What Counts as a Resolution When AI Support Agents Bill Per Outcome?

Short answer
For most AI support agents, a billed resolution is a conversation where the agent gave an answer and no one asked for more help or a person, usually within a set time. It doesn't require the customer to confirm the problem was fixed. The exact rule belongs to each vendor: Zendesk adds an AI model's check of the transcript, Intercom deducts the charge if the customer returns to the same conversation, and Salesforce and Freshworks bill actions, conversations or sessions rather than resolutions.
Key takeaways
- A billed resolution is a metering rule, usually an answer sent and no further request within a set time, and it doesn't require the customer to confirm the problem was fixed.
- The rules differ on what starts the clock, what stops the charge and what happens when a customer returns, so per-resolution prices aren't directly comparable.
- Salesforce and Freshworks bill actions, conversations or sessions rather than resolutions, and Intercom and Front also bill certain hand-offs.
- Check the meter yourself with a test set of real conversations, a monthly sample of billed resolutions, repeat-contact tracking and CSAT.
- Get the definition in writing before you sign, including how reopened conversations, hand-offs and disputes are treated.
A billable resolution, or outcome, is the event an AI support vendor counts when it charges you for its agent. It is a metering rule built from signals a system can detect, such as an answer sent, no hand-off requested and a timer expired. A support leader usually means something else by resolved: the problem is fixed, the customer agrees, and they don't need to contact you about it again.
These definitions are public, and vendors publish them in their own help centers. A meter needs a rule it can run on every conversation, so this is a practical design, not a hidden catch. But one word covers different rules, so two per-resolution prices may not buy the same thing. We are Webify.AI. We build custom chatbots and connect them to these platforms, so we have an interest, and where a platform's own agent is the better buy, we say so. Every definition below was read on the vendor's own page on 5 November 2026, so check again before you sign.
Key facts at a glance
- Zendesk bills only Verified resolutions, which must pass an AI model's check of the transcript (Zendesk).
- Intercom charges for a customer who leaves after Fin's answer without asking for more help, and deducts the charge if they return to the same conversation for help (Intercom).
- HubSpot sets resolution status 72 hours after the customer's last response, and only a hand-off started by the customer stops a conversation from counting (HubSpot).
- Some billed outcomes aren't resolutions. Intercom bills a configured Procedure handoff at $0.99, and Front bills a Handoff level at $0.39 (Front).
How each vendor defines the unit it bills
In this table, "quiet period" means how long the vendor waits before it treats the conversation as finished. "None stated" means we didn't find one on the page we read, not that the product has none.
Swipe the table sideways to see every column.
| Vendor | Billing unit | Definition | Quiet period | Source |
|---|---|---|---|---|
| Zendesk AI agents | Automated resolution; only the Verified tier is billed | The agent responded, the customer did not request further assistance, and an AI model confirmed the request was satisfactorily resolved | Email: 72 hours after the first email. Messaging: 2 hours after the first message, up to 72 on request. Voice: at hang-up | Zendesk |
| Intercom Fin | Outcome, at most one per conversation | Resolution: after Fin's last answer, the customer confirms it or exits without requesting further help. Procedure handoff: a configured Procedure ends in a hand-off | No hour count stated | Intercom |
| HubSpot Customer Agent | Resolved conversation | The agent shared a content source or performed an action, with no hand-off started by the visitor. A qualified or disqualified lead also counts | 72 hours after the last visitor response | HubSpot |
| Salesforce Agentforce | Action, or conversation | Action: "a specific function that an AI agent executes on the platform". Conversation: not defined on the pricing page or rate card | None stated | Pricing; rate card |
| Gorgias AI Agent | Automated interaction, plus a ticket fee | A customer request fully handled by AI Agent, with no human agent needed | 72 hours | Billing; metrics |
| Freshworks Freddy AI Agent | Session | All interactions between an end user and the agent within a time window | None. The session lasts 24 hours from the start of a chat, or 72 hours from the first email | Freshworks |
| Help Scout | Resolution, once per session | An AI chat session resolved without human help: Scout answered, and the customer didn't ask for more help afterward | None stated | Help Scout |
| Front Autopilot | Conversation, at the highest level reached | Resolution: no human replies, and no follow-up message from a teammate within the window | 24 hours on Resolve, SMS, WhatsApp and custom channels. 72 hours on email and portal | Front |
For list prices, and the routes for connecting a custom chatbot to each platform, see our guide to which platforms integrate custom AI chatbots.
