Safety stock is a formula, not a feeling

Ask why an article carries a safety stock of 400 units, and the honest answer in most mid-sized companies is some version of: “It has been that way for years.” Usually there was a stockout once, somebody raised the buffer, and the new number became permanent.

This is the safety stock ratchet: after every incident the buffer goes up, and there is no mechanism that ever brings it down. Multiply that across a few thousand articles and a few years, and a remarkable share of working capital is parked in insurance against risks that may no longer exist — while the articles that actually deserve protection often have too little.

What safety stock is actually for

It helps to separate two jobs your inventory does. Cycle stock covers expected demand between two replenishments — that part is simple arithmetic. Safety stock exists for exactly one reason: things vary. Demand swings, suppliers deliver late. If demand were perfectly stable and your supplier perfectly punctual, the correct safety stock would be zero — for every article.

That leads to the most important insight in this whole topic: safety stock should be sized by variability, not by volume. A fast-moving article with stable demand needs surprisingly little buffer. A medium mover with erratic demand needs a lot. Gut feeling reliably gets this backwards, because intuition anchors on how important an article feels, not on how unpredictable it is.

The standard formula, in plain words

The textbook formula is:

Safety stock = z × σ × √(lead time)

Three ingredients:

• z — your service level dial. It converts a management decision (“we want to serve 98% of demand from stock”) into a number: roughly 1.65 for 95%, 2.05 for 98%, 2.33 for 99%.
• σ — how much demand actually swings, measured as the standard deviation of demand per period, from your own history.
• √(lead time) — the longer you are exposed before a replenishment arrives, the more room variability has to hurt you. (Lead time here is in the same period units as σ.)

A worked example: an article sells 100 units per week on average, with a standard deviation of 30 units. The supplier needs four weeks. For a 98% service level: 2.05 × 30 × √4 ≈ 123 units.

Notice what is not in the formula: the average demand of 100 never appears. Two articles selling 100 a week can correctly need wildly different buffers. And notice the price of perfection: the same article at 95% needs 99 units, at 99% it needs 140. The last percent of availability is always the most expensive one — which is why service levels should be chosen consciously, per segment, not inherited.

Mistake 1: one service level for everything

A single company-wide service level quietly overprotects trivial articles and underprotects critical ones. The standard fix is ABC/XYZ segmentation: ABC ranks articles by value, XYZ by demand variability. Then set targets per segment — say 99% for stable A-articles, 90% for erratic C-articles, or make the latter to order entirely.

The important point: this is a management decision about cash versus availability. The formula does not make the decision — it executes it consistently across ten thousand articles, which no gut feeling can.

Mistake 2: ignoring supplier lead-time variability

The simple formula assumes the lead time is fixed. In reality, “four weeks” often means “between three and eight”. There is an extended version of the formula that adds a lead-time variance term, and in many businesses that term dominates — your supplier’s unreliability drives more buffer than your customers’ demand swings.

Two practical consequences. First, measure actual receipt dates against promised dates; most ERPs have this data and nobody looks at it. Second, sometimes the cheapest inventory reduction is not a better formula but a conversation with one supplier about delivery reliability.

Where the formula breaks

The formula assumes demand is roughly bell-shaped around its average. For lumpy, intermittent demand — spare parts, B2B businesses where three customers place large orders a few times a year — that assumption fails, and the formula will mis-size buffers in both directions. A usable rule of thumb: if more than half of your demand periods are zero, use methods built for intermittent demand instead. Knowing where a method stops working is part of using it properly.

The fix, in four steps

1. Segment your articles (ABC/XYZ) from 24 months of demand history.
2. Decide service levels per segment — explicitly, with management, as a cash-versus-availability trade-off.
3. Recalculate safety stocks and reorder points from measured demand variability and measured lead times.
4. Automate the recalculation to run quarterly. Demand patterns drift; numbers set once are wrong within a year.

Takeaway — Pull ten random articles and ask where their safety stock numbers come from. If the answer involves a person who no longer works there, you are very likely carrying buffers that are too high in some corners and dangerously thin in others. The formula is not perfect — but it is consistent, explainable and adjustable. A feeling is none of those things.