For twenty years, business software was priced the same way: a monthly fee for each person who logs in. That model fit tools that people used. It fits less well when the software does the work itself. If an AI agent answers half of a support team's tickets, the customer needs fewer seats, and a vendor paid per seat earns less the better its product performs.
A growing number of AI companies now price on results instead. This guide explains what outcome-based pricing is, how it compares with seat and usage pricing, which companies already use it, and what makes an outcome countable. It then covers the ways the model goes wrong, how a buyer should evaluate an outcome-priced contract, and how a vendor can make the move.
Key takeaways
- Outcome-based pricing charges for a defined result, such as a resolved conversation, instead of for users or usage.
- AI pushes software toward it because agents reduce the number of seats a customer needs.
- The hard part is the definition: an outcome must be unambiguous, measurable and confirmed by proof.
- Buyers should ask who decides an outcome happened, how disputes are settled and what spend looks like at peak volume.
What is outcome-based pricing?
Outcome-based pricing is a model in which a customer pays a set amount each time the software achieves an agreed result. The unit of value is the result itself: a support issue resolved without a person, a meeting booked, a cancellation saved, an invoice collected. If the result does not happen, the customer usually pays nothing for that attempt.
Sierra, the customer-service AI company, described the idea in a December 2024 post. Like consumption pricing, it varies with use, but it "is tied to tangible business impacts," and in the company's words, "Sierra gets paid only when we complete a task for you" (Sierra).
The idea is older than AI. Performance marketing has long been paid per conversion, and some consultancies take a share of savings. What is new is that software can now finish whole units of work, so it can be paid per unit of work finished. That is also why outcome pricing tends to follow autonomy. The more of a process runs from start to proof without a person, the easier it is to count each finished unit, which is the path we describe in the autonomous company.
Seat-based vs usage-based vs outcome-based pricing
Each model answers a different question: who uses it, how much does it run, or what did it achieve?
| Model | What you pay for | Good for the buyer when | Risk for the buyer |
|---|---|---|---|
| Seat-based | Each user, per month | Usage is heavy and predictable | Paying for seats nobody uses |
| Usage-based | Calls, tokens, actions or conversations | Volume is low or spiky | Bills that grow with activity, useful or not |
| Outcome-based | Each agreed result | The result is clear and valuable | Disputes over what counted, and hard-to-forecast spend |
| Hybrid | A platform fee plus results or usage above a tier | Both sides want a floor and upside | More terms to negotiate and check |
In practice many contracts are hybrids. A platform fee covers fixed cost and a base volume, and outcomes or actions are billed above that line.
Why AI is moving software away from seats
Three forces push in the same direction.
Agents shrink seat counts. Sierra's post argues that established customer-service vendors are "trapped in a conflict: the more effective their AI becomes, the fewer contact center seats their clients need." A seat model punishes the vendor for doing its job. The same logic applies to every role where AI employees take over recurring work: the customer buys fewer licences, and the value moves to the work itself.
AI has real running costs. Bessemer Venture Partners' February 2026 pricing playbook notes that AI companies see gross margins of 50 to 60%, compared with 80 to 90% for classic software, and that AI-native companies are "abandoning seat-based SaaS pricing" for usage, output and outcome models (Bessemer Venture Partners). When each task costs compute, pricing has to track tasks.
Buyers want proof of value. Many companies ran AI pilots in 2024 and 2025. Bessemer observes that many of those pilots reach their first renewals in 2026, where pricing will need to show measurable value. Paying per result is the most direct way to show it. This links to a broader shift we cover in service as software: when software delivers the service, it starts to be priced like one.
Who prices AI on outcomes today?
Customer support moved first, because a resolved ticket is one of the easiest outcomes to define.
- Intercom. Its pricing page lists its Fin AI agent at "$0.99 per outcome". An outcome counts when the customer confirms the problem is fixed, when they do not ask for more help after Fin responds, or when Fin completes a defined workflow, and Intercom charges once per conversation (Intercom).
- Zendesk. In September 2024, Zendesk introduced pricing under which customers "only pay for issues that are fully resolved by AI" (CFOTech Asia).
- Sierra. Prices most conversations by outcome, and notes that "if the conversation is unresolved, in most cases, there's no charge."
- Salesforce. Moved its Agentforce pricing in May 2025 from $2 per conversation to Flex Credits, at $500 per 100,000 credits, with one action using 20 credits, or $0.10 (Constellation Research). That is usage pricing at the level of individual actions, a step between seats and outcomes.
The pattern is that outcomes win where the result is simple to see. Sierra itself notes that "routing or greeter-style interactions may align better with consumption-based pricing." Not every task has a clean finish line.
What counts as an outcome?
Every outcome-priced contract stands or falls on its definition. Results sit on a ladder, and the higher the rung, the more valuable and the harder to attribute.
Bessemer's playbook puts the test plainly: outcome pricing fits "when the outcome is unambiguous and measurable," when the vendor is confident in performance and can absorb cost variance. Three questions make that concrete:
- Is there proof? A sent email, a closed ticket, a booked meeting or a paid invoice can be checked. "Customer felt helped" cannot.
