Outcome-based pricing
Paying when the AI does the job -- per outcome, not per token or seat -- has gone from a single live example to a documented pattern across the corpus. Twenty-four of 338 corpus companies (about 7%) now price on outcomes; the absolute count has roughly tripled as the corpus grew, yet it remains a minority model. It is a category standard in customer support, where Ada, Gorgias, Gladly, Kustomer, Lorikeet, Intercom Fin (at $0.99) and Parloa all bill per resolution, with Rox (per Agent Action in sales), Pixee (per code fix) and Digits (accounting outcomes) extending it beyond support.
What's happening — and why
What's happening: a growing set of products charge by the result the AI produces, not the work it does to get there. Twenty-four of the 338 corpus companies — about 7% — now meter on an outcome unit, up from a single live example when this trend was first logged; the absolute number has roughly tripled as the corpus expanded, but outcome pricing is still a minority model. In customer support it is now the default for new entrants — Intercom's Fin bills $0.99 per resolved conversation, and Ada, Gorgias, Gladly, Kustomer, Lorikeet and Parloa all meter on resolutions. Beyond support, Rox bills per Agent Action for sales tasks, Pixee per merged code fix, Digits on accounting outcomes, and the per-task unit has spread into pharma (Recursion milestones), AI training data (Mercor, Micro1) and contact-center QA (Observe.AI).
Why it works here: outcome pricing needs a result that's cleanly measurable and attributable to the AI. A resolved support ticket is about the cleanest such unit there is, which is why the support category converged on it first. Open-ended generation still doesn't qualify — but wherever an agent performs a discrete, checkable task, the outcome is becoming the meter.
How it works
Evidence over time
22 supporting · 7 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Intercom | Mar 2024 | Fin AI agent priced per resolution ($0.99) — you pay only when the bot actually resolves a conversation, not per seat or per message. The first live outcome-priced product in the corpus. |
| Rox | Jun 2026 | AI sales-agent swarm metered in Agent Actions — each task (account research, meeting brief, pipeline generation) costs a defined number of actions scaled to complexity and model used. Tagged outcome-based in taxonomy. Free Starter (2k actions/mo), Core $50/mo (5k actions), custom Enterprise. |
| Gorgias | Jun 2026 | Published a transparent flat per-resolution AI Agent price alongside its tiers — outcome (resolution) billing made fully public in customer support. |
| Ada | Jun 2026 | Outcome/per-resolution model (third-party data ~$1–$3.50/resolution); the resolution unit is the core meter, now sales-gated. |
| Lorikeet | Jun 2026 | Repriced to channel-based per-resolution credits with a 2–3x list increase — resolution as the billable unit, now segmented by channel. |
| Kustomer | Jun 2026 | Hybrid seats + resolutions — per-resolution outcome billing layered onto seat-based CS pricing. |
| Parloa | Jun 2026 | Quote-only model metered on media-minutes and resolutions — voice CS agent billed on the resolved outcome. |
| Pixee | Feb 2026 | Pivoted to outcome-based, quote-only pricing in security — billed per fix/resolution (auto-remediated vulnerability) rather than per seat or scan. |
| Digits | Apr 2026 | Launched outcome-based Enterprise pricing for accounting firms — billing tied to delivered accounting outcomes, not seats. |
| Recursion | Jun 2026 | Pharma AI company billing on milestones and outcomes from drug-discovery partnerships — the most capital-intensive outcome-based billing in the corpus, where the 'outcome' is a clinical-trial milestone worth tens of millions of dollars. |
| Mercor | Jun 2026 | AI training marketplace billing per-task (sales-only) for AI data labeling and evaluation — outcome-based billing applied to human+AI annotation work. |
| Micro1 | Jun 2026 | AI labeling and tasks platform metered per-task (sales-only) — confirms the per-task unit extends into AI training data markets. |
| Observe.AI | Jun 2026 | Per-agent + per-interaction pricing for contact-center AI quality assurance — the interaction as a billable outcome unit in enterprise voice operations. |
| HubSpot | Jul 2026 | Moved Breeze Agents to outcome pricing: the Customer Agent bills $0.50 per RESOLVED conversation (50 HubSpot Credits, down from $1.00 per conversation) and the Prospecting Agent $1.00 per recommended lead (100 credits) — a 'you pay when it works' resolution unit on a horizontal CRM platform, explicitly positioned to undercut Salesforce Agentforce's per-conversation meter. |
