Intercom Fin: How to Hand a Conversation to a Human

Almost every complaint about an AI support agent is really a complaint about handover. The bot answered when it should have stepped aside, or it stepped aside so eagerly that nothing got automated. Fin gives you fine control over both directions, but the controls are spread across a few screens.

This guide covers what triggers a handover, where each control lives, how to collect useful information on the way out, and what a handover does to your bill.

Get the mental model right and the rest is fast. Intercom frames the split cleanly in their documentation: escalation rules and guidance decide when escalation happens, and workflows decide what happens after it.

What actually triggers a handover in Fin?

There are several triggers and only some of them are things you configure. Intercom's documentation describes both automatic behaviours and settings you control, and the automatic ones catch people out.

• A teammate replies. If a human sends a reply in the conversation, Fin stops responding and control passes to the person. No setting required.
• The customer asks for a person. Intercom's documentation notes that if a customer's first message is a request along the lines of wanting to talk to someone, Fin follows the handover path rather than trying to generate an answer.
• Escalation guidance fires. Natural language instructions you write about when Fin should step aside.
• An escalation rule matches. A condition on customer or conversation data, for example a plan attribute or an order value.
• Negative feedback. Per Intercom's workflow documentation, a customer giving negative feedback follows a handover path that you configure in the workflow.

That last one matters. The path exists whether or not you have designed it, so an unconfigured workflow is still a workflow, just one nobody chose.

Where do you configure escalation?

Both escalation controls sit in the same place: Train, then Escalations. Intercom's documentation splits them into guidance and rules, and the difference is whether you are describing a situation or matching a data field.

Guidance is more flexible than it first looks. Intercom's documentation describes instructing Fin to escalate straight away, to offer escalation as an option, to ask a follow up question before escalating, or to avoid escalating for a given situation. That fourth option is the one teams forget, and it is often the one that lifts automation rate.

Creating an escalation rule

1. Open Train, then Escalations — Both guidance and rules live on this screen. Read what is already active before adding anything, since contradictory instructions are a real failure mode.
2. Click + New — Start a new rule rather than editing an existing one. Intercom's guidance best practice documentation recommends one focused instruction per entry instead of stacking conditions.
3. Build the condition — Rules match on data attributes, and Intercom's page describes conditions drawn from Users, Company and conversation data, including custom attributes you have defined.
4. Choose the audience — Use the Audience dropdown. Intercom notes the default audience is Everyone, so leaving it alone applies the rule to your whole customer base.
5. Choose the channels — Use the Channels dropdown to limit the rule to the surfaces you mean. A rule written for chat can behave oddly on email.
6. Save, then enable — These are two separate actions in Intercom's described flow. A saved rule that was never enabled does nothing, which is the most common reason a new rule appears to be ignored.

How do you control what happens after the handover?

That is the workflow layer. Go to Fin AI Agent, then Workflows, and look for the step Intercom calls Let Fin handle. Everything about the moment Fin steps aside is configured on that step.

Intercom's documentation describes several options on that step, and they are worth going through individually rather than accepting the defaults.

• Set expectation for human support. Tell the customer what happens next so a handover does not feel like being dropped.
• Ask for more information before handover. Includes an option to collect more information when a customer asks to speak to the team, so the human inherits context instead of starting cold.
• Follow up behaviour. Intercom's page describes options for Fin confirming whether the user still needs help, Fin offering escalation, or no follow up at all.
• Email specific settings. A configurable wait duration for follow ups, plus CC behaviour and reply scope options.
• Ask for conversation rating. CSAT collection, available across channels.
• Auto-close abandoned Fin conversations. An inactivity timer with configurable closure scenarios.
• Pass to a reusable workflow. Hand the conversation to another automation instead of duplicating routing logic in every branch.

Because escalation can be branched, you can route by reason. Intercom's documentation describes creating workflow branches based on which escalation was applied, so a billing escalation and a bug report do not have to land in the same queue.

Can Fin pause and ask a human mid-task?

Yes, and this is a genuinely strong piece of the product. Intercom documents human in the loop approvals for Fin Procedures, where Fin pauses part way through and waits for a teammate to approve before continuing. Refunds and compliance sensitive requests are the examples their documentation gives.

Per that page, the step is called Loop in teammate and agent and is inserted in the procedure editor. You write the question the teammate sees, define what response to collect, assign a reviewer, and set a wait time with an escalation owner in case nobody answers. The reviewer sees a card asking for their decision, and can either submit a response or take over the conversation outright.

Does a handover cost you anything?

Intercom's outcomes documentation, read on 17 August 2026, lists escalation as a non-billable outcome. Spam is also listed as non-billable. Billable outcomes on that page include resolutions and procedure handoffs.

That distinction has a real design consequence. A procedure that you configured to end in a handoff is treated differently from a plain escalation, so the way you build the path changes the bill, not just the customer experience. Check the current definitions on Intercom's own pages before you model your costs, because these categories have been revised.

There is a different way to think about escalation

Step back and look at what you have built by the end of this. A trigger layer, a destination layer, branch logic per escalation reason, an approval step for anything risky, and a set of rules that have to stay consistent with each other as your business changes.

This kind of set-up work is exactly what we removed at Sistava. Instead of wiring triggers to destinations and keeping the branches consistent, you hire an AI Employee and tell it in plain English what it should handle and when it should come to you. It asks when it is unsure rather than following a rule you wrote six months ago, and there is no flow to maintain.

To be fair to Fin, explicit rules have a real advantage: they are auditable. If a regulator or a finance lead asks why a refund conversation went to a human, a named rule is a better answer than a judgement call. Pick the model that matches how much you need to explain.

What goes wrong most often?

Four things account for most broken handover setups, and all four are quick to check before you start debugging anything deeper.

• Saved but not enabled. Intercom's flow has save and enable as separate actions on escalation rules.
• Audience left as Everyone. A rule meant for enterprise accounts applied to every customer, because the dropdown defaults to everybody.
• Contradictory guidance. Intercom's own best practice page warns about conflicting instructions, and with up to 100 active pieces that is easy to do by accident.
• No handover path designed. The negative feedback path exists by default and needs configuring in the workflow, or the conversation lands somewhere nobody chose.

Sources

• Intercom Help, Manage Fin AI Agent's escalation guidance and rules
• Intercom Help, Use Fin AI Agent in Workflows
• Intercom Help, Human-in-the-loop approvals for Fin Procedures
• Intercom Help, Fin AI Agent outcomes
• Intercom Help, Hand over Fin AI Agent conversations to another support tool
• Intercom Help, Fin Guidance best practices