Safety Stock Calculator
Three formulas, one calculator. Pick the method that matches your data, set a service level, and see what each extra point of service actually costs you in stock.
Calculate Safety Stock and Reorder Point
Fields update as you choose a method. Everything runs in your browser: no sign-up, no server call, and nothing you type is sent anywhere.
Formula applied
Fill the fields to see the formula with your numbers substituted in.
What this means
Enter your figures to see how this buffer behaves against your demand and lead-time pattern.
The Three Safety Stock Formulas, and When Each One Applies
Most pages present one formula as though it were the formula. It is not. The three below answer slightly different questions, and picking the wrong one is the most common reason a buffer that looked right on a spreadsheet fails in the warehouse.
| Method | Formula | What it needs | Use it when | Where it fails |
|---|---|---|---|---|
| Basic (max minus average) | (Max daily sales × Max lead time) − (Avg daily sales × Avg lead time) | Four numbers off a sales report and a receiving log. No statistics. | You are setting a first buffer or your history is too short to trust a standard deviation. Also useful as a sanity ceiling. | No service level built in. The answer is whatever your worst observed month happened to be, so one freak week permanently inflates the buffer. |
| Statistical, demand variability only | Z × σ(demand) × √(Lead time) | Daily demand deviation, average lead time in days, and a service level. | The supplier is genuinely reliable: a domestic transfer, a local contract manufacturer, or a lane where lead time barely moves. | Assumes lead time is fixed. If your supplier is ever late, this understates the buffer, sometimes several times over. |
| Full formula (demand and lead time) | Z × √( LT × σ²(demand) + Demand² × σ²(LT) ) | Everything above plus the deviation of lead time in days. | Cross-border sourcing, customs clearance, port congestion, and any lane where the arrival date genuinely moves. Most cross-border inbound qualifies. | Needs real receipt history. Guessing σ(LT) low is the usual way to get a comfortable but wrong answer. |
Why the third one matters for cross-border sourcing
The full formula is the only one that reacts to an unreliable supplier. Set the lead-time deviation to zero and it collapses back to the simple formula exactly. Every day above that adds buffer.
In cross-border lanes the lead-time term usually dominates. In the worked example below it accounts for 95% of the variance under the square root, which is why the simple formula returns barely a fifth of the stock actually required.
Service Level to Z-Score Table
Both statistical formulas need a Z score. It is the number of standard deviations of cover that delivers your chosen cycle service level, taken from the one-tailed normal distribution.
| Cycle service level | Z score | Expected stockouts | Typical use |
|---|---|---|---|
| 50% | 0.00 | 50 of 100 cycles | No buffer at all. Safety stock is zero by definition. |
| 80% | 0.84 | 20 of 100 cycles | Low-margin C-class items where substitution is easy. |
| 85% | 1.04 | 15 of 100 cycles | Slow movers and long-tail SKUs you stock but will not protect heavily. |
| 90% | 1.28 | 10 of 100 cycles | B-class items with steady demand and a supplier who can expedite. |
| 95% | 1.65 | 5 of 100 cycles | The usual target for A-class items, and the default most operators land on. |
| 97% | 1.88 | 3 of 100 cycles | High-velocity marketplace listings where losing the buy box is expensive. |
| 98% | 2.05 | 2 of 100 cycles | Top revenue SKUs and lines under a retailer fill-rate agreement. |
| 99% | 2.33 | 1 of 100 cycles | Critical items where a stockout carries a contractual penalty. |
| 99.9% | 3.09 | 1 of 1,000 cycles | Effectively never short. The money is better spent on lead-time reliability. |
Z values are the inverse normal distribution rounded to two decimals. One convention to be aware of: at 95% the exact value is 1.6449, and the long-standing practice is to round it up to 1.65. This calculator and the table above both use 1.65.
Cycle service level is the probability of not stocking out during a replenishment cycle. It is not fill rate, which measures the share of unit demand met from stock and is usually higher for the same buffer.
Why 99% Service Costs So Much More Than 95%
Service level and stock do not rise in a straight line. Z climbs gently through the middle of the range and steeply near the top, so the last few points of service are the expensive ones.
| Service level | Z | Safety stock | Step up from previous | Stock value |
|---|---|---|---|---|
| 50% | 0.00 | 0 units | — | $0 |
| 80% | 0.84 | 620 units | +620 units | $8,675 |
| 85% | 1.04 | 767 units | +148 units | $10,740 |
| 90% | 1.28 | 944 units | +177 units | $13,219 |
| 95% | 1.65 | 1,217 units | +273 units | $17,040 |
| 97% | 1.88 | 1,387 units | +170 units | $19,415 |
| 98% | 2.05 | 1,512 units | +125 units | $21,171 |
| 99% | 2.33 | 1,719 units | +207 units | $24,062 |
| 99.9% | 3.09 | 2,279 units | +561 units | $31,911 |
Calculated with the full formula on the example inputs: 120 units average daily sales, 21-day average lead time, demand standard deviation 35 units, lead-time standard deviation 6 days, unit cost $14. Change the calculator inputs and this table follows them.
90% to 95%
+273 units
Five points of service for 29% more buffer. Usually the best-value step on the curve.
95% to 99%
+502 units
Four points for 41% more buffer. Worth it on hero SKUs, hard to justify catalogue-wide.
