What Is Safety Stock? The Formula, and a Worked Example

Safety stock is extra inventory held to absorb the difference between what you forecast and what actually happens. Most definitions stop there. The quantity you actually order depends on a service level you have to choose and on variability you have to measure, so that is what the rest of this covers.

What is safety stock?

Safety stock is the quantity of a SKU you plan to still have on hand at the moment a replenishment order arrives. If everything goes to plan, you never touch it. It exists for the weeks when demand runs hot, or the container clears customs late, or both at once.

It sits underneath the reorder point rather than beside it:

Two consequences follow from that structure, and both matter more than the definition:

• Safety stock is not "extra stock you happen to have." It is a planned quantity with a stated purpose. Inventory sitting in a warehouse because a forecast ran high is overstock, and it buys you nothing.
• Safety stock does not fix a bad forecast. It covers the residual error of a reasonable one. If your forecast comes in low month after month, the answer is the forecast, not a bigger buffer.

What does safety stock actually protect you against?

Two sources of variability, and they compound.

Demand variability

Your weekly sales are a distribution, not a number. A SKU that sells a steady volume every week needs very little buffer. One that sells nothing for a fortnight and then clears a pallet after a creator posts about it needs a great deal more, at exactly the same average.

The measure that matters is the standard deviation of demand over a period matched to your lead time. Averages hide the precise thing you are buying protection against.

Lead-time variability

Most brands never measure this one, and it is usually the bigger term in the formula. A supplier who quotes four weeks and delivers in four weeks every time costs you almost nothing in buffer. A supplier who quotes four weeks and delivers anywhere between three and eight forces you to hold stock against the worst case, every cycle, forever.

Your purchase order history already holds this data: promised date against received date, per supplier, per SKU. If you have never pulled it, that is the highest-value hour of analysis available to you.

How do you calculate safety stock?

Two methods are worth knowing. One is fast and wrong in a predictable direction; the other needs data you may not have yet.

The days-of-cover method

Safety stock = average daily demand × days of cover

Pick a number of days you want to keep selling through if a delivery slips, multiply by average daily sales, and hold that. It takes a minute per SKU and gets a brand out of the worst trouble.

What it does not do is account for variability. The steady SKU and the volatile SKU get the same buffer, so you carry too much of one and too little of the other. It also treats the days of cover as a judgement call rather than as the consequence of a service-level decision, which means nobody can say what the buffer is buying.

The service-level formula

Safety stock = Z × √(LT × σD² + D² × σLT²)

The structure says something useful before you put any numbers into it. The left-hand term inside the root scales with lead time; the right-hand term scales with the square of average demand. On a high-volume SKU with an unreliable supplier the lead-time term dominates everything else, which is why chasing forecast accuracy on those SKUs pays back less than chasing the supplier does.

Several variants of this formula exist, each needing different data, and they are laid out in safety stock formulas compared. The step-by-step version of the calculation is in how to calculate safety stock.

One data-hygiene point changes the answer more than the choice of method does: strip stockout periods out of the history before computing σD. Weeks when you had nothing to sell recorded low demand because there was no stock, not because nobody wanted it. Leave them in and you size the buffer from the very weeks the buffer was supposed to prevent. The same cleaning step is the first thing demand forecasting asks for.

What service level should you pick?

Service level is the probability that you do not stock out during a replenishment cycle. It is a decision about money and it belongs to you, not to the formula. Each step up costs more inventory than the step before it.

Two things about that table are worth saying plainly.

The relationship is not linear. Moving from 90% to 95% costs a modest amount of extra inventory. Moving from 95% to 99% raises the Z factor by more than two fifths again, and the buffer with it. Very high service levels on slow-moving SKUs are where working capital quietly goes.

And the target should differ by SKU. A hero SKU a retailer will charge you back for justifies a different number from a seasonal variant. Setting one service level for the whole catalogue is the same mistake as setting one days-of-cover figure, wearing better mathematics.

How much safety stock does one SKU need?

