Ordering in time while retaining a buffer against uncertainty.
All notes

Supply chain

Safety stock and reorder point: formulas and example

Calculate safety stock and the reorder point for one spare part. Compare demand and supplier delays, distinguish cycle service from fill rate, and price a higher service target.

A distributor needs enough spare parts to cover the wait for replenishment. Its buyer has to allow for two uncertainties: how much customers will need and when the supplier's delivery will become usable.

Safety stock is extra inventory held to cover uncertainty in demand and replenishment. The reorder point combines expected demand during lead time with that buffer. The calculation below uses one spare part to compare demand variation, delivery variation and the cost of a chosen service target.

The example product

A distributor stocks a fast-moving spare part. From the last six months of data:

Input Value
Average daily demand (d) 120 units
Standard deviation of daily demand (σd) 30 units
Average supplier lead time (L) 9 days
Standard deviation of lead time (σL) 2 days
Target cycle service level 95% (z = 1.645)
Order quantity (Q) 1,200 units

Step 1: demand during the lead time

The reorder point exists to cover the wait between placing an order and receiving it. On an average day, that's:

Expected demand during the lead time

d × L = 120 × 9 = 1,080 units

With constant demand and lead time under this model, the buyer would reorder at 1,080 and receive the delivery as stock ran out. The buffer covers departures from that expected path.

Step 2: safety stock when only demand varies

The textbook starting point assumes the lead time is fixed and daily demand varies independently from day to day. Over L days, the variation grows with the square root of L, not with L itself, because good and bad days partly cancel out.

Demand variability only

σ over the lead time = σd × √L = 30 × √9 = 90 units

Safety stock = z × 90 = 1.645 × 90 = 148 units

Step 3: add lead-time variability

The supplier's lead time also varies: sometimes 7 days, sometimes 12, around the 9-day average. A delayed receipt leaves the distributor serving demand for longer. The next calculation includes both, assuming demand and lead-time variation are independent and the normal approximation is suitable.

Demand and lead-time variability

σ = √( L × σd² + d² × σL² )

= √( 9 × 30² + 120² × 2² ) = √( 8,100 + 57,600 ) = 256 units

Safety stock = 1.645 × 256 = 422 units

Look at the two terms under the square root. Demand variation contributes 8,100. Lead-time variation contributes 57,600, seven times more. The safety stock nearly triples, from 148 to 422 units, and almost all of the increase comes from the supplier, not the customer.

In this example, improving supplier delivery reliability has the larger effect. Cutting σL from 2 days to 1 would bring the safety stock down to about 247 units, a 41% reduction, with the same demand inputs.

Step 4: the reorder point

Reorder point

ROP = demand during the lead time + safety stock = 1,080 + 422 = 1,502 units

Place the next order when inventory position reaches 1,502. Calculate that position from on-hand stock plus outstanding replenishment, less backorders; confirm the existing receipts before creating another order.

Reorder point 1,502 Safety stock 422 Lead time, 9 days Order placed 0 Day 10 Day 20 Day 30 0 600 1,200 1,800 Buffer for bad days
The classic sawtooth. Stock falls by about 120 units a day. The order goes out at 1,502 and arrives nine days later, on an average cycle just as stock reaches the safety stock. On a bad cycle, the buffer absorbs the difference.

What service level really costs

The service level decides z, and z doesn't grow in a straight line. The higher you go, the more each extra point costs.

Cycle service level z Safety stock (units)
90% 1.282 328
95% 1.645 422
98% 2.054 526
99% 2.326 596
99.9% 3.090 792
0 200 400 600 800 80% 85% 90% 95% 100% Cycle service level units 95%: 422 99%: 596 99.9%: 792
Safety stock against service level for the example. The curve is gentle up to about 95%, then climbs steeply.

Put a price on it. If the part costs €20 and holding inventory costs 25% of its value a year, each unit of safety stock costs €5 a year. Moving from 95% to 99% adds 174 units, or €870 a year for this one part. For 2,000 parts with those same costs and uncertainties, the increase would exceed a million euros a year. Real items differ, so assess the service requirement and cost for each group.

A common approach is to set targets by ABC class (high-value or high-margin items first), and adjust for how critical the item is to the customer. Spare parts that stop a production line may justify 99%. Slow-moving accessories may be fine at 90%.

Cycle service level is not fill rate

Agree which service measure the team is targeting:

  • Cycle service level The share of replenishment cycles that end without a stock-out. That's what the z in the formula targets.
  • Fill rate The share of demand shipped straight from stock. It's usually what customers actually feel.

They can be very different numbers. In the example, a 95% cycle service level means 1 cycle in 20 runs short. But when a cycle does run short, it's usually by a few units, not by all 1,200. Using the standard normal loss function, the expected shortage per cycle is about 5 units:

Expected fill rate at 95% cycle service level

Expected units short per cycle = σ × G(z) = 256 × 0.0209 ≈ 5.4 units

Fill rate = 1 − 5.4 ÷ 1,200 ≈ 99.6%

So a 95% cycle service level delivers roughly a 99.6% fill rate here. If your contract promises a 98% fill rate, you may need far less safety stock than a 98% cycle service level would suggest. Agree on the definition before agreeing on the number.

Getting the inputs right

The formula is only as good as what you feed it:

  • Use forecast error, not raw demand variation, if you forecast. If demand has a trend or seasonality you already predict, the uncertainty is the error around the forecast, not the swings in demand itself.
  • Match the time buckets. If you measure weekly variation, convert with the square root: σdaily ≈ σweekly ÷ √(days per week you sell).
  • Measure lead time from order to available stock, including receiving and quality checks, not just the supplier's quote.
  • Check the normal assumption for slow movers. For items that sell a few units a month, the normal distribution fits poorly. Poisson-based or simulation methods work better.
  • Be careful at low service levels. Research by Chopra, Reinhardt and Dada found that below roughly 50% to 70% cycle service level, the normal approximation can give the wrong advice, even suggesting that more reliable suppliers need a higher reorder point. The formulas here are for the 90% to 99% range most planners work in.
  • Recalculate regularly. Monthly or quarterly, and after any big change in supplier or demand pattern.

Before claiming the forecast improved, check what its score measures. The MAPE/WAPE comparison and calculator shows why two metrics can choose different winners. For slow movers with quiet weeks, see forecast accuracy with zero demand.

Compare changes to the buffer inputs

Use the model to compare the changes the team can actually make:

Lever What it changes Effect
More reliable supplier deliveries σL Biggest reduction in this example
Shorter lead time L Less demand exposure while waiting
Better forecasts σd (forecast error) Smaller buffer for the same service
Differentiated service targets z Stock where it earns its keep
Reviewing more often The period of uncertainty Smaller buffer in periodic-review systems

For this part, check the supplier's actual delivery spread before changing the buffer. Recalculate after any agreed delivery improvement, and track shortages alongside the stock reduction.

Why busy machines make customers waitInventory and lead time are two sides of the same flow problem. Little's Law explains the other side.

Sources

The product and its numbers are illustrative.

NextWhat to automate first: a scoring model for processes, and when to use RPA, workflows or AI agents

Discuss an operations problem

I work with operations and transformation teams on operational excellence, digital transformation programs, supply chains and industrial AI. Send me the problem you’re working on and the result you need.

Abolfazl ShirkavandImage preview unavailable