How is AI changing software pricing?

AI agents do work that people used to do, so customers need fewer seats. Vendors are adding usage, credit, hybrid and outcome-based models that charge for work done.

What is outcome-based pricing?

The customer pays only when a defined result is achieved, such as a support ticket resolved by an AI agent without human help. It needs a precise, agreed definition of the outcome.

Is per-seat pricing dead?

No. Seats still fit products where people do the judgement work. Bain found in October 2025 that most software vendors adding AI kept seats and added an AI meter on top, a hybrid model, rather than replacing seats entirely.

What is the difference between usage-based and credit-based pricing?

Usage-based pricing bills for each unit consumed after the fact. Credit-based pricing sells a prepaid pool of credits that the customer spends across actions, which gives the buyer a fixed budget and the vendor upfront cash.

Why does inference cost matter for software pricing?

Inference cost is the cost of running an AI model each time it acts. It rises with use, so a flat price on heavy AI use can reduce gross margin.

How should a scale-up test a new pricing model?

Define the value unit, model the cost to serve at heavy use and confirm the unit can be measured. Then offer the new model to some new customers, keep existing contracts unchanged and compare win rate, deal size and margin.

How do investors look at usage-based revenue?

Investors usually separate committed revenue from variable usage revenue and check gross margin after AI inference costs to judge how recurring and profitable it is.

Insights / AI and software pricing

By Niraj GemawatPublished 3 Nov 2026

From seats to outcomes - how AI is changing software pricing

AI is moving software pricing away from charging per user and towards charging for the work the software does. When an AI agent resolves a support ticket or completes a task that a person used to handle, the customer needs fewer seats, so a per-seat price stops tracking value. Vendors are responding with usage, credit, hybrid and outcome-based models, and each one changes revenue predictability, gross margin and how buyers approve the purchase.

Title graphic reading From seats to outcomes: AI and software pricing, with a network of connected nodes on a navy background

Why does per-seat pricing weaken when AI agents do the work?

Per-seat pricing works when value grows with the number of people using the product. AI agents break that link, because the more work the software does on its own, the fewer people a customer needs to log in.

An AI agent is software that carries out a task with little or no human input, such as answering a customer, updating a record or drafting a reply. If an agent handles a share of the work that ten support staff used to do, a customer that buys more automation may buy fewer seats. The vendor delivers more value and earns less revenue.

Bain & Company made this point in October 2025 in Per-seat software pricing isn’t dead, but new models are gaining steam. The article argues that AI changes both the value software delivers and its cost structure, which makes a seat-only price a poor fit in some cases. It also notes that seats still fit products where people do the judgement work. The founder’s question is which parts of the product still map to people and which now map to work done.

What are the main pricing models for AI software?

There are five common models: seat, usage, credits, hybrid and outcome-based. They differ in how predictable revenue is, how well they protect gross margin and how easily a buyer’s procurement team can approve them.

Inference cost is the cost of running an AI model each time it answers or acts. Unlike most traditional software costs, it rises with every task, so a flat price on heavy use can erode gross margin.

Seat-based. A fixed price per user per month. Revenue is predictable and procurement teams know how to approve it. The risk is margin: a heavy AI user costs more to serve than a light one at the same price, and seat counts can shrink as automation grows.
Usage or consumption-based. The customer pays per unit used, such as a message, an API call or a document processed. Revenue tracks cost, which protects margin, but it moves with the customer’s activity and is harder to forecast. Buyers often find open-ended bills hard to approve.
Credits. The customer buys a pool of prepaid credits and spends them across different actions. Credits give the buyer a fixed budget line and the vendor upfront cash, while still linking spend to use. They need clear rules on what each action costs and what happens to unused credits.
Hybrid. A platform or seat fee plus a usage or outcome meter for AI work. This keeps a predictable base and captures upside from automation. It is more complex to explain, bill and sell.
Outcome-based. The customer pays only when a defined result is achieved, such as a resolved ticket. Price aligns directly with value, which helps the buyer justify the spend. The vendor carries the risk of work that does not succeed, and both sides must agree on exactly what counts as an outcome.

How are large software vendors pricing AI agents?

Several large vendors have published AI pricing that moves away from seats alone, usually by adding a meter on top rather than replacing seats outright. The examples below are described from each company’s own announcements, for identification only.

Salesforce. On 15 May 2025 Salesforce announced Flex Credits for Agentforce, a consumption model in which each agent action uses 20 credits, priced at US$0.10 per action, alongside its earlier conversation-based pricing of US$2 per conversation. Per-user licences remain, and customers can convert between licences and credits.
Intercom. Intercom prices its Fin AI agent per outcome. Its pricing page lists US$0.99 per outcome, counts a resolution when the customer asks for no further help after Fin’s last answer, and charges once per conversation.
Zendesk. On 28 August 2024 Zendesk announced outcome-based pricing for its AI agents, under which customers incur costs only for issues the AI resolves autonomously, with an in-product dashboard to track usage and forecast spend.
Microsoft. Microsoft priced Microsoft 365 Copilot at US$30 per user per month in July 2023, a seat model. In January 2025 it introduced Microsoft 365 Copilot Chat, in which agents are priced on a metered basis.

Support work, where a resolution is easy to count, has moved furthest towards outcomes. Broad productivity tools have kept seats and added metering for agents.

How widely have pricing models changed so far?

