Written and maintained by CASRAI Editorial Board
Last updated
Short answer
The safety stock formula depends on what varies. If only demand varies, use SS = Z × √(PC/T1) × σD. If only lead time varies, use SS = Z × σLT × Davg. If both vary independently, combine them under one square root; if they move together, add the two results instead. Z comes from the service level you choose (Z = 1.65 at 95%). The formulas are given by Peter L. King in APICS magazine (2011).
Definition and scope
King defines safety stock simply as inventory carried to prevent stockouts. It exists because demand fluctuates, forecasts are wrong and lead times vary. This page covers the statistical formulas for a consumable that is reordered repeatedly: gloves, pipette tips, culture media, dressings. It does not cover shelf-life, storage or clinical use of any product. For the buffer-versus-safety distinction, see safety stock vs buffer stock; for the simpler fixed-quantity approach, see par levels for medical supplies.
Rules of thumb such as holding 10 or 20 percent of cycle stock are, in King’s words, generally poor performers because they ignore how variable demand and lead time really are. That is why a formula is worth using.
The safety stock formulas and what each symbol means
All three formulas below are from King (2011). A safety stock calculation uses one of these three formulas, depending on which variation you can measure.
Symbols
- Z: the number of standard deviations that matches your target cycle service level (table below).
- PC: the performance cycle, meaning the whole time from deciding what to order to having the goods stored and ready for the next cycle. It includes the review period, communicating the order, supplier processing and delivery, not only transit.
- T1: the time unit in which σD was measured (for example one week).
- σD: standard deviation of demand per T1. σLT: standard deviation of lead time. Davg: average demand per unit of time.
Demand variability only
SS = Z × √(PC/T1) × σD. If σD is weekly and PC is three weeks, the weekly figure is multiplied by √3, not by 3.
Lead-time variability only
SS = Z × σLT × Davg.
Both vary, independently
SS = Z × √[(PC/T1) × σD² + (σLT × Davg)²]. King states this assumes the two are independent and normally distributed. If they are not independent, he gives the sum instead: SS = Z × √(PC/T1) × σD + Z × σLT × Davg. For independent variables the combined result is lower than the sum of the two parts.
Z-score and service level table
King’s table converts a cycle service level, the share of replenishment cycles that end without a stockout, into Z.
| Cycle service level | Z |
|---|---|
| 84% | 1.00 |
| 85% | 1.04 |
| 90% | 1.28 |
| 95% | 1.65 |
| 97% | 1.88 |
| 98% | 2.05 |
| 99% | 2.33 |
| 99.9% | 3.09 |
The relationship is not linear. Going from 95% to 99% raises Z from 1.65 to 2.33, about 41% more safety stock for the same variability. King says 100% is statistically unattainable and that typical goals run 90% to 98%. He also suggests setting Z by product group (strategic importance, margin, dollar volume) rather than one number for everything.
Worked example with illustrative numbers
These figures are invented for illustration only; they are not data from any lab or clinic. A consumable has weekly demand with σD = 20 boxes and average demand Davg = 120 boxes per week. The performance cycle is 3 weeks (T1 = 1 week), the lead-time standard deviation is 0.5 week, and the target is 95% (Z = 1.65).
| Case | Calculation | Safety stock |
|---|---|---|
| Demand variability only | 1.65 × √3 × 20 | about 57 boxes |
| Lead-time variability only | 1.65 × 0.5 × 120 | 99 boxes |
| Both, independent | 1.65 × √(3 × 20² + (0.5 × 120)²) = 1.65 × √4,800 | about 114 boxes |
| Both, not independent | 57 + 99 | about 156 boxes |
Two lessons follow. The independent result (114) is lower than the sum (156), so assuming independence matters. And here lead-time variability dominates, which is common where a supplier is unreliable. In King’s own film-roll example the opposite held: demand variability had almost ten times the effect of lead-time variability. Which source of variation dominates has to be measured, not assumed.
Cycle service level versus fill rate
Cycle service level counts how often a cycle has a stockout, whatever its size. Fill rate is the share of demand volume that was available when requested. King notes that fill rate is often the better measure because cycle service level ignores how big a stockout was. See fill rate and stockout for definitions in a hospital supply setting.
At a 95% cycle service level, King says that in roughly half of cycles the safety stock is never touched, in about 45% it is used and sufficient, and in about 5% a stockout still occurs. The goal is to prevent most stockouts, not all of them.
Which formula to use: a decision table
| Data you actually have | Formula |
|---|---|
| Usage history only, supplier delivers on a fixed schedule | Z × √(PC/T1) × σD |
| Steady usage, but dated order and receipt records show variable delivery | Z × σLT × Davg |
| Both histories, and late deliveries are not linked to busy periods | Z × √[(PC/T1) × σD² + (σLT × Davg)²] |
| Both histories, and they plausibly move together | The sum of the two single-source results |
The independence condition is King’s. Whether delays and demand are linked in a particular stockroom is something only its own records can show.
Data readiness checklist
- Demand and σD recorded in the same time unit (T1) that you use in the formula.
- Lead time measured over the whole performance cycle, including review and receiving, not only carrier transit.
- A target service level chosen per item group, then looked up in the Z table.
Checking that it works
King recommends monitoring after the levels are set: is safety stock consumed in about half of cycles, and are service targets met? If not, do root-cause analysis before adjusting the numbers. When the calculated level is too high, he points to improving the process, for example expediting, before simply carrying more stock. Chopra, Reinhardt and Dada (2004) add a caution: the normal approximation behind these formulas can mislead. They show that the usual conclusion, that cutting lead-time variability always lowers inventory, is flawed in the service-level range where many companies operate. For a firm operating just below that threshold, reducing lead time rather than its variability is the right lever. Treat the output as a starting estimate to be checked, not a guarantee. For related methods, see kanban inventory and PPE stockpile planning.
Common misunderstandings
- “A percentage of usage is good enough.” King says such rules of thumb generally perform poorly.
- “Higher service is just a little more stock.” Z rises nonlinearly; 99% needs about 41% more safety stock than 95% at the same variability.
- “Use the supplier’s quoted lead time.” The formula uses the whole performance cycle, and σLT needs real receipt records.
- “Safety stock removes all stockouts.” It is designed to prevent the majority; 100% is unattainable.
- “Both sources always add.” Only if they are not independent; otherwise combine under the root.
FAQ
What is the safety stock formula?
There are three forms: Z × √(PC/T1) × σD for demand variability, Z × σLT × Davg for lead-time variability, and the combined root-sum-of-squares form when both vary independently (King, 2011).
How do I do a safety stock calculation?
Pick a service level, look up Z, measure σD and σLT from your records, then apply the matching formula. The worked example above shows each case with numbers.
What Z-score should I use?
The one for your target cycle service level, for example 1.65 at 95% or 2.33 at 99%. King reports typical goals of 90% to 98%.
What is the difference between cycle service level and fill rate?
Cycle service level counts cycles without a stockout; fill rate measures the share of volume available, so it reflects stockout size.
Sources
- Peter L. King, CSCP, “Understanding safety stock and mastering its equations,” APICS magazine, July/August 2011, pp. 33-35. web.mit.edu/2.810 copy. Checked 3 November 2026.
- Chopra, S., Reinhardt, G. and Dada, M., “The Effect of Lead Time Uncertainty on Safety Stocks,” Decision Sciences, vol. 35, no. 1, Winter 2004. Kellogg School of Management copy. Checked 3 November 2026.
