Use this article to set up Human in the Loop with escalation rules or escalation guidance, answer Fin's questions as a teammate, understand how Human in the Loop affects billing and reporting, or troubleshoot why Fin did or didn't ask for help.
Until now, Fin had two options: resolve the conversation, or hand the whole thing over. With Human in the Loop, there’s a third. When a conversation needs a review, some information, or an approval, Fin asks a teammate for exactly what it’s missing, with the context and a draft reply ready to go. Once the teammate answers, Fin writes and sends the final reply and owns the resolution. For your customer, it’s one continuous conversation.
Note:
Human in the Loop is currently in beta and available on email and chat with the Intercom helpdesk.
In Escalation Rules, Human in the Loop is called Ask for input; on escalation guidance, it's the Ask a teammate for input before escalating setting; in the inbox view it's Needs teammate input (in the Fin for Service folder); in reporting it's Human in the loop.
Ask for input isn't available in AU workspaces (workspaces hosted in Australia), because confidence scoring isn't supported in that region. This is expected behavior, not a permissions or setup issue. In AU workspaces, the Fin will section doesn't appear in the escalation rule editor. See the FAQ at the end of this article.
Why use Human in the Loop?
Most teams hold Fin back from some conversations for one of three reasons. Human in the Loop is built for all three:
Sensitive topics. Refunds, payments, account changes, anything you don't want sent without a person's sign-off. Fin drafts the reply and asks a teammate to approve the decision.
Complex conversations. Exceptions, judgment calls, several questions at once. Fin drafts what it can and asks a teammate for the one decision Fin can't make.
Data Fin can't reach. The status of an order, a booking or a payment that lives in a system Fin isn't connected to. Fin asks a teammate to check, then finishes the answer with the real detail.
What your team gets out of it:
One decision instead of the whole conversation. These are conversations your team would have received anyway as a handover. Now a teammate answers one question and Fin does the rest.
Your team stays in control. A teammate can take over at any time, with the full context.
Fin needs your team less over time. Every answer shows you what Fin is missing. Add the fact to Fin's knowledge, or connect the data, and lower the threshold as Fin's drafts prove right. Human in the Loop is the bridge from handing a topic over to Fin resolving the topic itself.
Human in the Loop in escalation rules and guidance vs in Procedures
This article covers Ask for input on escalation rules and escalation guidance: you set the conditions or describe the situation, and Fin works out what to ask. Procedures have their own Human in the Loop step, where you write the exact question at the exact point in the procedure. If you know precisely when a teammate is needed, use a procedure. To catch everything else on a topic, use an escalation rule with Ask for input selected, or escalation guidance with Ask a teammate for input before escalating turned on.
If a procedure triggers, the procedure handles the conversation and Human in the Loop on your escalation rules and escalation guidance doesn't apply. Ask for input only steps in when Fin is writing its own answer. So if a teammate gets asked on a conversation you expected a procedure to handle, check the procedure's trigger: the procedure didn't run.
How does Human in the Loop work?
How does Fin decide when to ask a teammate?
For every answer Fin is about to send, Fin assigns a confidence score. Teammates see it as a badge on Fin's draft, from Low to Very High.
If the conversation matches your rule and Fin's confidence is under the threshold you set, Fin holds the answer. The threshold is how sensitive the rule is: higher means Fin asks more often, and at High Fin checks every matching answer. Human in the Loop only applies to answers. If Fin's reply is an offer to hand over (for example, because the customer is upset or asks for a person), Fin hands over as usual and doesn't ask a teammate.
Fin decides whether to ask based on its confidence in the specific answer it's about to send, not on whether you have content on the topic. Fin can have a relevant article and still not be confident it applies to this customer's situation, for example when the customer asks Fin to undo an action.
What does the teammate see when Fin asks a question?
