Outcome-Based Pricing: Examples & Companies

What is it

Outcome-Based Pricing is a pricing model where the customer is charged per business outcome — a resolved support ticket, a converted lead, a closed sale — rather than per unit of input.

The defining move is to meter the result instead of the resource. Seat-based pricing charges for access. Usage-based pricing charges for consumption — tokens, API calls, minutes — whether or not that consumption produced anything. Outcome-based pricing charges only when the vendor delivers a measurable business result, which is why it is often called the holy grail of usage-based pricing: vendor revenue and buyer value move in lockstep.

The model took hold first in AI customer support, where the “outcome” has a natural unit: the resolved ticket. Intercom defined the category with Fin AI Agent at $0.99 per resolution — the customer pays nothing for an AI conversation that fails to solve the issue, and Intercom charges the same $0.99 across its Essential, Advanced, and Expert seats and its standalone Fin product. Lorikeet states the promise even more bluntly on its pricing page: it only charges for successfully resolved tickets, and refunds tickets it handled poorly. Gorgias bills a flat $0.90 per resolved conversation as an AI Agent add-on layered on its ticket-metered helpdesk tiers.

The model is no longer confined to support. Pixee charges per resolved security vulnerability in application security, and Digits reserves a custom outcome-based band for its largest (500+ client) accounting firms. At the far end of the scale, drug-discovery labs such as Recursion and Isomorphic Labs charge nothing per seat and everything per milestone — an upfront fee plus payments that trigger only as an AI-designed molecule clears each clinical gate. The common thread across all 24 companies is a countable, attributable result. The hard part, in every domain, is defining that result precisely enough that buyer and seller agree on what counts.

How it works

An outcome-based model has to answer three questions: what is the outcome, what does it cost, and how is it counted. The outcome is the billable unit; the rate is the price per unit; the counting rule is the contractual definition that determines when a charge fires. In practice, the counting rule is the term buyers and vendors spend the most time negotiating — the headline rate on a pricing page is a number, but the definition of “resolved” is the price.

The unit math is deceptively simple — and that simplicity is the selling point versus token accounting:

Unit math: monthly bill = resolved_outcomes × rate. Intercom: 5,000 resolutions × $0.99 = $4,950. Gorgias: 1,500 resolved conversations × $0.90 = $1,350. Gladly: 10,000 assisted conversations × $0.60 = $6,000.

Two structural wrinkles recur. First, most vendors layer the outcome charge on top of a fixed component, so the model reads as hybrid: Intercom charges $29–$132 per seat plus $0.99/resolution, and Gladly charges a $180–$210 per-Hero seat package plus the $0.60 Sidekick AI conversation charge — at moderate volume the Sidekick line ($6,000 on 10,000 conversations) can exceed the entire human-seat line. Second, vendors that sell on annual commitments express the outcome rate through a credit pool. Lorikeet is the clearest case: Start ($1,500/mo) buys an 18,000-credit annual pool and Scale ($4,000/mo) buys 48,000 credits, and each resolution consumes 0.95 credits for chat/email/SMS or 1.50 for voice on Start, dropping to 0.80 and 1.20 on Scale. The outcome is still the meter; the credit is just the accounting wrapper.

Companies using this

Twenty-four companies in the corpus bill on outcomes. The largest cluster is customer support — Intercom, Intercom Fin, Ada, Gorgias, Gladly, Kustomer, Lorikeet, Forethought, Maven AGI, Sierra, Yellow.ai, Zendesk AI, and Decagon — all converging on per-resolution or per-conversation billing for AI-deflected support tickets. Beyond support, the model extends into application security (Pixee), accounting (Digits), sales and cloud cost (Rox, Usage AI), legal and radiology AI (EvenUp, Rad AI), and drug discovery and medical AI (Exscientia, Insilico Medicine, Isomorphic Labs, Recursion, Viz.ai). The table below shows each company’s full pricing-model classification, billing units, and verification date side by side.

Patterns observed

Customer support has become the bellwether for outcome-based pricing. Thirteen of the outcome-priced companies sell support automation, and they cluster on the same unit — a resolution. Where Intercom and Gladly once stood mostly alone, the per-resolution standard has been adopted by Forethought, Maven AGI, Zendesk AI, and Decagon as well. The interesting competitive question is no longer whether to use outcome pricing, but how to define the outcome and at what rate; on both dimensions the field is still fractured, which is exactly the signal we track in the outcome-based-pricing trend.

