The Purpose of Safety Stock Is to Control Stockout Risk — Not Eliminate It

The purpose of safety stock is to control stockout risk, not eliminate it

The textbook multiple-choice version of this question, the one showing up in study guides right now, has four options, and the correct one is "control the likelihood of a stockout due to variable demand and/or lead time." Not eliminate it. Not replace failed units. Not protect against a demand collapse. Control the likelihood.

That distinction matters more than it looks. Safety stock is a probability tool, not a guarantee. You pick a service level — say, a 95% chance of not running out before the next shipment lands — and the formula tells you how many extra units that confidence costs in inventory. Push the target to 99% and the number climbs fast, because you're now covering rarer, more extreme combinations of slow shipments and demand spikes at the same time.

The wrong answers on that exam question are wrong for specific reasons: safety stock doesn't replace defective units, that's a quality and returns problem. It doesn't protect against a sudden demand decrease, that's a forecasting and liquidation problem, arguably the opposite failure mode of a stockout. And it can't eliminate stockout risk entirely, because nothing short of infinite inventory does that. You're always trading carrying cost against service level, never removing the trade.

A worked example: turning the formula into a number

Take a SKU with average daily demand of 40 units and an average supplier lead time of 14 days. If both numbers held steady every cycle, you wouldn't need safety stock at all, just reorder at 560 units (40 × 14) and never run out. The problem is lead time doesn't hold steady, and neither does demand.

Say the lead time has a standard deviation of 2 days, sometimes it's 12, sometimes it's 17, and you want a 95% service level, which corresponds to a Z-score of roughly 1.65. Using a standard formula (Z-score × standard deviation of lead time × average demand):

Safety stock = 1.65 × 2 × 40 = 132 units

Your reorder point becomes 560 + 132 = 692 units. That's the level that triggers the next purchase order, not a number you sit at forever. Push the service level to 99% (Z ≈ 2.33) and the buffer grows to roughly 186 units, a 41% jump in stock for a 4-point gain in service level. That trade-off is the whole point of the formula, and it's the part every dictionary-style definition skips.

Where the math breaks down

Every formula above assumes something. The fixed formula assumes demand and lead time barely move. The average-max formula accounts for lead-time swings but not demand swings. Most calculators default to a normal distribution, which fits a steady seller fine and fits almost nothing about an intermittent SKU, one with long flat stretches and occasional spikes needs a different distribution entirely, or the buffer comes out wrong in both directions at once.

The mistake we see most often, across the more than $500M in managed revenue Full Circle has managed across 100+ brands, is sizing safety stock off average lead time instead of lead-time variance. A supplier that "usually" ships in 14 days looks safe until the one time it ships in 26, and the buffer sized for the average disappears mid-launch, right when a stockout does the most damage to rank.

We've made a version of that mistake ourselves: applying one service-level target across a whole catalog because it's simpler to manage, then finding out the SKU with the volatile supplier and the SKU with the reliable one were never going to need the same buffer. Epic Gardening's seasonal seed kits and HexClad's core cookware line don't carry the same risk profile, even when the same spreadsheet formula gets applied to both.

When your safety stock number turns out to be wrong

If a SKU stocked out anyway, don't just raise the buffer and move on, find out which input was wrong first. Check whether the standard deviation of lead time reflects what actually happened over recent cycles, not what the supplier promises on a contract. Check whether demand had a real trend or seasonal shift the average is smoothing over. And check whether this was a safety stock problem at all, sometimes it's a reorder point that fired late because a human was slow to act on it, which no amount of buffer inventory fixes.

On Amazon specifically, the other place this breaks is the fee side. Oversized safety stock held "just in case" runs straight into long-term storage fees and aged-inventory surcharges, and the removal-versus-liquidation decision gets harder the longer it sits. The formula only tells you the stockout-risk half of the trade; someone still has to weigh that against what the extra units cost to hold.