In August I argued that per-order pricing punishes your best months, because the month you close the most is the month your software bill hurts the most. The standard answer from vendors, and I include us in that, is tiered pricing: pay a fixed amount for a band of volume, move bands as you grow. Tiers can fix the problem. They can also reproduce it with extra steps, and from the outside the two look identical on a pricing sheet. We are in the middle of building volume tiers for Autopilot right now, so I have been staring at this harder than usual. Here is what a tier should actually buy you, and the five places a tier structure goes wrong.

The Point of a Tier Is Predictability

Start with what you are trying to get. Not the lowest unit price; that is what per-order already optimizes, and it is why per-order hurts. What you want is a bill that moves slower than your revenue. A busy month should feel like a busy month, not like a penalty. A slow month should not strand you paying for capacity you didn’t use. Everything below is a test of whether a given tier design delivers that or quietly takes it back.

Test One: Where the Breakpoints Sit Relative to Your Months

Vendors set tier boundaries at round numbers because round numbers look clean on a sheet. Your volume does not arrive in round numbers. Before you evaluate any tier structure, pull your last twenty-four months of order counts and find three figures: your median month, your 75th-percentile month, and your 90th-percentile month. Then lay the vendor’s breakpoints over them.

The failure case is a breakpoint sitting just below your 75th-percentile month. That means roughly one month in four, you are paying the next tier up for a handful of orders over the line. Over a year, that is three or four months at the higher rate to cover maybe 10% more volume. That is per-order pricing in disguise; it has just moved the pain from every order to every busy month. A tier structure built for you has its breakpoints comfortably above the busy months you actually have, not the median month the salesperson asked about.

Test Two: Which Way the Ratchet Turns

Read the clause about moving between tiers, because it is rarely symmetric. The common design moves you up automatically the month you cross a line and moves you down only on request, or only at annual review, or only with notice. Each of those is defensible on its own. Together they are a ratchet, and a ratchet is the per-order penalty in slow motion: every surge is permanent, every lull is temporary. Ask the vendor to say in writing how and when you move down, and whether it happens without you having to catch it. If the answer involves you noticing and asking, assume you won’t, because nobody does during a busy quarter.

Test Three: Overage Is a Price or a Penalty

Most tiers include some overage mechanism for orders beyond the band. The question is the rate. Divide your tier fee by the orders it covers to get your effective unit cost. If the overage rate is at or below that number, overage is just a price: you pay for what you use, at roughly what you were already paying. If the overage rate is above it, sometimes well above, the tier is a trap at the margin. Your busiest months are being charged more per order than your ordinary ones, which is exactly the thing tiers were supposed to end. A vendor who prices overage as a penalty is telling you they expect to make money on your surges, and you should believe them.

Test Four: What Is Inside the Band and What Is Metered

A tier defined on orders can still meter everything else. Summaries, document storage, user seats, API calls, support tiers. None of that is illegitimate, but it changes the arithmetic, because the predictable fixed fee turns out to cover the part of the product that doesn’t scale while the part that does is still billed per unit. Ask for a list of every metered item and run your realistic month through all of them, not just the headline. The June 16 post on reading an AI vendor’s pricing sheet made the same point from the other direction: the number on the front page is rarely the number you pay.

Test Five: The Downgrade Clause When the Cycle Turns

This one nobody reads until it matters. Volume in this industry is cyclical in a way most software customers’ volume is not. A refi wave ends and order counts fall by a third in a quarter. If your contract locks you in a tier for the term, you are paying for 2026 volume through a 2027 trough. The May 26 post on building a stack that survives the next refi cycle argued for architecture that flexes; the contract has to flex with it. Find out whether you can step down a tier mid-term, how much notice it takes, and whether there is a floor below which you can’t go. A tier structure with no graceful path down is a bet on volume that the vendor has placed with your money.

Including Ours

I said we are building Autopilot tiers now, so I’ll say what we are holding ourselves to, and you can hold us to it. Breakpoints wide enough that a normal busy month doesn’t bump you. Overage priced at the tier’s unit rate, not above it. A path down that doesn’t require you to notice and ask. And a list of what is metered that fits on one line, because if it doesn’t, the tier is hiding something. If the sheet you end up seeing from us fails any of those tests, I would rather hear about it than have you quietly conclude we didn’t mean it.

The broader point stands for any vendor. A tier is a promise about the shape of your bill. Before you sign one, make the vendor draw that shape over your real months, including the three best ones, and see whether it still looks like predictability.

See how Autopilot pricing is structured →