Per-Resolution Pricing: Examples & Companies

What is it

Per-Resolution Pricing is a billing unit unique to AI customer-support products, where the vendor charges only when an AI agent resolves a customer issue without escalation.

It is the canonical way to implement outcome-based pricing in customer service: instead of paying per seat, per token, or per API call, the buyer pays per resolved conversation — and pays nothing when the AI hands off to a human. Intercom defined the category in 2023 by pricing its Fin AI Agent at a flat $0.99 per resolution, the same rate across every plan tier and for standalone Fin running inside Zendesk or Salesforce. The pitch is structural alignment: Intercom is the rare AI vendor whose financial model only makes money when the AI actually works.

Fifteen of the corpus’s companies now meter on resolutions, and the cluster spans the whole go-to-market spectrum. At one end sit self-serve vendors that publish the rate — Intercom Fin and Yellow.ai both expose $0.99 per resolution publicly. In the middle are hybrids that graft a resolution meter onto an existing platform — Zendesk AI, Gorgias, Forethought, Kustomer, and Gladly. At the far end are pure-outcome, sales-gated agents — Sierra, Decagon, Ada, Maven AGI, Lorikeet, and Parloa — plus Pixee, which extends the unit outside support entirely by billing per resolved vulnerability. The tightness of that cluster makes per-resolution the bellwether for the corpus’s outcome-based pricing trend: when one category converges this hard on an outcome unit, the rest of AI software is watching.

The hard part is definitional. A “resolution” is whatever the vendor says it is, and that single choice decides whether the vendor’s incentives align with the buyer’s or quietly diverge from them. Decagon is candid about it — its own glossary admits “what counts as a resolution” is genuinely contested for abandoned chats and partial answers. The rest of this page is mostly about who counts what.

How it works

The mechanic is simple — bill = resolutions × rate — but the design lives in two decisions: what counts as a resolution, and whether the resolution charge sits on top of a seat fee, a ticket fee, or nothing at all.

Worked example (Intercom Fin). A support team on the Advanced tier with 10 agents and 5,000 Fin resolutions in a month pays 10 × $85 in seat fees plus 5,000 × $0.99 in resolution fees — $850 + $4,950 = $5,800, of which 85% is the outcome line, not the seat line. That ratio is the whole point: as the AI deflects more, the seat fee shrinks to a floor and the resolution meter becomes the bill. You can model exactly this split in the Intercom pricing calculator.

Worked example (Yellow.ai free-then-meter). A team on Yellow.ai’s Free tier resolving 2,000 conversations in a month pays for none of the first 500 (included) and 1,500 × $0.99 = $1,485 on the overage — the same $0.99 unit as Intercom, but exposed on a self-serve tier with no seat fee underneath it at all.

Worked example (Lorikeet credits). A team on Lorikeet’s Scale plan (48,000 credits/year) resolving 3,000 chat tickets in a month draws 3,000 × 0.80 = 2,400 credits against the annual pool — and pays for none of the tickets the AI failed to resolve, because Lorikeet refunds resolutions the customer is unhappy with.

The “engaged”/“assisted” framing at Kustomer and Gladly is the loosest definition in the table: a conversation the AI touched can count, not only one it closed end-to-end. At the strict end, Sierra and Decagon bill only when the agent resolves an issue with no human help — and Zendesk AI sits apart again, billing automated resolutions only above a plan allowance, so successful deflection grows the bill above the included floor. For the mechanics of metering these tiers, see usage thresholds and overage alerting.

Companies using this

Fifteen corpus companies meter on resolutions. Fourteen are AI customer-support products — from self-serve, published-rate vendors like Intercom Fin and Yellow.ai, through hybrids like Zendesk AI, Gorgias, and Forethought, to pure-outcome sales-gated agents like Sierra, Decagon, and Maven AGI — and Pixee extends the same outcome logic to security, billing per resolved vulnerability rather than per resolved ticket. The table below shows each vendor’s resolution definition, published rate, and what (if anything) the resolution charge sits on top of.

Patterns observed

• The published rate has converged on $0.99. The two vendors that expose a self-serve rate — Intercom Fin and Yellow.ai — land on the exact same $0.99 per resolution, and Gorgias sits a dime under at $0.90. When independent vendors price the same unit inside a 10-cent band, the market has found a reference price, and Intercom’s 2023 number is clearly the anchor everyone else calibrates against.

• Most of the cluster keeps the rate private, and opacity scales with deal size. The self-serve vendors publish; the enterprise ones don’t. Sierra, Decagon, Ada, Maven AGI, Forethought, and Parloa all bill per resolution but quote every deal — third-party estimates put Decagon near a $95K/year floor and Parloa above $300K/year. Outcome pricing and price opacity travel together here: a buyer can know the unit is a resolution and still be unable to model spend before entering sales.

• The resolution sits on a different base at each vendor — and the base is shrinking. Intercom Fin layers it (optionally) on a seat fee, Zendesk AI grafts it onto per-agent Suite seats, Forethought blends it with a platform access fee, Gorgias puts it on a ticket-metered tier, and Lorikeet and Ada put it on nothing at all. The trajectory is clear: the fixed base is becoming a floor while the resolution meter becomes the bill, exactly the dynamic the outcome-based pricing trend tracks.

