Safety stock is the buffer of units you keep above what you expect to sell, so you don't run out while the next replenishment is on its way. The classic formula is SS = Z × σ × √L. Its parameters are the factor for the service level you want (Z), how much your sales vary from one week to the next (σ) and how long the replenishment takes to arrive (L).
On Black Friday all three can change at once: sales vary more, goods can take longer to arrive and a stockout costs more. With a normal week's buffer, you can fall short just when you sell the most.
The hard part isn't the formula but having the data to feed it: the sales of each size in each store, week by week. With the product's total you'd get a buffer that doesn't tell you how much to keep of each size.
What is safety stock and how is it calculated?
We use the formula as Peter L. King explains it in APICS Magazine, the magazine of the professional association for planning and supply chain management. To use it you need to understand two concepts: how variable your sales are and the service level you choose.
Let's imagine Paula runs a coat brand with two stores, Centre and North, and an online shop (the figures are made up). Her bestseller this winter is a black puffer jacket, and the warehouse sends each store an order every Monday, which arrives the same day.
Sales variability (σ)
Over the last eight weeks, the Centre store has sold between 12 and 28 puffer jackets in size M a week: 20 on average. If Paula keeps only those 20 in the store, she'll run out of the M before the next delivery in any week when sales beat the average.
The standard deviation (σ) measures how far sales usually stray from that average. For the M in Centre it's 6 units; any spreadsheet's standard deviation function calculates it. The more irregular your sales, the bigger the buffer you need.
Service level (Z)
The service level (King calls it the cycle service level) is the percentage of replenishment cycles in which you don't run out of stock before the next delivery arrives. No buffer reaches 100%, and each extra point costs more units than the last. Each level has its Z factor:
| Service level | Z | Buffer for the M in Centre |
|---|---|---|
| 90% | 1.28 | 8 |
| 95% | 1.65 | 10 |
| 98% | 2.05 | 12 |
| 99% | 2.33 | 14 |
| 99.9% | 3.09 | 19 |
Going from 90% to 99% nearly doubles the buffer, from 8 to 14 units, to cut stockouts from 10 in every 100 weeks to 1. Is it worth it? It depends on the product.
King puts the usual targets between 90% and 98% and recommends setting them by product group, according to their importance, their margin or their sales value.
Applied to Datcisions' methodology, which concentrates the budget on the models that bring 80% of revenue, that means a higher level for those models, where a stockout at full price is a lost sale, and a lower one for secondary styles, where the buffer ends up in the sales.
Paula chooses 95% for the puffer jacket: it's her bestseller, and in a normal week, if the M sells out, more arrives the following Monday. With L = 1 week, the time until the next delivery arrives (L is in weeks because σ is weekly):
SS = 1.65 × 6 × √1 ≈ 10 units.
Every Monday, Paula tops the M in Centre up to 30 units: the 20 she sells on average until the next delivery, plus the 10 of safety stock. According to King, at 95% the buffer isn't touched in half the weeks, is used in another 45 in every 100 and runs out in about 5.
The North store sells 6 M a week, with σ = 3. Its buffer: 1.65 × 3 ≈ 5 units. Same puffer jacket, same size, half the buffer. If you calculate safety stock from the brand's total, you can't tell how much to keep in each store.
Why does the formula fail on Black Friday?
Because the formula assumes sales go up and down around a stable average, and on Black Friday the average itself changes. With the usual buffer, Paula would have 10 units to spare in the week when her forecast could be furthest off.
Last Black Friday, Centre sold 60 puffer jackets in size M that week, three times a normal one. Safety stock doesn't cover that rise; the forecast does: Paula needs the 60 she expects to sell in the store. The buffer covers what the forecast gets wrong, and that's where all three variables can change:
- σ can grow with volume. Missing by 30% on 20 units isn't the same as on 60. In this example, Paula assumes the deviation triples with sales, from 6 to 18. It's an assumption: to check it she needs the history of several campaigns and high-selling weeks, size by size and store by store.
- L gets longer. Last year, the following Monday's delivery arrived a week late: the warehouse was packing online orders at the same time. And if the M sells out on Friday, no more arrives until the campaign is over.
- Paula raises Z. A stockout on Black Friday is expensive: it's a week of heavy traffic, perhaps paid for with advertising, and there's no other like it to win that sale back. For this product, she goes from 95% to 98%.
| Scenario | Z | σ | L | Buffer |
|---|---|---|---|---|
| Normal week | 1.65 | 6 | 1 | 10 |
| More variability | 1.65 | 18 | 1 | 30 |
| Longer lead time | 1.65 | 18 | 2 | 42 |
| Higher service level | 2.05 | 18 | 2 | 52 |
The buffer for the M in Centre goes from 10 to 52 units, more than five times a normal week's, and it comes on top of the forecast sales for those two weeks. If Paula applied that increase to the whole catalogue, she'd fill her stores with stock she won't sell.
