Nobody decides their service level on purpose. It gets decided anyway, by feel, in the size of the pile you keep "just in case." A brewery keeps an extra pallet of cans because running out mid-canning-day is a disaster. A hardware store keeps three extra boxes of the fastener contractors buy on the way to a job. A vet clinic keeps more of one injectable than it will use in a month, because the day it's out is not a day anyone wants to explain.
That pile has a name, safety stock, and a price. What almost nobody knows is that the price does not rise evenly. The first stretch of certainty is cheap. The last stretch is where the money goes, and most owners are buying at the expensive end without ever having chosen to.
What the buffer is actually for
Your reorder math can cover the average. If you sell 50 a week and delivery takes two weeks, you order when the shelf hits 100, and in an average world that works forever. The buffer exists because no week is average. Sell 65 while you wait, twice in a row, and the average math leaves you apologizing.
So the buffer is insurance, and the thing it insures is the gap between an ordinary wait and a bad one. Its size follows from three numbers. How much your demand swings, how long you wait for delivery, and how sure you want to be. The first two are facts about your business. The third is a choice, and it is the one doing most of the pricing.
The cost of never running out
Say being covered 19 waits out of 20 takes a buffer of 80 units. That is 95%, and for a business that orders monthly it works out to one scramble every year or two. Moving to 99% pushes the scramble out toward once a decade, and the price of the move is about 40% more buffer, another 32 units sitting there through every ordinary week, earning nothing, to cover the one week that mostly never comes.
The reason is the shape of rare events. Covering the common swings is cheap because they are common and small. Each further step of certainty has to cover a rarer, wilder week, so each step costs more stock than the one before it. Perfect certainty is not on the menu at any price. Past some point you are no longer buying protection. You are storing anxiety in case form.
None of this says 99% is wrong. For the vet's injectable, it is probably right, and for a berry supplier's flats in June it is probably absurd. The point is that it is a purchase, with a price, and it deserves the ten seconds of thought any other purchase gets. What does a stockout of this item cost me, honestly? A lost sale, a substitution, an apology? Or a canceled procedure? Those are different answers, and they justify different piles.
The strange arithmetic of waiting longer
Here is the result that surprises everyone. If your supplier's lead time doubles, the buffer you need grows by about 1.4 times, well short of doubling with it.
The mechanism is worth having in plain terms, because it changes how you feel about long waits. Over a longer wait, you get more weeks of demand, and they don't conspire against you. A hot week lands next to a quiet one and they partially cancel. The total across four weeks wobbles less, proportionally, than any single week does, so the buffer that covers the wobble grows slower than the wait itself. Statisticians would say the variability grows with the square root of the time. You can just remember that waits stack, but their surprises partly cancel.
Two practical things fall out. A longer lead time is less costly to insure than it feels, so a cheaper supplier who takes four weeks instead of two is not asking you to double your buffer, only to grow it by 40% or so. And the reverse holds. Cutting your lead time in half does not halve the buffer either. If you are paying for speed mainly to shrink the pile, that trade is smaller than it looks.
What actually moves the number
Notice what the buffer math never asks about: how much you sell. It asks how much your demand swings. A steady 200 a week needs almost no buffer at all. An average of 50 that lands anywhere between 10 and 120, the App State home-game pattern half the businesses in Boone live with, needs a serious one. Volume is not risk. Volatility is the thing being insured, which is why the same arithmetic hands a food truck a bigger buffer than a dental supply shelf ten times its volume.
That also means the cheapest way to shrink a buffer is rarely to accept more stockouts. It is to shrink the swing or the wait: a supplier who says Tuesday and means Tuesday, a pre-order habit for the game-weekend spike, a second source for the one item whose lead time is the problem. The buffer is the receipt for the variability you have decided to live with. Change the variability and the receipt rewrites itself.
If you want your own numbers, the calculator prices the dial for you: your swing, your wait, and the certainty you pick, turned into units on a shelf. Try 95 and 99 and look at the gap between them. Then put a cost on the stockout you are protecting against, because that number is not in any formula, and it is the one that decides whether the last few points of sureness are insurance or anxiety.