Pay-Per-Resolution: A 2026 Buyer's Guide to Outcome-Based AI Agent Pricing

What outcome-based pricing actually means

Outcome-based pricing charges you only when an AI agent delivers a defined business result, rather than for access to the software or for the number of actions it performs. In customer support, where the model first took hold, the outcome is almost always a resolution: the agent handles a customer conversation end to end, and if no human has to step in, you pay a fixed fee for that single resolved issue. If the agent hands off to a person, or the customer leaves unhappy, most vendors charge nothing at all.

The idea is not new in theory — performance-based contracts have existed for years — but applying it to autonomous software at scale is a genuinely 2026 phenomenon. Sierra, the enterprise agent company co-founded by former Salesforce co-CEO Bret Taylor and former Google executive Clay Bavor, built its entire commercial model around it, charging clients only when the agent resolves an issue without human escalation. That approach helped Sierra cross $150 million in annual recurring revenue within eight quarters and raise a $950 million round in May 2026 at a reported $15.8 billion valuation. When a company grows that fast on a pricing model, the rest of the market notices. Intercom renamed its corporate entity to Fin in May 2026 to put its outcome-priced agent front and center, and Salesforce's acquisition and pay-per-resolution launch a month later confirmed the direction of travel.

What the market actually charges in 2026

Concrete numbers matter more than the theory, because the per-resolution price is the lever that decides whether this model saves or costs you money. The customer-service agents that popularized outcome pricing now cluster in a fairly narrow band, roughly $0.50 to $2.00 per resolved conversation, with the exact figure depending on the vendor, your volume commitment, and how much of the platform you buy.

A few details are worth pausing on. Intercom's standalone Fin, which layers onto an existing help desk, carries a 50-outcome monthly minimum of about $49.50, so even a very small operation has a modest floor rather than a large commitment. HubSpot's decision to cut its Customer Agent price to around $0.50 per resolution in April 2026 shows that the per-outcome rate is now a competitive battleground, not a fixed cost of doing business. And Salesforce's Help Agent explicitly does not bill when a customer asks for a human or leaves negative feedback, which is the kind of clause you want written down rather than assumed.

If you are weighing an autonomous agent against keeping the work with people, the per-resolution price is only half the comparison; the other half is the fully loaded cost of the human alternative. Our breakdown of the cost of an AI SDR versus a human SDR walks through that kind of like-for-like math, and the same discipline applies to support, scheduling, and any other workflow where an agent is priced by result.

The one word that decides your bill: "resolution"

Here is the uncomfortable truth every buyer should internalize before signing an outcome-based contract: there is no industry-standard definition of a resolution. Each vendor writes its own, and each one has an obvious incentive to define it broadly. The pricing model looks clean and fair on the marketing page, but the entire economics of your contract live inside a definition you have to negotiate.

The most common approach ties a resolution to a quiet period. The agent answers a customer, and if that customer does not reopen the conversation within a defined window — 72 hours is a frequent choice — the interaction is booked as a billable resolution. That sounds reasonable until you consider the edge cases:

• False-positive resolutions. The agent gives a confident but wrong answer, the customer gives up rather than replying, and the quiet period elapses. You are billed for a "resolution" that actually failed.
• Deflection versus resolution. Some vendors count a deflected conversation — one that never reached a human — as an outcome even if the customer's problem was not really solved.
• Disqualifications and handoffs. A few products bill when the agent disqualifies a lead or routes a query to a documented procedure, expanding what you pay for beyond true resolutions.
• Re-opens outside the window. A customer who comes back on day four starts a "new" conversation, which can be billed a second time.

None of this makes outcome pricing a bad deal. It makes it a deal you have to read carefully. The protection is simple in principle: get the definition in writing, insist on a dispute-and-clawback process for false positives, and confirm that escalations and negative-feedback interactions are genuinely free. This is the same due-diligence instinct we cover in our guide to buying an AI agent without getting burned — the demo always looks flawless, so the contract is where you protect yourself.

