Consumption-based pricing: how usage-based billing works

Usage-based pricing means the bill depends on how much a customer uses. Pricing consultancy Simon-Kucher defines consumption pricing as a model where customers are charged based on their actual usage of a product or service. Consumption-based, usage-based and pay-as-you-go pricing are, in practice, the same idea.

You already know it from utilities: you pay for the electricity you use. In software, the unit is usually something the customer can see and count. The AI model providers themselves price this way: OpenAI and Anthropic charge per million tokens of text in and out.

There are five common ways to charge for usage. Stripe’s pricing-model docs list flat rate, per-seat, tiered and usage-based models, and name fixed fee and overage, pay as you go and credit burndown as the usage-based options.

The prices in the examples are illustrations, not recommendations. Set yours from your own cost per use and what your audience will pay.

Because the cost of an AI app grows with every request, not with every customer. A normal app costs about the same to run whether a user logs in once or a hundred times. An AI app pays the model provider each time.

Take a chat reply that sends 2,000 tokens and receives 500. As of October 2026, that costs about $0.009 on Claude Sonnet 5.5 ($2 in, $10 out per million tokens) and about $0.0015 on OpenAI’s gpt-5-mini ($0.25 in, $2.00 out), by our calculation from the official prices. A user sending 900 replies a month costs about $8.10 on the first model. On a flat $9 plan with unlimited use, that customer is barely profitable; one who sends twice as many messages loses you money.

Usage pricing fixes this by tying revenue to cost. Light users pay little and stay; heavy users pay for what they use. We go deeper on AI costs in what is an AI wrapper.

It gets more complex when one app mixes features. A chat reply, a generated image and a short video can cost very different amounts. Either give each feature its own unit, or use credits where each action spends a different number of credits, and tell customers the cost before they press the button.

Stripe bills usage through meters: you send it each customer’s usage, and it adds it up and invoices at the end of the billing period. Stripe’s docs describe four parts: ingestion, product catalog, billing and monitoring.

1. 1

   Create a meter

   A meter tracks one kind of usage, such as images generated, and how to add it up: sum the values, count the events, or take the last value.

2. 2

   Attach a usage-based price

   Create a product with a recurring, usage-based price linked to the meter, for example per unit or tiered, billed monthly.

3. 3

   Subscribe the customer

   Put the customer on a subscription with that price. You can mix it with a flat monthly price for a hybrid plan.

4. 4

   Send meter events

   Each time the customer uses the feature, your app sends a meter event with the event name, the customer and a number. You can also add usage by hand or by CSV in the Dashboard.

5. 5

   Invoice and monitor

   At the end of the period Stripe rates the usage and creates the invoice. You can set alerts for customers who cross usage thresholds.

Two details matter for AI apps. First, Stripe supports prepaid billing credits through credit grants, but in its basic usage-based billing they are applied when the invoice is created, so customers can go past their balance during the month. Second, Stripe now points new integrations to Metronome, a Stripe product, for real-time metering and real-time credit burndown. As of October 2026, Stripe Billing costs 0.7% of billing volume on pay as you go, on top of card processing. If you would rather not wire meters yourself, an AI app builder with billing built in can do it.

For most consumer AI apps, a subscription that includes a monthly credit allowance, with top-up packs for heavy users, is the safest default. It gives you predictable revenue and caps your cost per customer.

Skip pure usage pricing when your cost per use is tiny and steady, or when your buyers need one fixed number to approve. A simple flat plan is easier to sell and to understand.

• Your AI use per customer is low and predictable, such as one summary a week.
• You sell to businesses whose budgets need a fixed monthly cost.
• Your audience is new to paying for apps and would be put off by meters.
• You cannot yet show customers their usage clearly inside the app.

Even then, keep a fair-use limit in your terms and watch your heaviest users. Moving from flat to hybrid pricing later is easier if the limit exists from day one.

The two big ones are bill shock for your customers and margin leaks for you. Simon-Kucher notes that customers like predictability and that unpredictable bills can cause frustration and churn, and that revenue drops when usage drops.

• Bill shock: warn customers before they run low or cross a limit, and let them set a cap.
• Margin leaks: track your AI cost per customer, not just your total bill. A few heavy users can wipe out the profit from everyone else.
• Confusing units: “1 credit = 1 image” beats “credits based on tokens”. Name the unit in words your audience uses.
• Model price changes: your supplier’s prices move. Leave room in your margin and review prices when models change.
• Late billing: if usage is billed at the end of the month, a customer whose card fails may already have used it. Prepaid credits avoid that.

More ways to charge are in our guide on how to monetize an app, and our guide on how to build an AI app covers pricing at launch.