Resolution rate is the share of customer conversations that end with the customer’s problem solved: resolved conversations divided by all conversations in the same period. Deflection rate is the share of customers who looked for help and never opened a conversation with a person, usually because a help center article or a bot answered first. Automation rate is the share of conversations an AI agent or a workflow handled from start to finish. The three are often quoted as one number. They are separate measurements, and one month of tickets gives a different figure for each.
Vendor-claimed resolution rates do not compare with each other, because every vendor defines a resolution its own way and many of them bill by that definition. Queue Bench does not rank resolution rates. It ranks the published price behind them on the AI price per outcome leaderboard, with each vendor’s billed unit printed next to its price.
Three rates, three questions
| Rate | Formula | Counts as a success | What it misses |
|---|---|---|---|
| Resolution rate | Resolved conversations divided by all conversations | The problem is solved, by a person or by AI | Reopened conversations, unless you count them back |
| Deflection rate | Help sessions with no conversation divided by all help sessions | The customer did not reach a person | Whether the customer got an answer or gave up |
| Automation rate | Conversations handled end to end by AI divided by all conversations | No person touched the conversation | Whether the answer was right |
A stricter version of the first, first contact resolution, counts only conversations solved in the first exchange. AI resolution rate needs a stated denominator. AI-resolved conversations divided by the conversations the AI touched gives a higher figure than the same count divided by every conversation whenever the AI skips some, so ask which one a vendor uses before reading its number.
How each vendor defines a resolution
These are the billed units behind the AI price per outcome leaderboard, with each vendor’s own wording where the wording matters.
| Vendor | Billed unit | What counts | Published price |
|---|---|---|---|
| Intercom (Fin) | Outcome | A resolution (no further help requested after Fin’s last answer), a Procedure handoff, a disqualification or a self-serve routing; at most one per conversation; nothing if Fin cannot answer or the customer asks for a human | $0.99 per outcome |
| Zendesk | Automated resolution | A request resolved by the AI agent without escalation to a human agent, checked by a language model when the conversation ends; counted per conversation | $1.50 committed, $2.00 pay as you go |
| HubSpot | Resolution | Support provided by the customer agent with no handoff to a human rep for 72 hours; qualifying a lead also counts | $0.50, paid in HubSpot Credits |
| Help Scout | Resolution | The customer gets an AI Answers reply and does not escalate, search the knowledge base, ask more questions or say they need more help; one per conversation | $0.75 per resolution |
| Jira Service Management | AI agent resolution | The AI agent handles an inquiry from the first question to a confirmed resolution without escalating, judged after 24 hours of inactivity | $1.00 per resolution |
| LiveChat | AI resolution | The AI directly solved a customer question; several questions in one chat count as one | $49.50 per pack of 50 |
| eDesk | Outcome | The AI Agent answers the customer or takes a real action for them; once per ticket | $0.99 per outcome |
| Lorikeet | Resolution | A ticket resolved successfully; unresolved tickets and escalations to a person are not charged | $0.99 per chat, email or SMS resolution on Start, plus a platform fee |
| Freshdesk | Session | Every interaction inside 24 hours of chat or 72 hours of email, billed whether or not the AI resolves anything | $49 per 100 sessions |
| Salesforce (Agentforce) | Conversation | The same $2 is labelled both Conversations and Help Agent Resolutions, and no end point for a billable conversation is stated | $2 per conversation |
Side by side, the definitions split on three points.
- When the clock stops. Zendesk judges at the end of the conversation, Jira Service Management after 24 hours of inactivity, and HubSpot after 72 hours without a handoff.
- What silence means. Intercom counts a conversation where the customer asks for nothing more after Fin’s last answer. Help Scout also rules out a customer who searched the knowledge base or asked a follow-up question.
- What else is billed. Intercom bills Procedure handoffs and routings at the same $0.99 as resolutions. HubSpot counts a qualified lead. Freshdesk bills a session whether or not anything is resolved.
Some vendors tie a rate to the price directly. Tidio sells a Premium plan with pay-per-resolution billing and a guaranteed 50% Lyro resolution rate, priced on request. Sierra says customers pay only when the software achieves specific, valuable outcomes, and publishes neither a price nor a definition of a billable outcome. A claimed rate carries its vendor’s definition with it, so two vendors’ rates cannot be set side by side.
How billing per resolution ties to the rate
Under per-resolution pricing, the vendor’s resolution count and the vendor’s invoice are the same number. Every conversation the vendor’s definition calls resolved is billed, and every one it does not is free. A looser definition raises the claimed rate and the bill together.
That makes the bill easy to estimate from your own volume. At 2,000 AI-resolved conversations a month, the published prices come to up to $1,000 on HubSpot before its included credits, $1,500 on Help Scout, $1,980 on Intercom, $2,000 on Jira Service Management and $3,000 on Zendesk at its committed rate, all before seats. The October 2026 AI price run also carries the bills at 500 and 10,000 conversations.
Session pricing breaks the link between rate and bill. Freshdesk and Freshchat bill $49 per 100 sessions whatever the AI resolves, so at the same traffic a low resolution rate costs as much as a high one. Platform fees sit on top for some vendors: Lorikeet adds $2,100 a month on its Start plan, paid annually, and Siena AI adds $750 a month to its $0.90 per automated ticket.
How to measure resolution rate on your own tickets
- Write the definition down before the first report: what closes a conversation, what reopens it, and how long you wait before calling it resolved.
- Count a reopened conversation back as unresolved if it reopens inside a window you set and keep.
- Report AI resolutions and human resolutions separately, over the same denominator.
- Read a sample of AI-resolved conversations every month. A customer who gave up can look the same as one whose problem was solved.
- Compare vendors during a trial with your own definition applied to both, and track the trend against your own past months.
What Queue Bench ranks instead
A resolution rate measured on one company’s tickets says little about the next company’s. A published price is the same list price for every buyer on a given date, so that is what the leaderboard ranks. AI price per outcome groups vendors by comparable unit, and vendors that publish no price, such as Sierra and Decagon, sit below the ranked rows as not published. Seats are priced separately on the seat cost leaderboard, under the same testing rules.
Resolution rate sits with first response time and CSAT among the customer service metrics that only a team’s own data can show. For buyers comparing AI agents, the rankings of AI help desk software and AI customer service software both order vendors by published AI price per outcome.