Why vendors abandoned per-seat pricing

Understanding why the industry moved helps you predict where it is going. Per-seat pricing was built for tools that make a named human more productive — a CRM seat, an email seat, a design seat. An autonomous agent does not occupy a seat; it does the work a seat-holder used to do. Charging per seat for software that removes seats is a contradiction that customers see through immediately, and it caps the vendor's revenue at exactly the moment the product delivers the most value.

Buyer sentiment reinforced the shift. Pricing surveys across 2026 report that seat-only vendors are increasingly disqualified early in enterprise evaluations, that consumption-based models are now the preferred structure for a plurality of buyers, and that hybrid pricing — a modest platform fee plus per-outcome or per-usage charges — has become the de facto standard for agentic products. The message from the buy side was clear: align the price with the value, or lose the deal.

This is a broader repricing of automation, not an isolated support-desk trend. We explored the wider commercial shift in the new economics of automation pricing, and outcome-based agent pricing is the sharpest edge of it. When the software is measurably doing a job, buyers and sellers both gravitate toward paying for the job done.

Outcome pricing versus the automation platforms you already run

A crucial point that trips up buyers: outcome-based agents and general automation platforms are priced on completely different logic, and most businesses end up paying for both. The agent bills for judgment work by result. The automation platform underneath bills for execution volume, regardless of whether any single run produced a business outcome.

Zapier's June 15, 2026 change is a good illustration of how usage models are getting more granular even outside the outcome world: its AI steps are now priced by model tier, so a premium model can consume roughly five times the tasks of a standard one for the same step. The lesson for buyers is to map the full stack. An outcome-priced support agent may sit on top of a task-priced automation layer that moves data into your CRM and a per-seat help desk your team logs into. Add all three before you decide whether the agent is "cheap."

How to run the numbers before you commit

Outcome-based pricing is not automatically cheaper or more expensive; it is a bet on your own volume and your agent's real success rate. The only way to know whether it wins is to model it against your actual data. Work through this sequence before any contract:

1. Count your resolvable volume. Take the monthly conversations, tickets, or requests the agent could plausibly handle — not your total volume, only the addressable slice.
2. Estimate a realistic automation rate. Vendors quote optimistic deflection numbers; use a conservative rate for the work you actually have, and ask for a pilot to measure it on your data.
3. Multiply out the monthly bill. Resolvable volume × automation rate × per-resolution price gives your expected charge. Then recompute it for a bad month where volume spikes 40%.
4. Compare against a flat or hybrid alternative. Put the outcome bill next to a fixed subscription plus any usage add-ons, and against the fully loaded cost of the humans doing the work today.
5. Price the false positives. Assume a percentage of "resolutions" were actually failures and add the cost you would pay for them absent a clawback clause. If that number is uncomfortable, negotiate the clause.

A quick rule of thumb: outcome pricing tends to favor buyers with low, spiky, or seasonal volume, because you never pay for idle capacity. Flat or hybrid pricing tends to favor buyers with steady, high volume, because the marginal cost of each resolution keeps climbing under a pure per-outcome model while a fixed fee amortizes. If your monthly resolvable volume is unpredictable, outcome pricing is a natural hedge; if it is large and stable, do the spreadsheet carefully, because a per-resolution rate can quietly exceed a flat plan at scale.

The risks buyers underestimate

Outcome-based pricing solves real problems, but it introduces its own. Going in clear-eyed about the failure modes is what separates a good purchase from a regret. The recurring ones in 2026:

• A definition written in the vendor's favor. If you accept the default resolution definition without negotiation, you inherit whatever maximizes the vendor's billable count.
• Budget unpredictability. A model that bills per result produces a variable monthly bill, which finance teams dislike. Insist on spend caps or alert thresholds so a viral spike does not blow your budget.
• Misaligned incentives on quality. A vendor paid per resolution has an incentive to mark more conversations resolved. Your clawback and quality-audit rights are the counterweight.
• Lock-in through data. Your agent's tuning, transcripts, and knowledge base live inside the vendor platform. Without export and portability rights, switching later is painful.
• Opaque success math. If you cannot audit how resolutions are counted, you cannot verify the invoice. Demand transparent, exportable outcome logs.