Fin assigns the conversation to the teammate or team in the rule and asks a question in the Inbox. Fin works out the question itself, from the customer's message, Fin's draft and Fin's knowledge: for example a booking reference Fin can't look up, or a sign-off on a draft Fin is unsure about. You don't write the questions.
The teammate sees Fin's original draft, Fin's confidence (click Show reasoning for why), and an Answer Fin's question panel with exactly what Fin needs. Conversations waiting for an answer are listed under Fin for Service › Needs teammate input in the Inbox.
What does my customer see while Fin waits for a teammate?
While Fin waits for a teammate, your customer sees:
On chat: sees "Let me check with the team. I'll be back shortly." The customer can keep typing; each new message is checked the same way.
On email: gets nothing until Fin replies.
What can the teammate do when Fin asks a question?
In the Answer Fin's question panel, the teammate can choose one of three options:
Approve original solution: Fin sends the draft as written.
Give Fin an answer: type the fact or decision Fin needs and click Submit. Fin writes a new draft using your answer.
By default, Fin shows you the new draft and you choose:
Approve: Fin sends the new draft.
Ask Fin to iterate: tell Fin what to change, and Fin writes another draft.
Other options: for example, take over.
If the rule's review cycle is set to Get back to customers right away, Fin sends the new draft straight away, without showing it to you first.
Other options › Take over from Fin: pick a reason, and the conversation is assigned to the teammate, with the full context. Fin stops replying.
When the answer is sent, Fin unassigns the teammate and carries on with the conversation. If the customer asks something else, Fin checks Fin's next reply the same way, so Fin may ask again. A new question on the same conversation goes straight to the teammate who answered the first one.
What happens if nobody answers Fin's question?
Important: If nobody answers before the timeout, Fin does what the rule's Timeout setting says:
Continue the conversation: Fin sends its latest draft (the original draft, or the new draft if a teammate already gave input) and carries on with the customer.
Handoff: Fin hands the conversation to your default assignee, a workflow, a team or a teammate.
Wait indefinitely: the conversation waits until a teammate answers.
What does my customer see on chat at each stage?
The table below shows what your customer sees in the chat Messenger at each stage of a Human in the Loop request, depending on whether the teammate approves, answers or takes over, or the timeout runs out.
When | Your customer sees |
Fin is waiting for a teammate | "Let me check with the team. I'll be back shortly." The message box stays available. |
The teammate approves or answers | Fin's reply, in the same conversation, from Fin, with "Reviewed by [teammate's first name]" under the reply |
The teammate takes over | "Waiting for a teammate" with the teammate's photo, then "[Teammate] joined the conversation" and the teammate's reply |
Time runs out | Depends on the Timeout setting: Fin's latest draft from Fin, or a handoff to your team |
The customer closes the Messenger while Fin waits | Nothing is lost: the reply is waiting when the customer opens the Messenger again |
On email, the reply footer says the answer was composed by Fin and reviewed by the teammate.
"Reviewed by" shows in the web Messenger, including in a phone browser. It doesn't show yet in the Messenger inside your iOS or Android app.
Note: Customers on email see nothing until Fin replies. When a teammate approves Fin's draft, or Fin sends its rewrite straight away after a teammate's input, the email footer names the reviewer, for example "This answer was composed by Fin, Acme's AI Agent, and reviewed by John." The email footer is text only, with no avatar. If "Show 'Composed by AI' in email footer" is turned off, the email shows no footer at all. Timeouts, take overs, and ordinary Fin answers don't name a reviewer.
How to set up Human in the Loop with an escalation rule
To set up Human in the Loop with an escalation rule:
Go to Fin AI Agent › Train › Escalation and click + New under Rules.
Name the rule and add conditions under When.
Under Fin will, choose Ask for input.
Set the Threshold, Who to ask and Settings (Priority, Review cycle and Timeout).
Click Save, then Enable.
Send a test question and check that it appears in Needs teammate input.
Each step is explained in detail below.