What the rate clustering conceals is a much wider spread in definition. The headline numbers sit in a tight band, but they are not metering the same thing. Kustomer charges per engaged conversation whether or not the AI actually resolved anything; Gladly calls its unit an “assisted conversation,” which is broader than a resolution; Yellow.ai meters a full resolution. This definitional divergence matters more than the rate spread: a buyer comparing two vendors at the same headline rate needs to understand that one may bill twice as many units on the same ticket volume. The number on the pricing page is a rate, not a price, until you have read the counting rule — a point our introduction to usage-based pricing makes about every value metric.

The pattern of domain migration is the structural story of the last twelve months. Drug-discovery companies — Exscientia, Insilico Medicine, Isomorphic Labs, Recursion — and medical AI such as Viz.ai apply outcome-based structures at an entirely different scale and timeframe than support automation. Their “outcomes” are drug-development milestones: Recursion’s Roche/Genentech deal carries $150M upfront and up to 40 programs each with $300M+ of milestone potential, and Exscientia’s Sanofi collaboration ran $100M upfront against up to $5.2B in milestones plus royalties. These are milestone-and-royalty structures, not unit-rate transactions. EvenUp and Rad AI take a middle position: EvenUp prices per case (historically ~$300 per demand package), while Rad AI’s Continuity module is sold on downstream-imaging revenue ($50–$250 net reimbursement per follow-up study) rather than a seat fee. The common thread across all these domains is that the result is countable and attributable; the execution scale ranges from milliseconds (a support deflection) to years (a drug approval).

Pure outcome billing — where the only charge is a per-outcome rate with no fixed component — is the rare challenger position. Ada and Pixee come closest, alongside the savings-share players. Most of the corpus runs hybrid-outcome structures instead: Intercom adds per-seat charges, Rox layers a base tier under its Agent Actions meter, credit-pool vendors anchor to an annual commitment, and the drug-discovery companies add substantial upfront and platform fees. The fixed floor de-risks vendor revenue in both directions — it guards against low-volume months and gives the buyer a predictable baseline — while the outcome meter preserves the incentive alignment that makes the model compelling in the first place.

Counterexamples & variants

Outcome-based pricing fails where the outcome can’t be cleanly attributed or counted, and the most instructive variants in the corpus are the ones that quietly retreat from it. Pixee makes the attribution problem explicit: rather than publish a price, its pricing page answers “What is a resolution?” in the FAQ, because the entire negotiation is over what counts as a fixed vulnerability — its quote is calculated off annual scanner findings (SAST + SCA). When the definition of the outcome carries the whole commercial relationship, vendors pull the number off the public page and quote it in sales, which is why Ada, Kustomer, and post-2024 Gladly are all sales-gated despite billing on outcomes. Gladly is especially striking: it published a clear $180–$210/Hero and $0.60/assisted-conversation table on gladly.com through September 2024, then withdrew all of it when it migrated to gladly.ai.

The most common variant is the credit-pool wrapper, which can blur the outcome guarantee. Lorikeet charges credits not only for resolutions but for non-resolution actions — routing, analytics tagging, and automated QA each draw fractional credits per ticket — so a buyer who reads “we only charge for resolved tickets” still sees credits consumed on tickets that weren’t resolved. That isn’t a contradiction so much as a reminder that the credit meter and the outcome promise are two different layers. Credit pools are particularly prevalent among vendors targeting enterprise buyers on annual contracts, because the pool structure gives procurement teams a budget ceiling to negotiate against while preserving the per-resolution meter beneath it. The usage invoicing and billing-cycles guide covers how these pooled counts roll into an actual invoice.

Drug discovery represents a structurally different variant that deserves its own reading. The milestone-and-royalty deals above are outcome-based pricing at a macro scale: the “outcome” is a milestone (a target validated, an IND filed, a Phase I trial initiated, a regulatory approval), and the “rate” is a negotiated ceiling that compounds over years. The eye-catching headline totals are contingent and cumulative across many milestones — most programs never trigger the full amount because most drug candidates fail. The defining challenge is attribution: when an AI system surfaces a candidate that a human team then develops over three years, determining exactly what the AI contributed to the eventual outcome is a contractual problem that dwarfs anything in the support-ticket world.