These are governable risks, not deal-breakers. The point is that outcome-based pricing moves the negotiation away from a sticker price and toward the definitions and controls in the contract. That is where a careful buyer wins or loses, and it is the same reason we tell buyers to treat the paperwork, not the demo, as the real product.

A buyer's checklist for outcome-based agents

Bring this list to every evaluation. If a vendor cannot answer these clearly and put the answers in writing, treat that as a signal about how the relationship will go once you are live and the invoices start arriving.

• Definition: Exactly what counts as a billable outcome, in writing, with examples of what does not.
• Quiet period: How long before an interaction is booked as resolved, and how re-opens are handled.
• Free events: Confirmation that escalations, human handoffs, and negative-feedback conversations are not billed.
• False positives: A dispute and clawback process for resolutions the customer later reopens or complains about.
• Spend controls: A monthly cap, a hard stop, or at least an alert threshold you configure.
• Transparency: Exportable outcome logs you can audit against the invoice.
• Portability: Export rights for your knowledge base, transcripts, and configuration, plus a defined offboarding path.
• Pilot: A trial period on your real data with a shared, measurable success metric before production volume.

FAQ

What is outcome-based pricing for AI agents?

It charges you only when an agent delivers a defined result, usually a resolved customer conversation, rather than for seats or a flat subscription. Fin sits near $0.99 per resolution, HubSpot cut its Customer Agent to about $0.50 in April 2026, and Salesforce launched a pay-per-resolution Agentforce Help Agent in June 2026 that only bills when the issue is fully resolved with no escalation.

How is a "resolution" defined?

There is no standard definition, which is the crux for buyers. Most vendors count a resolution when the customer does not reopen the conversation within a quiet window such as 72 hours, and some also bill deflections, disqualifications, or procedure handoffs. Pin the definition down in the contract.

Is outcome pricing cheaper than per-seat?

Only if your volume and resolution rate make it so. It favors low or spiky volume because you pay nothing for unresolved tickets, but at steady high volume a flat or hybrid plan can beat a per-resolution rate. Model both against your real numbers.

What are the biggest buyer risks?

A resolution definition written in the vendor's favor, paying for false-positive resolutions, unpredictable monthly bills, a vendor incentive to over-count resolutions, and data lock-in. Contractual definitions, clawbacks, spend caps, and portability rights mitigate all of them.

Why did vendors switch away from per-seat pricing?

Because seats make no sense for software that replaces work rather than assisting a user. Sierra popularized charging per successful resolution and crossed $150 million in ARR within eight quarters, and Salesforce's roughly $3.6 billion Fin acquisition in June 2026 confirmed outcome pricing as the new default for autonomous agents.

How does this compare to Zapier and Make pricing?

Those platforms bill for execution volume, not outcomes. Zapier charges per task and, from June 15, 2026, prices AI steps by model tier; Make charges per operation from about $9 per month for 10,000 operations. Outcome-priced agents sit on top of that layer, so many teams pay for both.

Should a small business pick outcome or flat pricing?

Small or seasonal operations usually benefit from outcome pricing and modest minimums, for example around $49.50 per month for 50 outcomes on standalone Fin. Steady high-volume businesses often do better on a predictable flat or hybrid plan. Decide with a spreadsheet, not a preference.

What belongs in an outcome-based contract?

A precise outcome definition, the quiet-period length, confirmation that escalations are free, a false-positive clawback process, a spend cap, exportable outcome logs, data-portability rights, and a pilot with a shared success metric before production.