The answer exists, but not in writing
How you'll know: Agents answer it the same way every time, and there's no article for it.
Fix: Write the article or snippet. It's the cheapest fix there is, and Fin picks it up straight away.
Intercom's chatbot today is Fin, an AI agent that answers from your help center and hands off when it can't, plus Workflows for the rule-based paths Custom Bots used to run. You pay for Fin by resolution, so the number that matters is how many questions it actually closes. The biggest lever on that number is rarely the bot's settings. It's what data Fin can reach.
Fin reads your help center and other content you give it, answers in your Messenger, email or other channels, and passes to a person when it can't resolve the question.
Workflows are the rules: routing, buttons, collecting details, and when Fin steps in. They replaced Custom Bots.
Data connectors, formerly Custom Actions, let Fin and Workflows call your own systems mid-conversation.
Settings › Integrations › Data connectorsMost teams use both: Workflows to collect and route, Fin to answer. Neither knows a customer's order or invoice until something connects them to where that lives.
Intercom charges $0.99 for each resolution and counts at most one outcome per conversation. Seats are billed separately.
| Conversations Fin replied to | 2,000 |
|---|---|
| Handed to a person (not a resolution) | −700 |
| Only a greeting, or Fin asked a question nobody answered (not billed) | −120 |
| Customer came back later for more help (resolution deducted) | −60 |
| Billable resolutions | 1,120 |
| At $0.99 each | $1,108.80 |
A resolution means the customer said the answer helped, or left without asking for more. Look at the 700 handoffs: every one you turn into a real answer saves an agent's time and costs less than that agent.
Pull a week of Fin handoffs and sort them. They usually fall into these groups.
How you'll know: Agents answer it the same way every time, and there's no article for it.
Fix: Write the article or snippet. It's the cheapest fix there is, and Fin picks it up straight away.
How you'll know: "Where's my order", "why was I charged twice", "when does my plan renew": Fin passes these because it can't see your order or billing system.
Fix: A data connector that looks up the order, invoice or subscription during the chat. This is where n-frames does most of its Intercom work, and the endpoint side is covered on our Intercom API page.
How you'll know: The customer wants to cancel, change an address or resend an invoice, and only an agent can click the button.
Fix: Let Fin or a Workflow call that action through a connector, with conditions, so an order that's already shipped can't be cancelled.
How you'll know: The agent asks the customer for the order number Fin already collected.
Fix: Store what the connector fetched on the conversation, so the agent sees it on arrival.
Fin is Intercom's AI agent. It answers customer questions from your help center content and any data it's connected to, resolves what it can, and hands the rest to a person.
Fin is priced at $0.99 per resolution, charged at most once per conversation, on top of your Intercom seats. Running Fin in front of another helpdesk has a monthly minimum of resolutions. The ledger above shows how a month adds up.
Not on its own. It starts with your content and what's in the conversation. To answer from account, billing or order data, it needs a data connector (what Intercom used to call a Custom Action) that calls your system during the chat.
It takes a real share of first replies, especially repeat questions your help center already covers. It's built to hand off what it can't resolve, so you still need people, usually fewer of them on the repetitive work.
No. Custom Bots were scripted, rule-based flows, and Workflows replaced them. Fin writes its answers from content and data instead of following a fixed script, and Workflows decide when it steps in.
It's as accurate as what it can reach. Up-to-date articles and live account data through connectors give reliable answers. Questions needing information it can't see are where it either hands off, correctly, or answers too generally.
Your developers can, or n-frames can: we build the endpoint in your order, billing or account system, set it up as a connector in Intercom, and test it against the questions Fin was handing off. See also Intercom integrations and Intercom and Salesforce.