| Gorgias | Jul 2026 | Raised its per-resolution economics: AI Agent overage past the bundled allotment went from $0.90 to $1.50 per automated interaction (+67%), with a modest included allotment (30 interactions on Basic) at ~$0.90 each. The per-resolution unit is settled; the dollar figure moved UP — the opposite direction to HubSpot's cut the same week. |
| Sierra | Jul 2026 | The outcome unit's time horizon expands from a conversation to a quarter. Sierra launched Horizon, replacing its Agent Data Platform line: Playbooks give an agent 'an outcome — a mortgage, a claim, a renewal — and they will plan and execute the work over weeks or months', with long-horizon planning and next-best-action decisioning. The company's own framing on the site-wide banner and homepage is explicit — 'It is Sierra's outcomes-based model at scale: you pay for results, not tokens' — and the homepage headline moved from 'Better customer experiences' to 'Better outcomes. Built on Sierra.' The number did not change; the unit's duration did. |
| Lorikeet | Jul 2026 | The corpus's first DURATION BOUNDARY on an outcome unit. All three plan cards (Start, Scale, Enterprise) now carry an asterisk on the 'Voice resolution' line resolved by a footnote: 'Voice resolution price is for resolutions up to 3 minutes long.' The credit rate itself held — 1.50 credits per voice resolution on Start, 1.20 on Scale, Custom on Enterprise — but no rate is published for resolutions past three minutes, so longer calls become a quoted item. Every other published number held (Start $1,500/mo for 18,000 credits/yr, Scale $4,000/mo for 48,000). |
| Observe.AI | Jul 2026 | First published per-interaction outcome rate — on AWS Marketplace, not its own site. observe.ai/pricing returned 404 again on this capture, while the VoiceAI Agents for Customer Support listing prices two usage dimensions with NO per-agent seat dimension at all: Number of Interactions $12.00 per unit and Minutes per Month $4.80 per unit on a 12-month prepaid contract ($24.00 and $9.60 for 24 months — exactly double, so the longer commitment carries no posted discount). Buyers prepay entitlement units against a stated quantity of use for the term. |
| RunPod | Jul 2026 | Outcome-CONTINGENT pricing reaches GPU compute. RunPod's new enterprise surface sells Runpod Overdrive on an explicit if-we-do-not-beat-your-baseline-you-pay-nothing guarantee, alongside four named commercial components (reserved baseline capacity, usage-based burst, committed-use pricing, post-paid billing on a single invoice) and clusters from 200+ on demand to 10,000 reserved. No enterprise dollar figures are published. The billable event is contingent on a measured performance outcome rather than on consumption — the outcome logic applied to infrastructure rather than to a support ticket. |
| Usage AI | Jul 2026 | Pure savings-share outcome pricing, with the rate withdrawn from its own site. The model is unchanged — a percentage of realised cloud savings, $0 platform fee, no minimum, no contract, full cashback on underused commitments — but the mid-2026 redesign deleted the inline fee calculator (a worked example putting a $1,200 fee on $8,000/mo of savings, ~15%) and the ICR-vs-ESR comparison. The FAQ now reads only 'a percentage of the savings we realize for you, billed monthly'; the only published rates are on AWS Marketplace: 20% of EC2 savings, 35% on RDS/ElastiCache/Redshift/OpenSearch. |
| HubSpot | Jul 2026 | Outcome prices proved stickier than the brand carrying them: HubSpot deleted the Breeze name from every surface for 'Agent Hub (Beta)' — adding HubSpot AEO and Agent Builder — while every per-outcome rate held exactly. Customer Agent $0.50 per resolution, Prospecting Agent $1.00 per recommended lead, Data Agent $0.10 per answer, all still billed against HubSpot Credits. A full rebrand with zero movement on the outcome meter, two weeks after the meter itself was introduced. |
| Yellow.ai | Aug 2026 | The first outcome-priced corpus company to reach a public listing with the outcome meter intact — a durability test rather than a pricing move. Yellow.ai agreed to go public via a SPAC merger with Bluerock Acquisition Corp (NASDAQ: BLRK) valuing the combined company at roughly $550M ($300M stock deal plus a $5M equity facility and a $50M PIPE note), confirmed by an official PR Newswire release and reported consistently by CMSWire, Stock Titan, TradingView, Investing.com and Unite.AI, several framing the strategy as using the listing to roll up BPO and outsourcing firms into AI-native operations. No pricing change was observed: the free tier, the $0.99 per-resolution overage and the sales-led Enterprise motion are all as last captured. Per-resolution economics have never been tested against quarterly public-market margin disclosure in this corpus; this is the case to re-read at first results. |
Counterexamples
- OpenAI · May 2026 — Prices inputs (per token) and seats, not resolved outcomes.
