AI products are a natural fit for usage-based pricing: tokens, requests, documents processed, minutes of audio. On AWS Marketplace, usage pricing means your product tells AWS how much each customer used, and AWS bills them. If those reports are wrong or late, you either overcharge customers or lose money.
1. Choose dimensions customers understand
A pricing dimension is the unit you charge for. Good dimensions are:
- Easy to explain — "per 1,000 requests" beats "per compute unit".
- Easy to measure from data you already log.
- Few in number. Two or three dimensions are plenty for most products.
For AI products, think about whether customers care about tokens at all. Many prefer a business unit like "documents" or "conversations", with token costs handled inside your pricing.
2. Design the metering pipeline
A reliable setup has four parts:
- Usage events written by your product every time billable work happens.
- An aggregation job that totals usage per customer, per dimension, per time window (hourly is a common choice).
- A sender that reports the totals to AWS using the AWS Marketplace Metering Service (the
BatchMeterUsageAPI for SaaS). - A ledger that records what was sent, what AWS accepted and what needs retrying.
3. Rules that catch teams out
- Send promptly. AWS rejects records that are too old — usage generally has to be reported within a short window of when it happened, with a small grace period at month end. Don't batch usage up for days.
- Small batches. Each call accepts a limited number of records and covers one product. Split large volumes into several calls.
- Retries are safe — if you're consistent. Sending the exact same record again is fine. Sending a record for the same customer, dimension and hour with a different quantity is rejected as a duplicate, so fix totals before you send, not after.
- Handle partial failures. The response tells you which records were processed and which need resending. Your ledger should retry only those.
- Use the current customer identifiers. AWS has been moving new SaaS integrations to account- and licence-based identifiers, so check the current requirement before you build.
4. Monitor it like billing — because it is
- Alert when a metering run fails or sends nothing for a customer who was active.
- Reconcile monthly: usage you recorded vs. usage AWS billed.
- Keep raw usage events long enough to answer customer billing questions.
Checklist before launch
- Dimensions agreed with sales and finance
- Aggregation tested with real traffic
- Retries, partial failures and month-end tested
- Alerts and a monthly reconciliation report in place
Our Listing Launch package includes building and testing this pipeline with your engineers.