How to create the rule and add conditions
Go to Fin AI Agent › Train › Escalation and click + New under Rules. Name the rule and set the conditions under When. A new rule applies to Chat, Email by default, and one rule can cover both channels. Untick a channel only if you want different settings per channel, for example a shorter timeout on chat.
Add conditions with + Add Filter, for example an AI attribute you've created, such as an attribute called "Topic" with the value Cancellations & refunds. In the condition list, AI attributes are under AI Agent. Keep your first rule to one or two topics.
Why don't I see my topics from Analyze? You create AI attributes under Fin AI Agent › Train › Attributes, and Fin detects them during the conversation.
You can also get there from the Escalation page by clicking the blue attributes link in the Rules description.
These aren't the Topics in Analyze, which are for reporting and can't be used as a rule condition.
How to choose Ask for input
Under Fin will, choose Ask for input (a new rule starts on Hand over).
In this example:
Match any means a conversation needs to match only one of the conditions.
Threshold: Very high means Fin always asks. Lower it once your team approves most drafts unchanged.
Ask All teammates sends the question to everyone. Choose a teammate or team to narrow it.
Configure opens the Settings drawer for Priority, Review cycle and Timeout.
How to set the threshold, who to ask and settings
Set up these three controls at the bottom of the escalation rule:
Threshold: "Fin will ask for input when its confidence is below this threshold." The slider goes from Low to High, and a new rule starts on Low. The threshold is how sensitive the rule is: higher means Fin asks more often. At the highest level, Fin always asks. Choose High to have every matching answer checked while you start, and lower it once your team approves most drafts unchanged.
Who to ask: a new rule asks All teammates. Pick a teammate or a team who will actually watch Needs teammate input: for example a dedicated team, your senior teammates only, or everyone on shift. See "Who does Fin assign the question to?" later in this article.
Settings: click Configure to open the Settings drawer.
It has three sections.
Priority: "Assign a priority to human in the loop cycles": None, Low, Medium, High or Urgent. Use it so teammates answer the most important asks first.
Review cycle: "When Fin generates a new answer based on your teammates' input, Fin will..." either Get back to customers right away (faster response times) or Show teammate for approval (full control on the final draft).
Timeout: "Set a timer for your teammates and tell Fin what to do if they don't answer in time." Set the hours and minutes under If there is no response after, then choose what Fin does: Wait indefinitely (Fin waits until a teammate answers), Continue the conversation (Fin sends its latest draft and carries on with the customer) or Handoff (Fin hands off to the default assignee, a workflow, a team or a teammate).
How to enable and test the rule
The Workflows group only appears when your workspace has a live workflow. To hand off to a workflow, create a reusable workflow in Fin AI Agent › Workflows › New workflow › Create from scratch › Reusable Workflow and set it live. Click Save, then Enable.
To test, send a calm question on the rule's topic that Fin can answer from your content, then check the question appears in Needs teammate input. Don't test with an angry message or "can I talk to someone?" as Fin will offer a handover instead of asking.
Tip: You can also set up Human in the Loop in the Fin step in Workflows. Already hand over a topic where your team usually just confirms something? Switch that rule's Fin will to Ask for input.
How to set up Ask for input on escalation guidance
Escalation rules match on attributes. For situations you can describe but can't capture with attributes, such as a customer asking for an exception to a policy, use escalation guidance. With Ask for input on, Fin asks a teammate before escalating, and the conversation stays with Fin.
Go to Fin AI Agent › Train › Escalation and click + New under Escalation Guidance.
Describe the situation in plain language, for example "Escalate if the customer asks for an exception to the refund policy, such as a refund for a missed hotel night."
Turn on Ask a teammate for input before escalating, then click Configure.
Set Threshold, Teammate to ask, Priority, Review cycle and Timeout. Each works the same way as in an escalation rule, described in "How to set up Human in the Loop with an escalation rule". New guidance starts at a Low threshold, and a 24-hour timeout that continues the conversation.