99% to 99.9%
+561 units
Under one point for a third more stock again. Fixing the lane is cheaper than this.
The cheaper lever nobody uses
Look at what sits under the square root: lead-time variance is multiplied by demand squared. Cutting σ(LT) from six days to two on the example inputs drops the 95% buffer from 1,217 units to 476 — a 61% reduction with no change to service level.
Booking fixed sailings, pre-clearing customs, or holding stock closer to the customer all attack that term. Raising the service level treats the symptom; reducing lead-time variance removes the cause.
An Overseas Supplier to a US Fulfilment Centre
A homewares brand sells one hero SKU across the US, replenished from an overseas Minh City supplier by sea freight. Twelve months of order history gives the figures below.
Average daily sales
120 units
Across all US channels combined, measured over trading days.
Average lead time
21 days
Purchase order to goods received at the US fulfilment centre.
Demand std deviation
35 units
Daily standard deviation of sales over the same twelve months.
Lead-time std deviation
6 days
Measured across 18 receipts. Peak-season sailings and customs drive most of it.
Peak daily sales hit 190 units and the slowest receipt took 34 days. At a 95% service level, here is what each method returns on exactly the same data.
| Method | Substituted | Safety stock | Reading |
|---|---|---|---|
| Basic | (190 × 34) − (120 × 21) | 3,940 units | Covers the worst historical case of both variables at once. Over three times the full formula, and 33 days of cover sitting idle most of the year. |
| Demand variability only | 1.65 × 35 × √21 | 265 units | Just over two days of cover. It ignores the fact that the ship can arrive 12 days late, so it is badly short for this lane. |
| Demand and lead time | 1.65 × √(21 × 35² + 120² × 6²) | 1,217 units | About 10 days of cover, roughly $17,000 at $14 a unit. The right answer for this lane, and the reorder point becomes 3,737 units. |
The spread between 265 and 1,217 units on identical data is the argument for choosing the method deliberately. The basic method's 3,940 units answers a different question: it protects against a worst case, not a stated service level.
Where the Numbers Come From
The formulas are trivial arithmetic. Almost every wrong answer comes from an input that was estimated, mismatched, or quietly measured over the wrong window.
Match your time units
If demand is per day, lead time and its deviation must be in days too. Mixing weekly demand with daily lead time is the most frequent error.
Deviation of demand
Take daily sales over at least 12 weeks. In a spreadsheet that is STDEV.P or STDEV.S across the daily column.
Deviation of lead time
Use actual order-to-received dates, not the supplier's quoted lead time. Ten receipts is a workable minimum; fewer than five is guesswork.
Exclude stockout days
Days with nothing to sell record as zero demand and drag the average down. Strip them out or the buffer shrinks every time it fails.
Before you trust the number
- Recalculate per SKU, not per category. Averaging a fast mover with a slow one suits neither.
- Re-run it quarterly, and immediately after a supplier, lane, or price change.
- Separate promotional and launch periods from the baseline before measuring deviation.
- On a review-period model, add the review interval to lead time before calculating.
- Sanity-check against days of cover. A 60-day buffer means the inputs deserve a second look.
- Confirm the unit of measure. Safety stock in cartons and demand in eaches is a costly mismatch.
- Agree who owns the number. Buffers drift when nobody reviews them.
When These Formulas Stop Working
All three methods assume demand during lead time is roughly normally distributed and reasonably stable. That assumption holds for a lot of steady ecommerce lines. It fails clearly in three situations, and using the formula anyway produces a confident wrong number.
Intermittent demand
Slow movers with many zero-sales days are not normally distributed. Deviation is large against the mean, so the formula returns an implausible buffer.
Instead: use an intermittent-demand method such as Croston's, or set the buffer from a target order-fill decision.
Strong seasonality
A twelve-month deviation blends peak and trough into one number. It over-buffers the quiet months and under-buffers the peak, which is backwards.
Instead: calculate a buffer per season, or apply the Z score to forecast error rather than raw demand.
Launches and promotions
A new SKU has no meaningful history, and a promotion is a deliberate spike rather than random variation. Neither is what these formulas describe.
Instead: plan that cover from the demand plan and supplier commitment, then start the statistical buffer once a baseline exists.
What safety stock cannot fix
Safety stock absorbs random variation. It does not absorb a supplier who is structurally late, a forecast biased low, or a lane with a peak-season capacity ceiling. Those show up as a rising buffer that never seems to be enough.
Order quantity is a separate decision. Economic order quantity logic sets how much to order and how often; safety stock sets the floor you never plan to touch.
Planning, Freight, and Fulfilment Tools
Safety stock sets how much buffer to hold. These free tools cover what it costs to move and store that stock once you have decided.
3PL Cost Calculator
Model storage, pick and pack, and receiving costs to price the buffer you just calculated.
CBM Calculator
Convert carton dimensions into cubic metres to see the space a replenishment order needs.
Purchase Order Template
Turn a reorder point trigger into a clean purchase order with the line detail your supplier needs.
Browse the full set on the free tools hub. If holding the buffer is the constraint rather than calculating it, see how Locad handles warehousing and storage and multi-market fulfilment.
Safety Stock Questions
Holding Less Buffer for the Same Service Level
Shorter, steadier lead times cut safety stock faster than raising your service level ever will. Talk to Locad about warehousing and fulfilment for your stock.