Guides
Why Lead-Time Variability, Not Average Lead Time, Sizes Your Buffer
Two suppliers can share the exact same average lead time and still need wildly different safety stock — because a buffer exists to cover the spread, and the average throws the spread away.
The average lead time lies to you
Ask a supplier how long shipping takes and you'll get a single number: fourteen days. It sounds like an answer. It isn't. Picture two suppliers who both average fourteen days. Supplier A lands almost every shipment within a day of the promise — thirteen days, fourteen, fifteen. Supplier B swings from eight days to twenty depending on the season, the port, and the mood of the freight market. Same average. Wildly different risk. If you plan both around "fourteen days," you will hold too little stock against Supplier B and get stocked out on the exact deliveries that arrive late.
The reason is baked into what an average is. The mean collapses a whole distribution of outcomes into its center point and discards everything about how far outcomes stray from it. But a safety buffer's entire job is to cover the strays — the late deliveries, not the on-time ones. So the one statistic you reach for first is the one statistic that contains none of the information the buffer is built to absorb. The average tells you where to expect the shipment. It says nothing about how wrong that expectation can be.
| Metric | Supplier A | Supplier B |
|---|---|---|
| Average lead time | 14 days | 14 days |
| Spread / consistency | ±1 day (illustrative) | ±6 days (illustrative) |
| Safety stock needed | Small | Large |
| What the average tells you | Nothing about the risk — reads "14" | Nothing about the risk — reads "14" |
Two suppliers, same average, different buffer (illustrative).
A buffer is sized by spread, not by the middle
Inventory splits cleanly into two jobs. Cycle stock covers expected demand over the expected lead time — the stock you'd hold even in a perfectly predictable world. Safety stock covers the gap between expected and worst-case: the demand that runs hot while a shipment runs late. The mean lead time sets your cycle stock. Only the variability sets your safety stock.
That's why the classic safety-stock formulas are built entirely out of standard-deviation terms:
SS = Z × σ_D × √L
The buffer scales with the spread of demand (σ_D), not its average. The Z is a service-level dial: 1.28 buys roughly 90% service, 1.65 buys 95%, 1.96 buys 97.5%, 2.33 buys 99%. Turn the dial up and you demand a fatter buffer to catch rarer bad outcomes. But notice what's absent from the term that grows the buffer: the mean. Move the average lead time and you move cycle stock, which you were holding anyway. Move the variability and you move the safety stock — the number this whole exercise is trying to size.
Why variable ship times punish you twice
Once lead time itself is uncertain, the math gets sharper — and less forgiving. The combined formula (the King equation) is:
SS = Z × √( σ_D²·L + μ_D²·σ_L² )
Two sources of risk live under the root — demand spread (σ_D²·L) and lead-time spread (μ_D²·σ_L²). The lead-time term is scaled by demand squared (μ_D²), so a faster-selling product pays far more buffer for the same erratic supplier. (Generic inventory theory.)
Read that second term slowly, because it's where variable ship times punish you twice. σ_L is the spread in your supplier's ship times. But it doesn't act alone — it's multiplied by μ_D², the square of your average demand. The intuition is plainer than the algebra: when a shipment is late, every extra day of lateness has to be covered by however fast the product sells. A slow mover barely notices a late truck; a fast mover burns through the shelf. So the same unreliable supplier costs a trickle of buffer on a slow SKU and a mountain of buffer on your best seller. Erratic ship times and high velocity compound — that's the double punishment.
The upside of understanding the mechanism is knowing which lever moves the number. Because σ_L sits inside the buffer term, shrinking your supplier's variability — a more consistent freight lane, a second source, a firmer commit — cuts the required buffer without touching your service level. You get the same protection for less capital. Chasing a faster average, by contrast, mostly moves cycle stock. The reliability of the ship time, not its speed, is the expensive variable.
Feeding a variance-aware buffer: forecast the demand side cleanly
You don't control your supplier's freight variance directly — but you fully control the demand half of that equation, and feeding it a clean signal is where the buffer math stops being theoretical. Both risk terms lean on demand: σ_D is how much your sales bounce around, and μ_D scales the lead-time penalty. Garbage in either one and the buffer is sized against a fiction.
This is the seam Assisty is built for. Assisty forecasts demand at the product and variant grain over a horizon you choose — so a size or color that moves fast gets its own number instead of hiding inside a parent average. That forecast reflects real demand, not stockout days: past out-of-stock days can be excluded, so a period where you couldn't sell doesn't quietly deflate your μ_D. The forecast flags items that will run out before the period ends — and those at-risk items carry straight into the Smart Buying Table, so spotting the risk and placing the order are one motion, not two tools. When a number looks off you can question it: Atlas explains the drivers behind a demand change, and what-if scenarios recompute the forecast and reorder needs instantly. The statistics above tell you why variability sizes the buffer; a clean, variant-level demand signal is what keeps the demand half of that math honest.