← All AI customer support software
What Counts as a Resolution?
Per-resolution pricing sounds like the fairest deal in software: you only pay
when the AI solves the problem. But every vendor decides for itself what
"solved" means, and the answer changes your bill, your dashboard and every
resolution rate you read in a case study. We read the billing docs of
13 AI support vendors to see what each one actually counts.
Checked against each vendor's own pricing page and help center in October 2026.
Five things we found
- Silence usually counts as success. Fin bills an "assumed resolution" when the customer leaves without asking for more help; HubSpot bills when nobody asks for a human within 72 hours; Gorgias bills when no agent replies within 72 hours. A customer who gives up looks the same as a happy one.
- Only Zendesk documents an LLM check before billing. Its AI agents bill only "Verified" resolutions, where a model reads the transcript and confirms the request was resolved. Ada uses a similar LLM check for its resolution metric but does not publish how it bills.
- Your dashboard and your invoice can disagree. Since May 2026 Zendesk's resolution rate includes Contained resolutions that it does not bill. Intercom bills Procedure handoffs as outcomes but leaves them out of Fin's automation rate.
- Many "per resolution" products bill per conversation. Freshdesk bills per session, Tidio per conversation with at least one AI reply, Agentforce per action or per conversation, and Ada and Decagon say most customers pick per-conversation pricing.
- Few vendors say what happens when the customer comes back. Intercom refunds the outcome; HubSpot can bill a second resolution; most of the rest do not document it.
Vendor by vendor
"Not documented" means we could not find it on the vendor's public pages, not
that the vendor has no rule: ask for it in writing.
- Billed per
- Outcome (a resolution is one kind of outcome)
- Published price
- $0.99 per outcome, $49/mo minimum including 50; voice $1.99
- Time window
- 24 hours of customer silence after Fin's last answer
- Customer comes back
- Refunded if the customer comes back to the same conversation, even in a later billing period
What counts: After Fin's last answer, the customer either confirms it helped or leaves without asking for more help (an "assumed resolution"). Replying to a greeting does not count. Asking for a human is free, but a handoff that a configured Procedure tells Fin to make is billed.
Sources:
Outcomes article, 24-hour rule, Pricing
- Billed per
- Automated resolution
- Published price
- $1.50 committed, $2.00 pay-as-you-go (pricing page; the help center describes a per-seat dollar allowance instead)
- Time window
- Ends 2 hours after the first message on messaging (configurable up to 72), 72 hours after the first email, at hang-up on voice
- Customer comes back
- Not documented; disputes go to Zendesk support
What counts: The AI answered, the customer did not ask for more help, and an LLM reviewing the transcript confirms the request was resolved (a "Verified resolution"). Conversations that fail that check are "Contained" and not billed.
Sources:
Resolution tiers, Reporting change, Pricing
- Billed per
- Three models: Flex Credits per action, per conversation, or per Help Agent resolution
- Published price
- $0.10 per action (20 credits), $2 per conversation, or $2 per Help Agent resolution
- Time window
- Conversations: 24 hours on most channels. Help Agent: a resolution covers up to 2 hours of chat or 10 minutes of voice; longer counts as more than one
- Customer comes back
- Not documented
What counts: Credits and conversations are billed whatever the outcome. A Help Agent resolution needs at least two turns, final feedback that is not explicitly negative, and an ending without escalation to a human (or explicitly positive feedback).
Sources:
Pricing, Conversation windows, Flex Credits and resolutions
- Billed per
- Resolution (paid in HubSpot credits)
- Published price
- $0.50 per resolution (50 credits); voice $0.50 per minute
- Time window
- Status is set 72 hours after the last visitor reply, then locked
- Customer comes back
- An email reply after 72 hours reopens the conversation, and another resolution can be billed
What counts: The agent shared a content source or took an action, and the visitor did not ask for a human within 72 hours of their last reply. Negative feedback on its own does not stop the charge, and a manual reassignment by your team is not a handoff. Qualifying a lead also counts.
Sources:
How resolutions work, Credits
- Billed per
- Automated interaction
- Published price
- $0.85 to $1.00 per interaction depending on plan, plus the helpdesk ticket fee
- Time window
- 72 hours with no human involvement
- Customer comes back
- A reply after a long gap (more than 10 days on email) opens a new ticket, which may be billed
What counts: A request fully resolved without a human agent, by AI Agent or by another automation such as Flows or Order Management. If an agent replies within 72 hours, it is billed as an ordinary ticket instead.
Sources:
Pricing and FAQ, How you are billed, Metric definitions
- Billed per
- Full resolution
- Published price
- About $1.20 on the $14,400/yr Starter plan, falling toward a $0.55 floor; 12-month commitment
- Time window
- Not documented
- Customer comes back
- Not documented
What counts: Yuma handled the ticket end to end and nobody, customer or agent, got involved. Counted once per ticket; escalations are free.