Zendesk: three tiers, one of them billed
Since 18 May 2026, Zendesk has sorted automated resolutions into three tiers (Zendesk). In an Assisted escalation, a person finished the job. In a Contained resolution, the agent handled the interaction and the customer didn't ask for clarification, give feedback or ask for a human. A Verified resolution also passes a check by a large language model (LLM), the type of AI model behind chat assistants, which reads the conversation to confirm the request was "satisfactorily resolved". Zendesk bills only the Verified tier (Zendesk).
Its timers run from the first message, not the last. Accounts that haven't moved to tiers stay on the older rules until they do. Zendesk's blog says a resolution "is often defined by a 72-hour quiet period" (Zendesk), so go by the help center article for your account.
Intercom Fin: the edge cases are written down
Fin costs $0.99 per outcome, charged at most once per conversation (Intercom). Intercom's outcomes page spells out the edge cases, which makes it a useful model for what to ask everyone else:
- A customer who "exits the conversation without requesting further assistance" after an answer is charged as an assumed resolution. Intercom suggests escalation rules so Fin hands off when it detects frustration, and those escalations aren't billed.
- If Fin asks a clarifying question and gets no reply, there's no charge.
- If a customer asks for a person, there's no charge, unless they then go inactive while Fin's collect information setting is on.
- If a teammate joins after Fin has answered and the customer hasn't asked for more help, it still counts as resolved.
HubSpot: 72 hours from the last response
Customer Agent costs $0.50 per resolution (HubSpot), paid as 50 credits (HubSpot). Its knowledge base has the working rule: a qualifying reply counts if "there is no qualifying handoff to a human agent within 72 hours of the last visitor response". Only hand-offs started by the visitor qualify, so manual assignment, including through workflows, doesn't by itself stop it counting. Once the status is set, "subsequent actions, such as messages from a live agent, a transfer request, or negative feedback, will not change the resolution status".
Salesforce Agentforce: activity, not outcome
Salesforce's two options don't try to measure a resolution. Under Flex Credits, each action is metered individually at 20 credits, with 100,000 credits for $500 (Salesforce), which Salesforce states as $0.10 an action (Salesforce). The other option is $2 per conversation, "regardless of the complexity of the task" (Salesforce). This is a different bargain: you pay for activity, and it's up to you to work out whether the activity fixed anything.
Neither the pricing page nor the Conversations rate card, dated December 2024, says when a conversation starts or ends. The rate card points to a usage-types article in Salesforce Help, which we couldn't read when we checked, so ask your account executive for that definition in writing.
Gorgias: a fee on top of the ticket
An interaction counts as automated if the customer "does not require the involvement of a human agent within 72 hours" (Gorgias). The price is $0.90 on annual plans or $1.00 on monthly plans, and each interaction "also counts as a helpdesk ticket and is billed accordingly" (Gorgias). If an agent replies within 72 hours, you pay only the ticket fee.
Freshworks: a session, whatever happens in it
"An AI Agent session represents all interactions between an end user and an AI Agent within a defined time window" (Freshworks). A pack of 100 sessions costs $49, which is $0.49 a session by our arithmetic. The definition has no outcome test, so measuring results is your job.
Help Scout and Front
- Help Scout charges $0.75 when Scout gives a response and the customer doesn't ask for more help after it. A greeting, or a follow-up question without a full answer, isn't billed (Help Scout).
- Front charges each conversation once, at the highest level reached: Triage $0.05, Handoff $0.39, Resolution $0.89. Its test watches your team: a conversation is resolved "if there is no follow-up message from a teammate" within the window (Front).