Click the status chip at the bottom left (it shows Disabled until you turn it on) and switch on Service agent under Availability. Then click Publish to Fin Main to set the guidance live.
If a conversation matches both a rule and escalation guidance, the escalation rule applies first. A matching rule always acts (Fin asks for input or hands over, as the rule says), while escalation guidance may only lead Fin to offer an escalation.
Who does Fin assign the question to?
Fin assigns each Human in the Loop question as follows:
Fin assigns the conversation to the teammate or team in the rule ("Fin assigned you" in the conversation).
If the rule asks a team, the team's usual assignment method applies, for example round robin or balanced assignment.
If Fin asks again on the same conversation, the new question goes straight to the teammate who answered the first one, skipping the team's assignment.
When the answer is sent, Fin unassigns the teammate.
When a teammate takes over, the conversation stays assigned to that teammate.
What are the best practices for Human in the Loop?
Human in the Loop works best as a confidence gate with an owner: a precise rule, a named person or online team watching the questions, and teammates who answer Fin rather than rewrite Fin. The most common ways it goes wrong are a rule on everything at the highest threshold (a human in front of Fin, not in the loop), nobody watching the queue so timeouts decide, and teammates taking over without giving Fin anything.
Start with what you hand over today. Pick a topic your team already gets as a handover, where the teammate usually just confirms something. Switching that rule's Fin will to Ask for input is the quickest win: your team gets the same conversations, with a draft and one question instead of a blank reply.
Start narrow, especially on chat. One topic per rule, on one brand or audience first. A rule on every conversation at High turns Fin into a triage inbox: your team reviews everything, including answers Fin was already getting right.
Name an owner. Ask a specific teammate or a team that's online, and make sure someone owns Needs teammate input. If nobody answers, the timeout decides, and Fin's original draft goes out unreviewed.
Set the timeout longer than your team's usual time to answer. The Human in the loop report shows average time to action per teammate. If drafts keep going out after the timeout, lengthen the limit or wait indefinitely.
Pick an approach. Ask only when unsure: a topic rule at a middle threshold. Check everything: a precise rule at High, for sensitive topics or while you're new to Fin. Plan to move from the second to the first.
Train your team to answer, not rewrite. Taking over without giving Fin anything costs almost as much as handling the email yourself, and teaches Fin nothing. If one sentence would fix the draft, give Fin the sentence. Pick 2–3 champions and walk through real examples. The key message: you're briefing Fin, not replying to a ticket.
Check Fin's new draft before approving. Fin combines your answer with your content, so read the new draft before you click Approve, especially when your answer is an exception to the usual policy.
Teach Fin with your team's answers. Every time Fin asks a teammate for help, Fin records what it was missing. If your team keeps giving Fin the same fact, fix the gap so Fin can answer on its own next time.
With Operator (Pro add-on), ask: "What does Fin keep needing help with when it drafts replies?" Operator lists the subjects Fin asked about, how often, and the type of gap:
Content: Fin was missing information. Add or update an article, guidance or a procedure.
Action/connector: Fin needed to look something up or do something in another system, such as checking a refund status. Add a data connector so Fin can do it itself.
Pick a gap and Operator suggests the fix. Each gap you close means Fin asks your team less often.
Give a rule about a week of conversations before you judge the rule.
Then lower the threshold. If your team approves most drafts without changes, lower the threshold one step so Fin answers those conversations itself, and check again two weeks later. If Fin's resolution rate holds, lower the threshold again. Each step down means Fin asks less often.
Don't build rules on upset customers. Fin offers upset customers a handover, so a rule on angry sentiment almost never asks.
Check for overlap with your existing escalation rules and guidance, and turn off rules that never fire.
If Fin is weak on a topic, fix the content first. Where Fin's knowledge is thin, the human-checked answers resolve far better than Fin alone, but that's your team propping Fin up. Add the content, then lower the threshold.
Where does Human in the Loop work?
✅ Human in the Loop works with Escalation Rules on email and chat, in conversations Fin is part of.
