Locus product guide · Updated
AI billing: how to charge customers for AI usage
To charge for AI usage, find your cost per unit of work (tokens, tool calls or finished tasks), pick a billing unit your customers can count, and add a markup that still covers failed runs and payment fees. Then collect up front with prepaid credits, or invoice usage in arrears with spend limits. A task that costs you $0.166 and sells at cost plus 50% leaves $0.072 of margin after card fees on $20 top-ups.
What are the ways to charge customers for AI usage?
There are five common methods. They differ in when you get paid and in who carries the cost when a customer uses more than expected. The examples come from vendor pricing pages. Prices checked November 10, 2026.
| Method | How the customer pays | Who carries cost risk | Published example |
|---|---|---|---|
| Prepaid credits | Buys a balance first; each use draws it down | The customer, since usage stops at zero | Lovable sells top-up credits that stay valid for 12 months |
| Subscription with included usage | A monthly fee covers an allowance; extra use is paid separately | You, inside the allowance | GitHub Copilot Pro is $10 a month with $15 of AI credits |
| Pay as you go in arrears | Usage is metered and invoiced at the end of the period | You, until the invoice is paid | Stripe basic usage-based billing invoices metered usage at period end |
| Per outcome | Pays only when the AI finishes a defined result | You, for every attempt that fails | Intercom charges $0.99 per Fin outcome |
| Seat plus usage | A per-user fee plus metered AI usage | Split by contract | Claude Enterprise is $20 a seat plus usage at API rates |
How do I work out what one unit of AI work costs me?
Add up every paid step in one unit of work. For an AI feature that is usually model tokens, paid tool or data calls, and the share of runs that fail and retry. Billing and card fees come off revenue later.
Example: a research task sends 20,000 input tokens and gets 2,000 output tokens back from Claude Sonnet 5.5. At $2 and $10 per million tokens, that is $0.04 plus $0.02. Three Brave web searches through Locus Pro add $0.0909, at 30.3 credits each. The task costs $0.1509. If 1 run in 10 fails and is retried, budget about $0.166 per finished task.
- Model tokens: input, output and cached input have separate prices. A Claude Sonnet 5.5 cache hit costs $0.10 per million tokens, against $2 for fresh input.
- Built-in tools: web search on the Claude API costs $10 per 1,000 searches plus tokens. OpenAI lists web search at the same $10 per 1,000 calls.
- Paid data APIs: search, scraping, enrichment and speech each have their own price per call.
- Failures: a run that times out still used tokens, unless your provider releases the charge.
How much markup do I need after fees?
Card fees take a fixed slice of every payment, so small payments cost more to collect. Stripe charges 2.9% plus 30 cents per successful card charge. On a $20 top-up that is $0.88, or 4.4% of revenue. On a $100 top-up it is $3.20, or 3.2%.
The table applies three markups to the $0.166 task above and subtracts a 4.4% card fee. Billing software adds its own fee on top: Stripe Billing takes 0.7% of billing volume on pay-as-you-go.
A 20% markup leaves little room for refunds, disputes and free trial usage. A flat fee per call or per task protects margin on very cheap calls.
| Markup on cost | Price per task | Margin before fees | Margin after a 4.4% card fee |
|---|---|---|---|
| 20% | $0.199 | $0.033 (16.7%) | $0.024 (12.3%) |
| 50% | $0.249 | $0.083 (33.3%) | $0.072 (28.9%) |
| 100% | $0.332 | $0.166 (50.0%) | $0.151 (45.6%) |
Should customers prepay or pay after they use AI?
Ask self-serve and new customers to prepay. Usage stops when the balance runs out, so an unpaid bill cannot grow. You also hold the cash before your providers bill you.
Invoice in arrears for enterprise contracts that need purchase orders and monthly invoices. Then you need a meter, a credit limit and alerts.
Check how your billing tool enforces a balance. Stripe's basic billing credits apply only when an invoice finalizes, so a customer can spend past the balance during the cycle. Stripe's Metronome product burns credits down in real time.
How do I bill customers for AI tool calls with Locus Pro?
Locus Pro Enterprise gives each of your customers a credit balance. Your server names the customer on each call with the X-Locus-End-User header. Locus checks that balance before the call runs and returns HTTP 402 if it is short. Otherwise it charges the catalog base price plus your markup, and the markup is your margin. The ledger and customer analytics group every charge by customer.
Markup can be a percentage, up to 100% by default, a flat amount per call, or both. Example from the Locus catalog on November 10, 2026: Brave web search has a base price of 30.3 credits, or $0.0303. With 20% plus $0.002 flat, the customer pays $0.03836 and you keep $0.00806.
Customers fund their balance by card through Locus-hosted checkout or an in-app widget. Or you bill them your own way and move credits to them from your pool. Enterprise costs $200 a month with $30 of included credits and up to 240 tool calls a minute.
How do I choose a way to bill AI usage?
Match the billing unit to what the customer values and to how your cost moves. Then pick the tool that enforces it.
- AI cost per user is small and even: a seat price with a fair-use limit is simplest.
- Cost varies a lot between customers: prepaid credits, or a subscription with included usage and overage.
- The AI finishes a result the buyer already counts, such as a resolved ticket: price per outcome, high enough to cover failed attempts.
- You sell only your own product's usage: a billing engine such as Stripe Billing, Metronome or Orb is enough. Locus Pro bills only calls made through its catalog.
- Your customers' agents call third-party paid APIs: Locus Pro Enterprise supplies the APIs and bills each customer at base price plus your markup.
Frequently asked questions
What is AI billing?
AI billing is how a product charges customers for the AI work it runs for them, such as model tokens, tool calls or finished tasks. It covers metering each unit, pricing it, collecting payment and showing customers what they used.
Should I show customers tokens or credits?
Show a unit they can predict. Few buyers can estimate tokens. Credits with a fixed dollar value, or a price per task, are easier to budget. Keep token counts in usage reports for technical buyers.
How do I stop one customer from running up a large AI bill?
Use prepaid balances, a monthly spend limit per customer, and a ceiling per call or per task. Locus Pro supports a budget per customer and a maximum charge per call, both checked before the call runs.
Do I need Stripe to bill for AI usage?
You need some way to take payment. Stripe Billing can meter and invoice usage. Locus Pro takes card top-ups through Stripe-hosted checkout and charges each call itself, so calls made through it need no meter events.
Implementation references
Use these first-party references for current request contracts and account requirements. Tool availability, prices, and negotiated terms can change.
Sources
Third-party details were checked on November 10, 2026. Vendors change pricing and features often, so confirm on their site before you decide.