Why the definition changes your bill
A customer who gives up can count as resolved
Picture a customer who reads a wrong answer, closes the chat and phones you instead. Inside the chat, there was no request for a person and no reply from a teammate. To a rule based on no further request, that looks the same as a customer who got exactly what they needed.
Intercom says this directly in its FAQ. The HubSpot, Gorgias, Help Scout and Front pages don't discuss the silent customer in those words, but our reading is that silence doesn't stop a charge under them, because each rule depends on a request or a human reply. Zendesk's model check adds a filter, but it evaluates the text of the conversation, not what the customer did next.
Waiting for a confirmation would have its own problem, since the meter would miss every satisfied customer who simply left. So treat the billed count as a meter reading, not a quality score.
What stops the charge differs by vendor
Some rules mainly watch the customer, and others watch your team. On Intercom, HubSpot and Help Scout, the customer's request is what counts. On Gorgias and Front, a human reply inside the window changes the billing. If your team often steps into AI conversations, that difference affects your invoice.
How reopened tickets are treated, where it's documented
- Intercom: if the customer returns to the same conversation for more help, the resolution is deducted "even across billing periods". The page gives no time limit (Intercom).
- HubSpot: a reply after more than 72 hours in an email thread reopens the conversation and resets the window, and another resolution uses more credits (HubSpot).
- Gorgias: a reply after a period of inactivity becomes a new ticket, after 10 days on email and 3 days on chat. It may be billable if your help desk replies (Gorgias).
The other pages we read don't describe reopening, so ask. And none of them says it links a later contact on another channel back to the billed conversation.
What a support leader means by resolved
| What happened | What a per-outcome meter may record | What a support leader would call it |
|---|---|---|
| Right answer, customer leaves without comment | Resolved | Resolved |
| Wrong answer, customer gives up and phones | Resolved, on rules that rely on no further request | Not resolved, plus a second contact |
| Customer opens a new chat next day about the same problem | Two separate conversations | One problem, not fixed the first time |
| Agent completes a configured hand-off | A billable outcome at some vendors | Useful triage, not a resolution |
Dashboards inherit the vendor's logic too, so a resolution rate on one vendor's dashboard can't be compared directly with a rate on another's.
How to check it yourself
Test real conversations before launch
Pull real conversations across your main topics, including the difficult ones: complaints, refund disputes and cases that should go to a person. A couple of hundred is a workable start, by our rule of thumb. Have someone who knows the content mark the right outcome for each, then run the agent against them as far as the product allows, for example during Intercom's 14-day free trial or HubSpot's 14 days of free access. Score whether the answer was right, whether it should have handed off, and whether you'd call it resolved.
Sample billed resolutions every month
Get the list behind the invoice. Intercom says you can find every conversation that generated a billable outcome in its performance funnel, Zendesk's usage dashboard links to the tickets behind the count, and HubSpot has resolution reports. Ask other vendors for the same.
Draw a random sample of about 100. By our arithmetic, that gives a margin of error of about plus or minus 10 percentage points at 95% confidence: 1.96 x √(0.5 x 0.5 / 100) = 0.098. Ask a reviewer who didn't configure the agent to mark each as fixed, probably fixed, wrong or incomplete, gave up, or should have gone to a person.
Then work out your cost per real resolution. As an illustration only, if reviewers confirmed 80 of 100 billed resolutions at Intercom's $0.99, the cost would be $0.99 / 0.80, about $1.24 per confirmed resolution. The 80 is an invented example, not a benchmark, and the $1.24 is our arithmetic, not Intercom's.
Track repeat contacts
Export billed conversations with a customer ID and match them against any contact from that customer within seven days, on any channel, including phone. Seven days is our suggestion, so pick what suits your product. Compare the result with conversations your team handled on the same topics, because an agent may be taking the easier ones.
Watch CSAT next to the billed count
CSAT, or customer satisfaction score, comes from a short survey after a contact. Track it separately for billed AI conversations and human-handled ones. Expect a blind spot, since a customer who gave up may not answer a survey either. If billed resolutions rise while CSAT or repeat contacts get worse, the meter and your customers disagree.
Questions to ask before signing
- Is the definition part of our contract, and how much notice do we get if it changes?