How do you raise the buffer without overstocking?
Raise it only where a stockout really costs you, work it out store by store and decide in advance what you'll do if there isn't enough stock for every store.
Only on items that keep their price during the campaign. On Black Friday you also sell stock from other seasons and items on promotion.
An extra buffer on an item at 30% off is stock that, if left over, will go to the sales at an even deeper discount. Keep the higher service level for items sold at full price or with only a small discount: Paula's puffer jacket, yes; last season's jumpers, no.
Store by store. With the same changes, North's buffer rises from 5 to 26 units (2.05 × 9 × √2), and Centre's to 52. Splitting it equally would leave North with units to spare and Centre with fewer than it needs. Replenish first where stock will run out soonest: rotation comes first, not volume.
If there isn't enough for every store, concentrate it. If the warehouse can't cover both stores' forecast and buffer, send the M to the store with the highest rotation. Each unit there is more likely to sell before the end of the season instead of ending up in the outlet.
When should you bring it back down?
At the first replenishment after the campaign, if sales and deliveries are back to normal. In 2026, Black Friday is on Friday 27 November and Cyber Monday on Monday 30 November. The buffer of 52 was worked out for a week that's over: if you keep it, Centre holds 52 M of safety stock to sell 20 a week.
Go back to the normal parameters (95%, σ = 6, L = 1, 10 units) and top the M in Centre up only to 30, its usual level; if it already has more, don't send it that size. If the promotions or the delays carry on, recalculate the buffer for those weeks.
If you keep replenishing with the campaign's buffer, whatever is left over will reach the January sales, and every unit sold at a discount is margin you don't get back.
Then check the result, as King recommends: was the buffer used in roughly half of the sizes and stores? It's normal for a size not to use its buffer; if it ran out in several, note which ones and look for the cause before changing anything. That record is your best forecast for next Black Friday.
Why does everything depend on your data?
The formula fits on one line. What decides whether it works is the data: the average and the deviation of each size in each store, last year's Black Friday, the real lead time of each delivery. With aggregated data, the buffer is too big where it doesn't sell and too small where it does.
That's where the profit is: selling at full price what you already had and not sending to the sales what you didn't need to buy or ship.
Datcisions is an AI merchandise planning tool for fashion brands. It connects to your Shopify store (Holded and Odoo are coming soon) and analyses the sales and stock of each size in each point of sale.
With Replenishment, it works out the units each store needs from its recent sales and allocates them from your warehouses. Its projection is adjusted for how your sales grew on those dates last year: exactly what changes before Black Friday. Your team adjusts any quantity and confirms it.
If you sell on Shopify, How to plan purchasing and replenishment on Shopify when you sell fashion shows which data to check before each order.
Frequently asked questions about safety stock
Which service level should I use?
King puts the usual targets between 90% and 98%, and 100% is unattainable. Set it by product group: higher for the models that bring the most revenue and sell at full price, lower for secondary styles, where leftover stock ends up in the sales.
What's the difference between safety stock and the reorder point?
The reorder point tells you when to order: when your stock, counting what's already on its way, falls to the forecast sales during the lead time plus the safety stock.
Paula replenishes on fixed days, so she uses a target level instead: every Monday she tops the M in Centre up to 30 units, the 20 of forecast sales plus the 10 of safety stock.
What if the supplier's lead time varies too?
King gives an extended version of the formula that adds the lead time's variability inside the square root: SS = Z × √(L × σ² + (σL × D)²), where σL is the standard deviation of the lead time, in weeks, and D is average weekly sales.
It only holds if sales and lead times vary for different reasons. If not, as on Black Friday, when the same campaign drives up sales and delays deliveries, you add the two buffers together, which gives a bigger buffer. To use it, record the real arrival date of each delivery: lengthening L just in case isn't the same as measuring how much it varies.
Does a 95% service level mean serving 95% of customers?
No. It's the percentage of replenishments in which you don't run out of stock, regardless of how many units are missing when you do. The percentage of demand you do serve is called the fill rate. According to King, with stable sales it tends to be higher than the service level, and with very irregular sales, lower.
Connect your online store and review, size by size and store by store, what to replenish before Black Friday. If you'd rather see it with someone from the team, book a demo.
Sources
- Peter L. King, "Crack the Code: Understanding safety stock and mastering its equations", APICS Magazine, July/August 2011, pp. 33–36. Reading for MIT course 2.810, consulted on 4 November 2026.