Sources:
Pricing
- Billed per
- Resolution
- Published price
- $0.99 chat, email or SMS and $1.50 voice on the Start plan ($2,100/mo, annual)
- Time window
- Not documented
- Customer comes back
- Not documented
What counts: Charged only for tickets successfully resolved; escalations and tickets you are unhappy with are free. Lorikeet says the customer holds a veto over what counts.
Sources:
Pricing
- Billed per
- AI conversation
- Published price
- About $0.20 per conversation, $200/mo minimum
- Time window
- 72 hours (resolution guarantee only)
- Customer comes back
- A reopen within 72 hours means no resolution under the guarantee
What counts: Its pricing page meters conversations the AI handles, while its llms.txt says "AI-resolved conversation". Its resolution guarantee is stricter: closed by the AI with no human intervention and not reopened within 72 hours.
Sources:
Pricing explainer, Guarantee terms
- Billed per
- Session
- Published price
- $49 per 100 sessions
- Time window
- 24 hours from the start of a chat; 72 hours from the first email
- Customer comes back
- Not documented
What counts: Not tied to an outcome: all interactions between a customer and the AI agent inside the window make up one session.
Sources:
Sessions
- Billed per
- Lyro conversation (pay-per-resolution on Premium)
- Published price
- From $32.50/mo for 50 conversations; Premium by quote
- Time window
- 15 minutes (resolution-rate metric only)
- Customer comes back
- Coming back to the chat later is still one conversation
What counts: Any conversation with at least one Lyro reply. Premium's pay-per-resolution definition is not published. For its quoted resolution rate, Tidio counts an interaction the customer considers successful with no follow-up within 15 minutes.
Sources:
Pricing and FAQ, Resolution rate
- Billed per
- Conversation by default; resolution-based pricing for some enterprises
- Published price
- Not published
- Time window
- Idle conversations end after 24 hours on chat and social, 72 hours on email
- Customer comes back
- Not documented
What counts: The billing definition is not published. For its Automated Resolution metric, an LLM classifies each conversation as resolved only if the answer was relevant, accurate, safe and contained.
Sources:
Pricing models, Automated resolution
- Billed per
- Per conversation (most customers) or per resolution
- Published price
- Not published
- Time window
- Not documented
- Customer comes back
- Not documented
What counts: Not published. Decagon itself warns that an upset customer who leaves can end up counted as a resolution.
Sources:
Pricing models
- Billed per
- Outcome, agreed per contract
- Published price
- Not published
- Time window
- Not documented
- Customer comes back
- Not documented
What counts: Criteria are agreed up front in each contract. Unresolved and escalated conversations are free "in most cases"; greeter or routing work may be priced per conversation instead.
Sources:
Outcome-based pricing
Resolution, automation, containment, deflection
These four rates get quoted as if they were interchangeable. They are not,
and the same deployment can score very differently on each.
- Resolution rate
- Resolved conversations as a share of the conversations the AI handled. The denominator leaves out everything routed straight to people, which is why it is usually the highest number a vendor quotes. Microsoft's Copilot Studio, for example, counts a session as resolved when the user is asked "Did that answer your question?" and either says yes or does not respond (source).
- Automation rate
- Fully automated conversations as a share of all support volume. Intercom defines it as involvement rate times resolution rate (source), and Gorgias divides automated interactions by all automated plus human-handled tickets (source). Some vendors use the term more loosely for any ticket the AI touched.
- Containment rate
- Self-service contacts completed without escalation to a live agent, as a share of self-service contacts started. A contact center metric, mostly quoted for voice bots and IVR (NICE definition).
- Deflection rate
- Requests handled in self-service that agents would otherwise have handled. Microsoft notes that each organization defines it differently, and some fold abandoned sessions into the calculation (source). Treat it as the weakest of the four.
See how these rates compare across published case studies on the
AI customer support benchmarks page.
What to get in writing before you sign
- The exact rule for a billable resolution, including whether a customer who goes quiet counts.
- The time window, and whether it runs from the first message or the last.
- What happens when the customer comes back, and whether a reopened conversation is refunded.
- Whether handoffs your own rules trigger are billed, and whether greetings, spam and tests are excluded.
- Whether the resolution rate on your dashboard uses the same definition as your invoice.
- How to dispute a charge, and whether you can export the per-conversation billing status.
Then put your own volumes through the cost calculator,
and compare published rates on the pricing table. A higher
resolution rate under per-resolution pricing means a bigger bill; whether that is
a good trade depends on what a human-handled ticket costs you.
A vendor that is missing here, or a definition that has changed?
Send us the link and we will update the table.