• Definition is the real product decision. Sierra and Decagon only bill when the AI resolves an issue end-to-end without human help — strict, buyer-friendly. Kustomer and Gladly bill on “engaged”/“assisted” conversations the AI merely touched — looser, and easier for the meter to fire. Zendesk AI bills only automated resolutions above an included allowance, and Lorikeet goes furthest the other way, refunding resolutions the buyer is unhappy with. The unit is nominally identical across all six; the firing rule is not.

• The strongest alignment claim wins the narrative. Every vendor in the cluster markets incentive alignment — Sierra’s “pay for a job well done” and Pixee’s “traditional tools profit when your backlog grows; we only profit when it shrinks” are the sharpest framings, while Lorikeet’s “you don’t pay for that ticket” guarantee is the strongest contractual version of it. Decagon even publishes a defense of resolution-based pricing in its glossary, turning the billing model itself into a sales asset.

Counterexamples & variants

The clearest variant is Pixee, which proves per-resolution isn’t a support-only unit: it bills per resolved vulnerability (calculated off annual SAST + SCA scanner findings), with unlimited developers included, framing the model as “pay per vulnerability resolved, not developer seats.” The outcome unit generalizes to any domain where an AI closes a measurable backlog item rather than consuming a seat — support tickets, security findings, or anything else with a countable definition of “done.”

The clearest counterexample to transparency is the sales-gated enterprise tier — Sierra, Decagon, Ada, Maven AGI, and Parloa. All meter on resolutions but publish no rate at all. Ada pivoted away from its old Core/Advanced/Pro tiers to an outcome model in 2023 yet shows only a “book a consultation” form; Parloa’s /pricing path 404s entirely; Sierra communicates its model as a philosophy (“pay for a job well done”) rather than a number. A buyer can know the unit is a resolution and still be unable to forecast spend before entering sales.

Kustomer and Gladly are the soft-definition variant — their “engaged”/“assisted conversation” meter charges for AI involvement, not strictly for an unescalated resolution, so the unit looks like per-resolution but can fire on conversations a human still finished. Zendesk AI is the incumbent-hybrid variant: it never fully commits to the outcome unit, keeping per-agent Suite seats ($19–$115/mo) for humans and billing automated resolutions only above a plan allowance — an outcome meter grafted onto a seat platform rather than replacing it, which lets Zendesk protect its seat base while answering pure-outcome challengers. And Forethought is the explicit-blend variant: rather than choose, it states outright that “our pricing model is a blend of platform access fees and an outcome-based pricing cost,” hedging predictability against value alignment inside a single contract.

What this means for buyers vs vendors

For buyers

Read the resolution definition before the rate. A “engaged conversation” at Kustomer or Gladly that fires whenever the AI touches a ticket can cost more than a $0.99 “unescalated resolution” at Intercom Fin that only fires on a clean deflection — the per-unit price is meaningless without the firing rule. Model your spend at your real deflection rate, not the vendor’s demo rate, and watch the base underneath: at Zendesk AI the resolution meter only kicks in above a plan allowance, so a rising deflection rate is exactly what grows your bill.

Treat the sales-gated vendors — Sierra, Decagon, Ada, Maven AGI, Parloa — as un-forecastable until you have a quote, and use the self-serve anchors (Intercom Fin and Yellow.ai at $0.99, Gorgias at $0.90) as your sanity-check reference price when you do. Watch for separate channel charges — voice generally carries higher economics than chat at nearly every vendor — the way any metered line shows up across usage invoicing and billing cycles. Ground your forecast in the introduction to usage-based pricing, settle on your value metric with choosing the right usage metric, and run the numbers in the Intercom pricing calculator.

For vendors

Per-resolution is the strongest alignment story in AI pricing — you only earn when you deliver — but the definition is a liability if it’s loose. A meter that fires on “assisted” conversations, as at Kustomer and Gladly, invites disputes the first time a buyer audits its bill against its actual deflection logs. The durable versions are the strict ones: Sierra’s and Decagon’s end-to-end-resolution rule, Intercom Fin’s unescalated-resolution rule, or Lorikeet’s refund guarantee.

Decide next whether to publish. The self-serve winners — Intercom Fin, Yellow.ai — turned a transparent $0.99 into a category anchor and a PLG on-ramp; the enterprise players — Sierra, Maven AGI, Forethought — keep the number private to tailor each deal to volume and channel mix. Either can work, but opacity costs you buyer trust and lengthens the sales cycle, so Decagon’s move — publishing a philosophy even while quoting privately — is a smart hedge. Whichever you choose, keep the resolution charge on top of a thin platform floor so you’re not fully exposed to a buyer’s deflection variance, and settle the resolution definition in the contract, not the marketing page — see the outcome-based pricing trend for how the category is hardening these definitions.

Explore this theme in the knowledge graph →