Guides
How to bill AI products: tokens, credits, seats and outcomes
November 16, 2026
AI products changed usage-based billing in two ways: the unit of value is new (tokens, model calls, agent runs, resolved tickets) and the cost of serving each unit is high and variable. Billing has to follow both.
The models in use
- Token billing. Input and output tokens priced per thousand or per million, often by model.
- Credits. A single currency covering several models and features, sold in prepaid packs. Credits hide model prices from the buyer and let the vendor change them.
- Seats plus usage. A seat price that includes an allowance, with usage billed above it.
- Outcome-based pricing. A price per result: a ticket resolved, a document processed, a meeting booked.
What the billing system needs
Real-time metering, because usage can spike in an hour. Prepaid balances with alerts and hard caps, because an unpaid spike is a direct loss when model costs are passed through. Flexible aggregation, to price per model or per feature. And, increasingly, cost and margin tracking: the model cost of each customer next to the revenue it brings.
Finance still applies
Credits sold up front are deferred revenue, and outcome-based contracts raise the question of when the outcome is delivered. AI companies selling to enterprises meet the same revenue recognition and audit requirements as any SaaS company.
Platforms built for it
Our page on usage-based billing software for AI companies lists the platforms that serve AI companies and document token billing, with their levels on credits, real-time visibility, alerts and margin tracking.
Was this guide useful?
Get the next guide and every ranking update by email.
Ranking updates only. Unsubscribe at any time. See our privacy policy.