How to Price AI Products: Cost, Consumption, or Outcome

Your CFO pulls you into a room. Revenue is up and to the right since you shipped the AI feature. Gross margin is down. Every new customer who turns on the AI workflow costs you money to serve, and the per-seat price you set last year does not cover it. This is the pricing problem that traditional SaaS never had to solve, and it is now sitting on your desk.

Software has near-zero marginal cost. One more user costs you nothing. That single fact is why per-seat pricing worked for two decades. AI products break that assumption. Every inference call has a cost. Every generated response, every document processed, every agent action carries a variable expense that scales with usage. If you price AI like traditional SaaS, you are selling a product whose cost to serve grows with adoption while your revenue stays flat. That is a margin trap.

Pricing an AI product is a decision about who absorbs the variable cost and how you capture the value you create. There are three models. Each makes a different tradeoff.

The Margin Problem Comes First

Before you choose a model, instrument your cost to serve. You cannot price what you cannot measure. For every account, you need to know the variable cost of delivering the AI workflow: inference, orchestration, any third-party model calls, and the compute around them.

Do not guess at this. Establish your baseline from your own usage logs and infrastructure billing. Track cost per account, not just cost in aggregate. The account-level view is what exposes the customer who is quietly destroying your margin while looking like a healthy logo in your revenue report.

Once you have cost per account, you can set a gross margin floor and hold every pricing decision against it. This is the number your deal desk defends. It is the number that tells you when a discount has crossed from aggressive into unprofitable.

Three Models, Three Tradeoffs

Cost-Plus and Consumption Pricing

You charge for what the customer consumes. Credits, tokens, generations, agent actions, documents processed. The price is anchored to your variable cost plus a margin.

When it works: Your cost to serve varies widely across customers, and you need every account to carry its own weight. Consumption pricing passes the variable cost through to the buyer, which protects your margin by design. The heavy user pays more because they cost more.

When it breaks: Buyers cannot forecast their spend. A CFO who cannot budget for your product will choose a competitor who offers a predictable number, even if the predictable number is higher. Consumption pricing also caps adoption, because every use of the product is a metered cost the customer feels. That works against the daily usage you need to drive retention.

The forecasting cost: Consumption revenue is harder to forecast and harder to compensate. Your AEs cannot commit a fixed ACV when the ACV depends on how much the customer uses. Your comp plan, your capacity model, and your forecast all inherit that variability.

Outcome-Based Pricing

You charge for the result, not the activity. Per resolved support ticket, per qualified lead, per completed task, per successful action. The price tracks the value delivered rather than the compute consumed.

When it works: The outcome is measurable, attributable to your product, and clearly worth more to the buyer than it costs you to produce. Outcome pricing is the strongest alignment between what the customer pays and what they get. It also reframes the buying conversation away from cost and toward value, which raises your ceiling.

When it breaks: Attribution is contested. If the customer can argue that the outcome would have happened without you, you will fight over every invoice. Outcome pricing also carries a margin risk that consumption pricing does not: if an outcome is cheap for the buyer to trigger but expensive for you to produce, you can lose money on a successful result. You need a tight link between the outcome you charge for and the cost you incur to deliver it.

The measurement burden: Outcome pricing only works if both sides trust the measurement. You need instrumentation that the customer accepts as the source of truth. Build that before you sign the contract, not after the first billing dispute.

Hybrid: Platform Fee Plus Variable

A base platform fee that covers your fixed cost to serve, plus a consumption or outcome component that captures expansion. This is where most AI companies with a real go-to-market motion are landing.

When it works: You want predictability for the buyer and margin protection for yourself. The platform fee sets a floor that guarantees each account covers its baseline cost. The variable component captures upside as the customer scales, without requiring an expansion sales cycle for every dollar of growth.

Example structure: A monthly platform fee that includes a defined allotment of usage, plus metered pricing above the allotment. The base creates a predictable number the CFO can budget. The overage creates automatic expansion as adoption grows. Set the included allotment so that a typical account lands comfortably above your gross margin floor at the base fee alone.

The hybrid model is not a compromise. It is a way to give the buyer the predictability they demand while keeping the margin discipline that AI economics require.

Packaging: Bundle or Meter

Beyond the pricing model, you face an AI-specific packaging decision. Do you bake AI into your existing tiers, or charge for it separately?

Bake it in when the AI feature drives adoption and retention, and when your cost to serve is low enough that including it does not threaten your margin at the tier price. Bundling removes friction and makes the AI a reason to buy the whole product.

Meter it separately when the AI carries meaningful marginal cost, when usage varies widely across accounts, or when the AI delivers value that a flat tier price cannot capture. A separate AI line item, priced on consumption or outcome, lets your heaviest users pay for what they use without forcing that cost onto everyone.

The wrong move is to bundle a high-cost AI feature into a flat tier and hope the average holds. Averages hide the accounts that are underwater. Meter the cost driver. Bundle the adoption driver.

Guardrails Your Deal Desk Needs

AI pricing needs deal desk rules that traditional SaaS never required, because a bad AI deal loses money on every unit of usage rather than just leaving margin on the table.

1. Set a gross margin floor per account and enforce it. No deal closes below the floor without CRO and CFO sign-off. This is not a discount policy. It is a solvency policy.

2. Cap unlimited commitments. Never sell uncapped usage at a flat price. If a customer demands "unlimited," structure it as a high allotment with overage above it, or price the cap into the fee. Uncapped flat pricing on a variable-cost product is how you fund your customer's growth out of your own margin.

3. Monitor cost per account after the sale. Pricing is not done at signature. Instrument ongoing cost to serve per account and flag the accounts that drift below your margin floor as their usage grows. These are renewal conversations you want to have early.

When to Change Your AI Pricing

Three signals that your model needs to evolve:

Gross margin declining as adoption grows. Your price does not track your cost to serve. You need a consumption or outcome component that scales with usage.

Buyers stalling on unpredictable spend. Your consumption model is costing you deals to competitors offering a fixed number. Add a platform fee to create the predictability the CFO needs.

Your best outcomes are your worst margins. An outcome you charge for costs more to produce than you priced it. Re-anchor the outcome price to your delivery cost, or add a consumption floor beneath it.

Start by measuring cost per account this quarter. You cannot price an AI product responsibly until you know what it costs you to serve one. Everything else follows from that number.