- Who confirms it? The customer, the end user, or a rule both sides agreed in advance.
- Is it attributable? If a person did half the work, the contract should say whether that counts.
Intercom's three-part definition is a good model: it names exactly which events count. The company autonomy metrics guide applies the same discipline inside a company.
Where outcome-based pricing goes wrong
- Soft outcomes. If the outcome is "a happy customer", nobody can audit the bill. Disputes follow.
- Silent customers counted as resolved. A user who gives up looks the same as a user who was helped. Definitions that count silence need sampling and review.
- Cost variance for the vendor. Bessemer warns that "a difficult customer issue could consume far more compute than anticipated." Vendors may respond by narrowing what they attempt.
- Unpredictable budgets for the buyer. Finance teams want a fixed number. Bessemer cites the case of Leena AI, whose early consumption pricing made customers wary of using the product until it moved to outcome pricing.
- Gaming. Any metric that triggers payment will be optimised. Without proof and review, both sides can drift toward what counts instead of what helps.
How buyers should evaluate an outcome-priced AI contract
Treat the outcome definition like a payment term. Before signing, get written answers to these:
- What exact event counts as an outcome, and where is it recorded?
- Can we see the list of billed outcomes with links to the proof for each?
- What happens when a person finishes the work the AI started?
- How are disputes raised and settled, and within what time?
- Is there a cap, a committed volume or a price that drops with volume?
- What do we pay for attempts that fail or are escalated?
- What would our bill have been last quarter at our real volume?
The last question is the most useful. Run the vendor's definition over a month of your own history before committing. For broader buying questions, see how enterprise AI will actually be adopted.
How vendors can move from seats to results
- Pick one countable outcome.
Choose the result customers already measure, with proof that lives in a system both sides can see.
- Instrument the proof.
Record each outcome with a link to its evidence, so every line on an invoice can be checked.
- Model your cost per outcome.
Include the hard cases. Price so a bad month does not wipe out margin.
- Offer a hybrid first.
A platform fee plus outcomes above a base volume gives buyers a budget and keeps the incentive aligned.
- Publish the definition.
Put it on the pricing page in plain words. Vague definitions cost more in disputes than they save in sales.
Measure the move with three numbers: the share of revenue from outcomes, the dispute rate on billed outcomes, and gross margin per outcome over time.
How GeniOS approaches outcome-based pricing
GeniOS is priced by plan today, as listed on the pricing page: Startup at $99 a month and Growth at $299 a month, with Enterprise coming soon. A predictable plan suits a company that is starting read-only and building trust step by step.
The product is built around the thing outcome pricing depends on: proof. GeniOS reads email, calendar, chat and the business tools a company already uses, forms the company's goals from its own emails, decks and meetings, and checks every team's pending work against them. Each item reaches one owner with what to do, why and a date, and the owner answers Delegate, Done, Later or Not relevant.
An item closes only on proof, such as the email that went out, and nothing is sent until a human approves. That record of closed work is what lets a company see results instead of activity, whichever way its software is priced.
Worked example: pricing an AI follow-up agent at a safety-shoe maker
The names, dates and numbers in this example are made up to explain the idea. Vaultex makes safety footwear in Greater Noida and was GeniOS's first design partner. Its sales lead, Harsh, is choosing an AI agent to chase quotes and buyer documents.
Two vendors offer contracts. One charges per seat. The other charges per outcome, defined as "a buyer request answered". Harsh tests the outcome definition against a real week:
- Bata wants a quote for 12,000 pairs of S3 safety shoes at $12 a pair, about $144,000, before Friday's vendor review, on 60-day payment terms.
- Adidas needs IS 15298 test reports for its vendor audit.
- Reebok asked when the Line 2 samples will ship.
The test exposes the definition. An agent could "answer" Bata within an hour with the old $12 price, and the vendor would bill it. But steel toe-cap costs rose 8% this week, so a fast answer at the old price would cost Vaultex margin on a $144,000 order. The Reebok answer depends on Anushka in production confirming the Line 2 date. A good outcome here is "the correct quote sent after the price and terms were checked", which an "answered" count cannot see.
Harsh negotiates a different unit: an outcome counts when the approved email is in the sent folder and the request is closed, with a monthly cap. That definition matches proof Vaultex can check, rewards correct work over fast work, and gives finance a number to plan around.
Conclusion
Outcome-based pricing is spreading because AI agents do work that used to require seats, and because buyers want to pay for value they can see. It works best where the result is unambiguous, measurable and backed by proof, which is why customer support moved first. For buyers, the definition is the contract: test it on your own history before signing. For vendors, publish it plainly and price for the hard cases. Either way, the companies that keep a clean record of closed, proven work will find it easier to buy and sell on results. To see how GeniOS keeps that record, see pricing or book a call.
Sources
- Sierra, Outcome-based pricing for AI agents, 10 December 2024
- Bessemer Venture Partners, The AI pricing and monetization playbook, 10 February 2026
- Intercom, Pricing, accessed 10 November 2026
- CFOTech Asia, Zendesk introduces outcome-based pricing for AI solutions, 26 September 2024
- Constellation Research, Salesforce revamps Agentforce pricing with Flex Credits, 15 May 2025