The honest counterexamples at the high-frequency end are products where the AI’s contribution can’t be isolated from the human’s. Rox sells an AI sales-agent swarm but meters “Agent Actions” — a usage unit whose cost scales with the work the agents do — rather than closed deals, precisely because a closed sale has too many non-Rox inputs to bill as a pure outcome; its “starts at $50/month” Core plan buys a 5,000-action pool that a heavy user can burn through in a single month. Usage AI sits at the other extreme, with the cleanest pure-outcome model in the corpus (a percentage of realized cloud savings, $0 floor) — but even there the watch-item is the definition of “realized savings,” because the baseline against which the percentage is computed is the lever the vendor controls. Both examples mark the boundary of the model: where the result is jointly produced or the baseline is contestable, the counting rule does all the work.

What this means for buyers vs vendors

For buyers

Read the counting rule before the rate. A $0.60 charge per engaged conversation (Kustomer) and a $0.99 charge per resolved ticket (Intercom) are not comparable until you know what triggers the meter and who adjudicates a disputed resolution. Ask: does a reopened or escalated ticket get refunded, are non-resolution actions billed separately, and can you audit the resolution log? For sales-gated vendors like Sierra, Decagon, and Maven AGI, the per-resolution rate itself is the negotiation, so anchor against the public benchmarks ($0.99 Intercom, $0.90 Gorgias) before you sign. Our introduction to usage-based pricing covers how to evaluate value metrics like these, and the usage invoicing and billing-cycles guide explains how outcome counts roll into an invoice. You can model a blended seat-plus-outcome bill with our pricing calculator hub, or start from the Intercom pricing calculator to see how $0.99/resolution stacks on top of seats at your ticket volume.

For buyers in drug discovery, radiology, and legal AI, the negotiation is categorically different from the per-ticket support world. Milestone definitions need the precision a patent attorney would use: what constitutes a “candidate identified,” who verifies that the IND was filed, what happens to milestone payments if the program is discontinued mid-pipeline. Attribution clauses deserve particular attention — most of these contracts include language about what percentage of the discovery must be traceable to the AI system’s work before a milestone fee applies. For ROI-framed products like Rad AI’s Continuity or Viz.ai’s NTAP-offset stroke suite (up to $1,040 per eligible Medicare use), scrutinize the reimbursement and volume assumptions baked into the vendor’s calculator — the headline “opportunity” is only as good as the conservative-case defaults. Buyers should insist on audit trails that document which parts of the pipeline the AI ran, what it produced, and how that output connected to the downstream event that triggers the charge.

For vendors

Outcome-based pricing only works when the outcome is countable, attributable to your product, and auditable by the buyer. If any of those break, you will spend the relationship arguing over the meter. Most successful implementations in the corpus add a fixed floor — seats (Intercom), a base tier (Rox), an annual credit commitment, or an upfront fee (the drug-discovery labs) — to de-risk revenue while keeping the outcome meter for incentive alignment. The rare pure-outcome plays like Usage AI work because the result (realized cloud savings) is unambiguous and lands as a single line item on the buyer’s existing cloud bill. Above all, invest in defining the outcome precisely: the counting rule is not fine print — it is the product.

The expansion of outcome-based pricing into drug discovery, radiology, accounting, and legal AI opens a larger opportunity surface for vendors in adjacent domains. Any workflow that produces a discrete, attributable, high-value result — a diagnosis flagged, a contract clause extracted, a compliance finding closed, a firm’s manual work measurably reduced (the basis of Digits’ enterprise band) — is a candidate for outcome metering. The barrier is not technical but contractual: the vendor must produce an audit trail both parties accept as the basis for billing, and the buyer must trust that the count is honest. A cautionary note from the corpus is transparency — even outcome-aligned vendors like Sierra and Decagon keep their per-resolution numbers private, which slows buyer trust. The vendors most likely to win the next wave are the ones that publish a flat, public rate the way Intercom and Yellow.ai do, and let the number itself become the category anchor.

Explore this theme in the knowledge graph →