where a stockout costs the most.
Limitations to know
- The formulas assume roughly normal demand. Very lumpy sellers (a few big orders a month) break that assumption; treat their numbers as a starting point.
- With less than about two weeks of in-stock history, variability can't be measured. A flat percentage of lead-time demand (for example 25%) is a safer temporary rule.