- Harvey · May 2026 — Per-seat enterprise pricing — value is billed by access, not by task completed.
- Anthropic · May 2026 — Pure per-token API + seat subscriptions; no outcome unit.
- 11x · Jun 2026 — Sells AI SDRs as 'digital workers' but prices on subscription seats, not per completed outcome — the agent framing without the outcome billing.
- Zendesk · Jul 2026 — A CX incumbent monetized AI through the SEAT, not the outcome, and did it aggressively: the Contact Center add-on rose from US$50 to US$83 per agent/month paid yearly (+66%), breaking the uniform US$50 price it had held across Copilot, the Workforce Engagement Bundle and Contact Center, and moved to its own pricing surface selling Minutes Blocks at US$33 and Quality Assurance at US$35 per agent/month. Copilot and the Workforce Engagement Bundle stayed at US$50; Support Team US$19, Suite Team US$55 and Suite Professional US$115 were unchanged. In the vertical where 74% of vendors bill resolutions, the largest incumbent raised a per-agent rate instead.
- Freshworks · Jul 2026 — Priced the AI per SESSION, not per outcome: the Freddy AI Agent add-on is now published at $49 per 100 bot sessions (~$0.49 each) identically on Growth, Pro and Enterprise, where a bot session is any unique end-user engagement — an engagement meter, not a resolution meter. The distinction matters commercially: a session that fails to resolve still bills. Seat prices unchanged ($9/$39/$59 per user/mo annual), and the $0 three-user Free plan was removed.
- Sequence · Jul 2026 — Agent framing with no outcome meter at all: Sequence shipped named agents for AR, reconciliation, invoice review, quote approvals, payment reminders and contract intake — reviewed in a Watchtower console, plus Sequence MCP for Claude/Cursor/ChatGPT — and priced none of them separately. Growth is still $799/month for startups under $1m annual revenue, with Core and Scale still bespoke platform fees assessed on projected billed revenue. The agents are a feature of the platform fee, not a billable outcome.
Trivia
-
Intercom's $0.99-per-resolution Fin pricing (March 2024) is the corpus's first live outcome-priced product, and its outcome definition — "no follow-up contact within 24 hours" — is the only published, time-bounded resolution standard in the corpus. That 24-hour window has become an informal reference point for the category: subsequent outcome-pricing discussions routinely invoke it when asking what "done" means for an AI agent in a support context.
-
Rox's "Agent Action" unit (June 2026) is the first outcome-pricing model in the corpus where cost scales explicitly with task complexity and model used — meaning two nominally identical Agent Actions can cost different amounts depending on which AI model handles them. This is a structural departure from Intercom's flat-rate resolution model and suggests outcome pricing will fragment into complexity-tiered outcome units as agent tasks become more heterogeneous.
-
11x is the corpus's sharpest illustration of the gap between agent framing and outcome billing: it sells AI SDRs as "digital workers" — the strongest possible outcome framing — but prices them on subscription seats, not per completed task. The contrast with Rox (same sales-automation use case, per-action billing) shows that outcome pricing is a deliberate structural choice, not an automatic consequence of positioning a product as an autonomous agent.
-
The 2026-06-08 re-test invalidated this trend's own "two adopters" framing in one stroke: per-resolution billing turned out to be the *default* in AI customer service, not a curiosity — Ada, Gorgias, Intercom, Kustomer, Lorikeet, Parloa, and Gladly all meter resolutions, meaning outcome pricing is simultaneously the rarest model corpus-wide and the category standard inside one vertical.
-
Outcome pricing and price gating move together: Ada, Parloa, Pixee, and Gladly all bill per resolved outcome AND hide the per-unit rate behind sales (Ada's ~$1–$3.50/resolution is only knowable from third-party data) — so the vendors most willing to tie price to delivered value are also the least willing to publish what that value costs, because the resolution rate is the negotiating lever.
-
Lorikeet (2026-06-07) is the first corpus vendor to *raise* outcome prices 2–3x while re-segmenting the unit by channel — evidence that as resolution billing matures it fragments (per-channel, per-complexity) rather than commoditizing toward a single flat rate, the opposite of what happened to per-token model pricing.
-
Outcome-based pricing stopped spreading and started concentrating. The tag has sat at exactly 24 of 353 companies (7%) across a 15-company corpus expansion and a 140-change window with no new adopter — yet inside AI customer service it is now the default unit of account: 14 of 19 `customer-service` companies bill resolutions (74%) against a 4% corpus base, a 17.3x lift. The rarest pricing model in the corpus is simultaneously the standard one in a single vertical.