- Does the clock start at the first message or the last response?
- What stops the charge: a customer's request, a teammate's reply, or either?
- If a customer gives up after an answer, or comes back later, are we charged?
- Are any hand-offs or non-support events, such as lead qualification, billed?
- Is every billed conversation counted, or is the number calculated from a sample?
- Can we export every billed conversation with an ID, and how do we dispute one?
- What other fees sit beside this unit, such as tickets or seats?
- What happens when the allowance runs out? Zendesk can pause the agent and route requests to people, HubSpot stops assigning it new conversations, and Freshworks' agent stops responding.
Does a custom chatbot avoid this?
Partly. A custom chatbot has no per-resolution meter, and you decide what resolved means. You pay for the build, model usage, hosting and maintenance instead; our planning estimates are in our cost guide. You still need the test set, the sampling and the repeat-contact tracking, and someone has to keep the system current, which our retainers cover.
With one channel, one help desk, modest volume and no data residency rule, a platform's own agent is often the faster route, and we'll say so on the first call, which is free. If you need one assistant across several systems, our AI chatbot integration page explains how we build the chatbot and connectors as one project.
Frequently asked questions
What counts as a resolution for an AI support agent?
It depends on the vendor. Most rules count a conversation where the agent answered and no one asked for more help or a person, often within a set window. Zendesk also requires an AI model to confirm the request was satisfactorily resolved. Salesforce and Freshworks bill other units instead.
Does a customer who gives up count as resolved?
They can. Intercom says a customer who leaves after Fin's answer without asking for more help is charged as an assumed resolution. Under Zendesk's tiers, the conversation is billed only if it passes the model's check.
Is the quiet period the same at every vendor?
No. HubSpot's 72 hours run from the last visitor response. Zendesk's run from the first email, and its messaging default is 2 hours from the first message. Front uses 24 hours on chat-style channels and 72 on email and portal.
What happens if a customer comes back after a billed resolution?
On Intercom, returning to the same conversation deducts the charge. On HubSpot, a late reply in an email thread reopens the conversation, and another resolution uses more credits. On Gorgias, a late reply becomes a new ticket. Other pages we read don't cover it, so ask.
Is per-resolution pricing better than per-action or per-session pricing?
Neither is better on its own. A per-resolution price ties the fee to a rule that resembles success, and a per-action or per-session price says nothing about results. Either way, measure outcomes yourself.
Sources
- Zendesk: About automated resolution tiers, checked 5 November 2026.
- Zendesk: Announcing changes to AI agent reporting, checked 5 November 2026.
- Zendesk: About the automated resolutions platform (prior to resolution tiers), checked 5 November 2026.
- Zendesk: Managing resolution allowances, checked 5 November 2026.
- Zendesk: Outcome-based pricing, checked 5 November 2026.
- Intercom: Fin AI Agent outcomes, checked 5 November 2026.
- Intercom: Pricing, checked 5 November 2026.
- HubSpot: Understand the customer agent, checked 5 November 2026.
- HubSpot: Analyze customer agent's performance, checked 5 November 2026.
- HubSpot: AI Customer Agent, checked 5 November 2026.
- HubSpot: Now you pay when the task is complete, checked 5 November 2026.
- Salesforce: Agentforce pricing, checked 5 November 2026.
- Salesforce: Conversations rate card, checked 5 November 2026.
- Salesforce: Flexible pricing blog post, checked 5 November 2026.
- Salesforce: Flexible Agentforce pricing announcement, checked 5 November 2026.
- Gorgias: How you're billed for using Gorgias, checked 5 November 2026.
- Gorgias: How metrics are calculated, checked 5 November 2026.
- Gorgias: Pricing, checked 5 November 2026.
- Freshworks: Purchase and manage Freddy AI add-ons, checked 5 November 2026.
- Freshworks: Freddy AI Agent sessions FAQs, checked 5 November 2026.
- Help Scout: AI Resolutions Pricing, checked 5 November 2026.
- Front: Automate complex workflows with Autopilot, checked 5 November 2026.