Say you sell a single flavour of protein bar through DTC and Amazon at an average of 100 units a day. Your co-packer's purchase orders have been landing at an average of 21 days. Cleaned of stockout weeks, the daily demand standard deviation works out at 30 units, and the receipt dates have varied with a standard deviation of 4 days.

Suppose you set a service level of 95% on this SKU, which puts Z at 1.64. Running the two terms through the formula gives roughly 694 units of safety stock — about seven days of cover. The interesting part is the split. The demand term contributes around a tenth of the variance inside the root; the lead-time term contributes the rest. Almost the entire buffer exists because the co-packer is inconsistent, not because the flavour is unpredictable.

For example, getting that co-packer to commit to a firmer receipt window, halving the lead-time standard deviation and nothing else, drops the buffer to a little under 400 units — a cut of roughly two fifths, with no change to demand and no extra cash committed. Buying more inventory is the expensive way to solve this problem, and it is the one most brands reach for first.

How is safety stock different from a reorder point?

Safety stock is a quantity you intend to still be holding. A reorder point is a trigger level that tells you to place the order.

Reorder point = (average daily demand × average lead time) + safety stock

The reorder point is always the larger number, because it has to cover the demand you expect during the lead time as well as the buffer underneath it. When stock on hand crosses the reorder point, you order. If the cycle runs to plan, the delivery lands with the safety stock untouched. If it does not, the safety stock is what you sell from while you wait.

Confusing the two produces a specific and expensive error: setting the reorder point equal to the safety stock. Do that and you place the order at the moment you were supposed to still have a full buffer, so you are short for the entire lead time on every cycle. The mechanics are worked through in how to set reorder points.

When does the formula give the wrong answer?

The service-level formula assumes demand is roughly normally distributed and that demand and lead time vary independently. Both assumptions break in ordinary CPG situations.

• A promotion is running. Promotional demand is not the same distribution as baseline demand, and a buffer sized on blended history covers neither well. Plan the lift separately — see forecast adjustment for promotions.
• The SKU is new. There is no σD to compute. Borrow an analogous product's variability and revisit once you have real weeks, as in forecasting demand for new products.
• Demand is intermittent. A SKU that sells in occasional case-pack lumps to wholesale accounts has a demand distribution the normal curve describes badly, and the formula will understate what you need.
• The product expires. Shelf life caps the buffer whatever the mathematics says. On a short-dated SKU, a high service level and a slow-moving line together produce write-offs rather than protection.
• The supplier's minimum order quantity is larger than the buffer. Then the buffer is not the binding decision and the order size is.

None of these make the formula useless. They mean its output is an input to a judgement, and the judgement is yours.

How often should you recalculate it?

Quarterly is a reasonable floor for a catalogue of any size, and four events should trigger a recalculation before the quarter is up.

• You changed supplier, or your supplier changed factory. The lead-time distribution has been replaced and the old σLT no longer describes anything.
• You added a channel. Retail and wholesale demand arrives in a different shape from DTC, so blended variability changes even when total volume does not. That interaction is covered in multi-channel demand planning.
• You are heading into peak. The cost of a stockout in peak is not the cost of a stockout in March, so the service level you are willing to pay for should not be the same either.
• A SKU's velocity has moved materially. Both terms in the formula depend on the demand level, so a SKU that has doubled needs its buffer recomputed rather than scaled.

The failure mode here is not picking the wrong method. It is computing a buffer once, typing it into a min-max field, and leaving it there for two years while the business changes around it.

Where a planning tool takes this over

Everything above is a per-SKU calculation with two inputs that drift constantly. On a catalogue of thirty SKUs it is a spreadsheet you maintain. On three hundred, across DTC, Amazon, retail and wholesale, it is a job.

That is the point where the calculation moves into software. Planster builds one demand forecast across your channels, nets it against what is on hand, on order and already committed, and turns the result into a ranked list of what to order — with the buffer recomputed from your own current demand and receipt history rather than from a number somebody typed in last year. The reorder recommendations are where that lands.

The honest version: if you have a handful of SKUs and one reliable supplier, a spreadsheet and the days-of-cover method are enough, and you should not buy anything. The calculation starts costing more than it returns somewhere around the point where you can no longer remember which SKUs are at risk.