-
`resolutions` is the corpus's tightest three-way marker. Of the 15 companies billing it, 14 are also tagged outcome-based, 14 sit in customer-service (Pixee, billing per fixed vulnerability, is the lone exception), and 14 have NO free tier — against a 61% corpus base for free tiers. If a vendor bills resolutions you can predict its category, its pricing model, and that you will not get to test the outcome definition for free.
-
The vendors most willing to tie price to delivered value are the least willing to publish it: 16 of the 24 outcome-based companies (67%) do not publish pricing, against a 29% corpus base. The 8 that do publish are almost all incumbents bolting an outcome onto a pre-existing public rate card — Intercom, Zendesk and Gorgias — rather than outcome-native vendors, so transparency here is inherited from the seat business, not chosen for the outcome one.
-
Sierra (2026-07-16) stretched the billable outcome from minutes to months. Horizon replaced its Agent Data Platform line and reframed the unit as a business goal pursued across days or weeks — "a mortgage, a claim, a renewal" — with the homepage headline moving from "Better customer experiences" to "Better outcomes. Built on Sierra." Every other outcome vendor in the corpus meters something that completes inside a single session; Sierra now meters something that may take a quarter.
-
Lorikeet (2026-07-21) put the corpus's first duration boundary on an outcome: the voice-resolution rate (1.50 credits on Start, 1.20 on Scale) is now footnoted "for resolutions up to 3 minutes long", with no published rate above that — so a four-minute call becomes a quoted item. List prices did not move (Start $1,500/mo for 18,000 credits/yr, Scale $4,000/mo for 48,000). The outcome unit is acquiring the same fine print that per-minute billing has.
-
Observe.AI publishes its outcome rate everywhere except its own website. observe.ai/pricing returned 404 again on 2026-07-22, while its AWS Marketplace listing prices Number of Interactions at $12.00 per unit and Minutes per Month at $4.80 per unit on a 12-month prepaid contract — with 24-month rates at exactly double ($24.00 and $9.60), so the longer commitment carries no posted discount at all. There is no per-agent seat dimension in the listing.
For buyers
Outcome pricing aligns spend with value — but scrutinise the definition: what counts as a 'resolution', who adjudicates it, and how the per-outcome price compares to your fully-loaded cost to do it another way. A loose definition can bill outcomes you wouldn't have paid a human for.
For vendors
Outcome pricing only works where the result is measurable, attributable and hard to game. You need event instrumentation, a defensible definition and dispute handling. Start with a narrow, checkable outcome (a resolved ticket, a completed task) before generalising.
Outlook — what to watch
Per-resolution is now the support default; the open question is whether the next agentic categories (sales, coding, finance, pharma) standardise on their own outcome unit the way support did. Watch the per-resolution price itself — as token deflation cuts the underlying cost, the $0.99 anchor has room to fall, and incumbents still on seat pricing will face pressure to add an outcome meter. The status would weaken if outcome adoption stalls as a share of the corpus, or if the support-resolution unit fragments so far (per-channel, per-complexity) that 'outcome' stops meaning a single comparable thing.
Bottom line
Outcome-based pricing has grown from a single live example to 24 of 338 corpus companies (about 7%) — roughly tripling in absolute adopters as the corpus expanded, while staying a minority model overall. Per-resolution billing is the de-facto standard for new customer-support AI, and the pattern holds wherever an agent performs a discrete, verifiable task.
FAQ
What is outcome-based pricing for AI?
Charging for the result the AI delivers -- a resolved support ticket, a completed sales task -- rather than for seats, messages or tokens. You pay when the job gets done.
How common is outcome-based pricing?
It's a minority model: 24 of the 338 corpus companies — about 7% — price on outcomes. That's up from a single live example when the pattern was first logged, so adoption has roughly tripled in absolute terms, but seat- and usage-based pricing remain far more common.
Who uses outcome-based pricing?
In customer support it's now standard — Ada, Gorgias, Gladly, Kustomer, Lorikeet, Intercom (Fin, $0.99/resolution) and Parloa all meter resolutions. Beyond support: Rox (per Agent Action in sales), Pixee (per merged code fix), Digits (accounting outcomes), Recursion (pharma milestones), and Mercor/Micro1 (per-task AI training data).
Where does outcome pricing work — and where doesn't it?
It needs an outcome that's measurable, attributable to the AI and hard to dispute. Support deflection (resolved or not) is the cleanest case, which is why support standardised on it first; open-ended generation doesn't qualify yet. Some agent vendors selling 'digital workers' (11x) still price on seats.