ACU: paying for AI without a surprise bill

The standard way to bill for AI features is to meter tokens and invoice monthly. It is also the way customers end up with a bill several times what they expected, from a process they did not realise was still running.

One unit, one price

An ACU is one credit at a fixed price. Every AI action carries a stated cost in ACU: a full event build is 35–45, regenerating one section is a fraction of that. You buy credits and spend them, and the price of a credit does not move because a model provider changed their rate card.

Quoted before it runs

The cost appears before you start, not after. A quote that arrives afterwards is a bill, and a bill you could not have declined is not a price.

Zero is a hard stop

When the balance reaches zero, AI actions stop. There is no overdraft, no auto-top-up you did not ask for, and no negative balance to settle later. Everything that is not AI keeps working — selling tickets, scanning at the door, taking payment — because those are the product and AI is an accelerant on top of it.

Why a credit rather than a token count

Tokens are a supplier's unit, not a customer's. They vary by model, they are impossible to estimate in advance, and pricing in them means every model change is a price change you have to explain. A credit is stable and comparable across features, which is what makes a budget possible.

The ledger shows every spend

Each deduction is a line: what ran, when, what it cost, and against which event. If a number looks wrong you can find the specific action that produced it rather than arguing with a monthly total.