Skip to main content

Trends, tech news, and tool comparisons.

Usage-based vs seat-based pricing for AI products

· 6 min read
Dealpuffin
Proof points and live coupons for early adopters

Use seat-based pricing when each person gets their own value and your cost per user is stable. Use usage-based pricing when your cost rises with every request, which is usually true of AI features. Many AI products land on a hybrid: a seat price that includes a pool of usage, with paid overage after that. Intercom and GitHub both do versions of this, and their pricing pages show how.

Facts below come from vendor pages read on 10 November 2026. Prices change, so confirm before copying a number.

Why AI products strain seat pricing​

A normal SaaS feature costs you almost nothing extra when a user clicks it twice as often. An AI feature does: every call has a model cost. Under flat seat pricing, one heavy user can cost you more than the seat brings in, while a light user subsidises them. That is the core tension.

The opposite risk is real too. Pure usage pricing makes bills unpredictable, and buyers dislike not knowing what they will owe next month.

The three models​

ModelYou charge forWorks well whenMain risk
Seat-basedEach named user per monthValue is per person; usage varies littleHeavy users erode margin
Usage-basedUnits consumed (requests, tokens, credits)Cost scales with use; buyers think in unitsBill shock; metering work
Outcome-basedA successful resultYou can define and verify a resultDisputes over what counts

Example 1: Intercom's Fin charges per outcome​

On its pricing page, Fin AI Agent is priced at $0.99 per outcome. A billable outcome is a resolution (no further help requested), a completed procedure handoff, a disqualification or a self-serve routing, and the page lists qualifications at $9.99. Intercom's helpdesk separately charges per seat, listed from $19 per helpdesk seat per month, so the company combines both models: seats for the humans, outcomes for the AI.

The lesson is that you can price the AI part on the thing the buyer cares about while keeping seat pricing for the part that behaves like normal software. It only works because "resolved" is something you can define and report on.

Example 2: GitHub Copilot sells seats with a credit pool​

GitHub Copilot's plans list Pro at $10 per user per month and Pro+ at $39, each with a monthly allowance of GitHub AI Credits (the page lists $15 and $70 of credits respectively). One credit equals $0.01, and when you hit the limit you can wait for the reset, set a dollar budget for extra usage, or switch to a cheaper model.

This is the hybrid most AI tools settle on: a predictable seat price, a usage allowance sized so the typical user never notices it, and metered overage for the heavy tail.

A decision rule​

Ask three questions in order.

  1. Does your marginal cost per use matter? If one request costs you real money, you need some usage component. If it costs you fractions of a cent, a flat seat may be fine.
  2. Who is the buyer, and what do they budget for? Teams buying for employees expect per-seat prices they can multiply by headcount. Teams buying automation expect to pay per result.
  3. Can you measure the unit honestly? If you cannot show a customer exactly what they were charged for, do not use usage pricing yet.

If the answers are "yes, yes, no", start with seats and a fair-use limit, then add metering when you can show it. The earlier post on pricing a SaaS product at the early stage covers the value-metric choice in more depth.

Worked example​

Say your AI feature costs you about $0.02 per run and you charge $20 per seat per month.

  • A light user runs it 100 times: $2 of cost, $18 margin before fees.
  • A heavy user runs it 2,000 times: $40 of cost, a $20 loss on that seat.

A hybrid fixes this without scaring buyers. Include 500 runs per seat, which covers the typical user, and charge overage per extra 100 runs. Most customers never see an overage, and the heavy user pays for what they use. The numbers here are my own illustration, not a benchmark.

What you need to build​

Seat pricing needs a subscription and a way to count active users. Usage pricing needs meters that record events, a way to aggregate them each billing period, and an invoice that explains them. Stripe documents this in its basic usage-based billing guide, and notes that Metronome, now part of Stripe, is its recommended starting point for new usage-based integrations. For tooling options after that acquisition, see usage-based billing after Metronome and Orb.

Add spending alerts and limits from the start. Intercom and GitHub both let customers set budgets, which is what makes unpredictable bills acceptable.

Who should pick what​

  • Solo tool with light AI use: seats, with a stated fair-use limit.
  • AI is the product and cost scales with use: usage or credits, with a monthly minimum so revenue is predictable.
  • AI replaces a measurable task: consider outcome pricing, but only if you can verify the outcome.
  • Unsure: hybrid, with seats plus an included allowance.

Frequently asked questions​

Should an AI SaaS charge per seat or per usage?​

Charge per seat if value and cost are per person and stable. Charge per usage if each request costs you money and customers think in units. A hybrid of seat plus included usage is the most common answer.

What is outcome-based pricing?​

It means charging only when the product delivers a defined result. Intercom lists $0.99 per Fin outcome, where an outcome is a resolution, procedure handoff, disqualification or self-serve routing.

How do I avoid bill shock with usage pricing?​

Include an allowance in the base price, show usage in the product, and let customers set alerts and spending limits. GitHub Copilot and Intercom both offer budget controls.

Do I need special billing software for usage-based pricing?​

You need metering and invoicing that handle events, which Stripe supports through its meters API and Metronome. A simple product can start with seats and add metering later.