Hybrid pricing is the common interim step, and full outcome-based pricing is still rare.

Bain’s October 2025 analysis of more than 30 software vendors adding generative AI found that about 65% had introduced a hybrid approach with an AI meter on top of seats, roughly 35% had raised seat prices and bundled AI into existing tiers, and none had moved fully to usage or outcome-based AI pricing. Maxio’s 2025 Pricing Trends Report, based on a Benchmarkit survey, found that 44% of SaaS companies charge for AI-powered features and that companies using hybrid models reported the highest median growth rate, at 21%. Looking ahead, Gartner’s July 2025 research, as cited in Deloitte’s 2026 TMT Predictions, expects that by 2030 at least 40% of enterprise SaaS spend will shift towards usage-, agent- or outcome-based pricing. That is Gartner’s view, not a certainty.

How should a scale-up choose and test a pricing change?

Start from the unit of value the customer already recognises, check that the margin works at heavy use, and test with a small group before changing the price list.

Name the value unit. Ask what the customer would say they are buying: a seat, a document processed, a ticket resolved, a meeting booked. If the AI does the work, the unit is usually a task or a result, not a login.
Model the cost to serve. Measure inference and infrastructure cost per task at light, typical and heavy use, and check gross margin for the heaviest users, not the average.
Check you can measure it. Usage and outcome pricing need reliable telemetry and billing. Bain’s analysis notes that most software companies lack this infrastructure. If you cannot count the unit cleanly, the customer will not trust the invoice.
Define the outcome precisely. Write down what counts, what does not, and how disputes are handled. Ambiguity here turns into renewal friction.
Keep a predictable base. For most scale-ups a platform fee plus a meter, or prepaid credits with a commitment, gives buyers a budget line and gives the company a forecastable floor.
Test with new customers first. Offer the new model to a segment of new deals or one market, keep existing contracts unchanged, and compare win rate, deal size, sales cycle and margin over two or three quarters.
Plan the migration. Decide how existing customers move, what protection they get on price and when. Bain points out that customers may pay more in the short term before automation reduces their other costs, so transition terms matter.

What does a pricing change do to forecasting, sales compensation and investor metrics?

Moving away from seats makes revenue more variable, changes what salespeople are paid for and changes how investors read the metrics. These effects need planning before launch.

Forecasting. Seat revenue is mostly known at the start of the year. Usage and outcome revenue depend on customer activity, so finance teams need a forecast built from usage data and committed minimums. Maxio’s 2025 report found that 73% of SaaS companies with usage-based models are actively forecasting variable revenue.
Sales compensation. If salespeople are paid on contract value at signing, they will push for large commitments whether or not customers use them. Many companies with usage pricing pay partly on committed spend and partly on realised usage over the first months. Bain notes that sales teams trained to sell seats need new training and incentives.
How investors read the numbers. Annual recurring revenue (ARR) is the yearly value of recurring contracts. With usage pricing, investors separate committed revenue from variable revenue, look at net revenue retention (NRR) month by month as well as year by year, and check gross margin after inference costs. A company that reports these clearly is easier to assess; how private equity firms value SaaS businesses explains why the quality of recurring revenue and margin weigh heavily in valuation.

Pricing is also part of defensibility. A product that charges for completed work must show that its results are reliable and hard to copy, a theme covered in what will make a SaaS business defensible and valuable.

Where does pricing fit in a growth plan?

Pricing sits inside the revenue engine, next to positioning, sales motion and customer success. A new model changes how the company sells, bills, forecasts and reports.

TGC Capital Partners, the strategic investment arm of Gateway Group and your Growth Engineering Partner, works with founders of software and technology scale-ups in Europe, North America and India on this kind of commercial change. Through 1X Revenue Engineering, TGC works with the founder’s team on pricing, packaging and the sales motion, and through 2X Software Engineering & Tech Excellence on the telemetry and billing a usage or outcome model depends on. Revenue is the outcome TGC measures, not a promise.

AI does not end subscription software, but it does make founders price the work, not just the login.

Frequently asked questions

How is AI changing software pricing?: AI agents do work that people used to do, so customers need fewer seats. Vendors are adding usage, credit, hybrid and outcome-based models that charge for work done.
What is outcome-based pricing?: The customer pays only when a defined result is achieved, such as a support ticket resolved by an AI agent without human help. It needs a precise, agreed definition of the outcome.
Is per-seat pricing dead?: No. Seats still fit products where people do the judgement work. Bain found in October 2025 that most software vendors adding AI kept seats and added an AI meter on top, a hybrid model, rather than replacing seats entirely.
What is the difference between usage-based and credit-based pricing?: Usage-based pricing bills for each unit consumed after the fact. Credit-based pricing sells a prepaid pool of credits that the customer spends across actions, which gives the buyer a fixed budget and the vendor upfront cash.
Why does inference cost matter for software pricing?: Inference cost is the cost of running an AI model each time it acts. It rises with use, so a flat price on heavy AI use can reduce gross margin.
How should a scale-up test a new pricing model?: Define the value unit, model the cost to serve at heavy use and confirm the unit can be measured. Then offer the new model to some new customers, keep existing contracts unchanged and compare win rate, deal size and margin.
How do investors look at usage-based revenue?: Investors usually separate committed revenue from variable usage revenue and check gross margin after AI inference costs to judge how recurring and profitable it is.