AI Business Models.

In this model, the customer pays only when the AI agent achieves an agreed result: a resolved support ticket, a recovered chargeback, a saved customer. Intercom, the customer service software company, charges $0.99 for each conversation its Fin agent resolves¹. Sierra, the customer service agent company co-founded by former Salesforce co-CEO Bret Taylor, says it reached $200 million in annualized revenue in May 2026^4,5. It's still rare as a company's main pricing model. In a 2026 survey of 230 software companies by Kyle Poyar, who writes the Growth Unhinged newsletter,, only 5% used outcome pricing as their primary model⁷. The appeal is obvious: no result, no bill. The hard part is agreeing on what a result is.

The vendor carries the risk

This is what separates outcome pricing from pay per job (model 4). When the agent fails, the customer doesn't pay, but the vendor still pays for the compute. Intercom doesn't charge when Fin hands a conversation to a person, when the customer asks for a human, or when the customer goes quiet after a clarifying question¹. That makes the outcome rate the vendor's most important number, because every point of improvement is revenue. Sacra, a research firm that tracks private tech companies, reports that Fin resolves 66% of conversations on average, and more than 20% of its customers see over 80%³. That's the improvement loop on the map: tune the agent, and more of its work becomes billable.

Everything depends on what counts

Intercom counts a resolution when the customer confirms the answer helped, or when they leave without asking for more help¹. Zendesk infers resolution from confirmation or inactivity, then checks it with an AI model after 72 hours⁶. Utpal Dholakia, a marketing professor at Rice University who writes about pricing, points out the problem: silence can't tell a satisfied customer from one who gave up, and the vendor gets paid either way⁶. Customers notice. On Intercom's community forum, one customer found 16 of 40 sampled tickets marked as resolved by Fin, but judged only about 3 to be truly solved. Intercom admitted it had misstated how one case was counted and said it was considering changes to how it measures resolutions². That's the trust loop on the map: disputes lead to a tighter definition of the outcome.

It works best for narrow, countable work

Scott Woody, CEO of the billing platform Metronome, calls outcome pricing "mostly a myth" for most companies. In his view, only vendors with monopoly-like power or very large contracts can measure outcomes well enough to price on them⁸. Dholakia argues it works best for narrow tasks and should be combined with traditional pricing for broader products⁶. Elena Verna, who leads growth at Lovable, says she loves outcome pricing "in theory" because it aligns the vendor's success with the customer's, but that it gets hard when outcomes are hard to define or the two sides disagree on what success looks like⁹. In practice most vendors mix models. Sierra blends volume pricing for routine conversations with outcome pricing for more complex ones⁴, and Zendesk charges about $2 per automated resolution pay-as-you-go, or about $1.50 with committed volume⁶. Some vendors take a share of the value instead. Chargeflow, which fights credit card chargebacks for merchants, takes 25% of the chargebacks it wins⁷.

Proven results earn new use cases

Outcome pricing is usually sold to larger companies through a pilot, and the pilot has to prove the outcome rate before the deal grows. Once it does, expansion comes from new channels and workflows. Sacra estimates Sierra went from about $26 million in annualized revenue at the end of 2024 to $100 million in November 2025⁴, with customers including ADT, Chime, Nordstrom and Rivian. The churn risk is different from other models. A customer who has learned which work an agent can do may decide to build its own agent, or switch to a vendor with a better price per outcome.