AI Pricing Strategy: Test the Contract Against Cost to Serve

An AI pricing strategy needs two linked decisions: what customers pay for, and how the contract behaves when their usage changes. For a U.S. AI company preparing an enterprise offer, finance should test the proposed package against customer value, delivery cost, and cash terms before the quote is approved. This is a management framework, not a market benchmark.

Choose a unit the buyer can recognize

Seats, completed workflows, processed documents, capacity tiers, and usage can each be reasonable customer-facing units. Ask whether the buyer can predict the bill, verify the count, and connect an increase in that count to value received. Define failed or repeated work explicitly. A workflow price is hard to defend if the contract does not say what counts as completed.

The internal cost unit may differ from the invoice unit. OpenAI’s API pricing lists separate input, cached-input, and output token charges. Amazon Bedrock’s pricing describes on-demand and provisioned-capacity options. Both pages were accessed October 28, 2026. They illustrate why provider invoices and capacity commitments must be mapped into a customer-level cost view; they do not establish the right price for any product. The choice of customer unit and review method below is Amafi Advisory practitioner judgment.

Customer-facing unit	Question before approval	Cost driver to keep inside the model
Seat	Can a small number of users generate disproportionate usage?	Attempts and model usage per active user
Completed workflow	Is completion defined, including retries and human review?	Attempts and review time per successful result
Usage or capacity tier	Can the buyer forecast when the next tier applies?	Provider charges, reserved capacity, and peak demand
Platform fee with included usage	What happens when an account exceeds the allowance?	Customer-level usage and service obligations

Use the AI unit economics guide to build a cost bridge from provider invoices and human delivery work before selecting a margin measure.

Run a contract stress test

Start with one proposed customer offer. Write down the annual or monthly fee, payment timing, included unit, overage rule, implementation scope, support promise, and any termination or repricing condition. Then calculate the same customer-level view for expected use and a plausible high-use case. Do not hide unused committed capacity in an average unit cost or assume that all customers have the same retry rate.

The numbers below are invented solely to show the calculation. They are not an Amafi client result, industry benchmark, or target margin.

Monthly case	Customer fee	Model, cloud, and review	Implementation and support	Delivery margin before shared overhead
Expected use	$10,000	$2,400	$1,600	$6,000, or 60%
High use under the same fee	$10,000	$5,200	$2,300	$2,500, or 25%

The management question is whether that change is an intentional commercial choice. The company might accept it for a defined reason, change the allowance, charge for excess use, narrow the service promise, or require approval for the exception. Assign an owner and review date to whichever choice management makes. Reconcile this delivery view to reported gross margin instead of presenting the two as interchangeable.

Make exceptions visible before signing

A pricing review should show the commercial team what can be quoted under standard terms and which changes need a decision. A short approval sheet can record:

1. the customer unit and how it is counted;
2. included usage, overage, and capacity boundaries;
3. implementation and support included in the fee;
4. expected and high-use delivery cost;
5. discounts, prepayment, and collection timing; and
6. the person who can approve a departure from standard terms.

Keep the assumptions beside the quote. A discount may be acceptable if the service obligation is limited; a higher fee may still be weak if it includes open-ended review work. The financial model guide explains how to connect pricing and usage assumptions to cash and operating scenarios.

Recheck the model after the first reporting cycle

Compare the quoted usage and delivery assumptions with observed attempts, successful units, provider charges, review time, support demand, and cash received. Keep the original assumptions so management can see which part changed. If usage rises faster than the billed unit, investigate product behavior and contract design before applying a blanket price increase.

A defined corporate finance project can establish the first cost bridge and approval sheet. A fractional CFO engagement can own the recurring review as contracts, usage, and forecasts change. To discuss a specific pricing decision, describe the offer and current data through the contact form.

Common questions

What is the best pricing model for an AI product?

There is no universal model. Choose a customer-facing unit that the buyer can understand, then test each package against expected and high usage, implementation, support, and cash collection. The contract should say what happens when usage exceeds the included amount.

Should an AI company price per token?

Use tokens as a customer-facing unit only when buyers can understand and manage them. Otherwise, keep token volume and provider charges in the internal cost model and price a unit the customer recognizes, such as a completed workflow or a stated capacity tier.

How should finance evaluate an unlimited-usage contract?

Estimate delivery cost across plausible high-usage cases, including retries, committed capacity, human review, implementation, and support. If the downside is not bounded, define a usage allowance, fair-use condition, capacity limit, or review trigger before approval.

This article is general business information, not accounting, tax, legal, investment, or securities advice. Qualified advisers should review financial-statement classification and contract terms.