✅ Escalation guidance, with Ask a teammate for input before escalating turned on.
Human in the Loop on chat
Use Human in the Loop on chats you already hand over. Your team would get these chats anyway, and the customer would wait for a handover anyway. With Human in the Loop, your teammate makes one decision and Fin finishes the conversation. Don't add chat to rules that cover topics Fin already resolves on its own.
One rule can cover chat and email. Add chat to an existing email rule only if the same topics and the same timeout make sense on both; otherwise make a separate chat rule.
Your customer is waiting live. The teammate sees a countdown on the question when you set a timeout. A short limit (for example 5 minutes) means the customer never waits long, but Fin does what the Timeout setting says when time runs out. Waiting indefinitely means every reply is checked, but the customer waits as long as your team takes. Choose a short limit for topics where Fin's drafts are usually right, and wait indefinitely only where your team is online and quick.
Route questions to a team that's online during chat hours, and make sure everyone on the team knows how to answer before you switch Human in the Loop on.
Human in the Loop on email
The customer sees nothing until Fin replies.
Waiting indefinitely means nothing goes out unreviewed, but the customer waits until someone answers. If you set a timeout, set it longer than your team's usual response time: Fin does what the Timeout setting says when time runs out. The Human in the Loop report shows your team's average time to action.
How to troubleshoot Human in the Loop
Rule isn't firing: is the rule enabled? Does the channel match? Does the conversation reach a Let Fin handle step in your Fin workflow? Is the threshold low? Set it to High to test: at High Fin asks on every matching answer.
The rule matched but Fin didn't ask: either Fin was confident, or Fin's reply was an offer to hand over rather than an answer. Handover offers skip the confidence check, so they never trigger a question. Fin offers a handover when the customer sounds upset, asks for a person, or asks something your content doesn't cover. Open the conversation to see which one happened.
Too many questions: lower the threshold, narrow the conditions, or add the missing information to Fin's knowledge.
Lots of drafts sent after the timeout: nobody's watching. Check Who to ask is a person or an online team, lengthen the timeout, or wait indefinitely.
Teammates mostly take over: show teammates Give Fin an answer in the teammate guide. If they take over because the draft is wrong on a whole topic, fix the content for that topic.
Every matching conversation asks, even good ones: the threshold is at High on a broad rule. Narrow the conditions or lower the threshold one step.
Finding past Human in the Loop conversations: Needs teammate input only shows questions still waiting. For all Human in the Loop conversations, open the Human in the loop report and drill in.
Fin asked, but a procedure should have handled the conversation: the procedure didn't trigger. Human in the Loop only applies when no procedure runs, so check the procedure's trigger and conditions.
Fin asked, even though you have content on the topic: Fin's confidence in this answer was under your threshold. Click Show reasoning on Fin's draft to see why. If Fin's drafts on this topic are usually right, lower the threshold one step. If the content doesn't cover cases like this one, add that to the content.
My Analyze topics aren't in the rule conditions: rules use AI attributes, not Analyze topics. Create an AI attribute (for example "Topic") with the values you need in Fin AI Agent › Train › Attributes.
Fin's new answer went out without a teammate approving it: the rule's review cycle is set to Get back to customers right away. Change it to Show teammate for approval in the rule's Settings drawer (click Configure).
I can't hand off to a workflow: the Workflows group only appears when the workspace has a live workflow. Set a reusable workflow live first.
How is Human in the Loop billed?
There's no additional charge for Human in the Loop itself. Billing stays per resolution, and the usual resolution rules apply: you're charged when the conversation becomes a Fin resolution (the customer confirms the answer helped, or doesn't ask anything else or leave negative feedback). If the customer is unhappy or asks for a person, there's no charge.
Whether you're charged depends on who ends up sending the resolution, not on whether a teammate was involved:
Teammate gives Fin an answer, Fin sends the reply: this is billed. The teammate supplies a fact or decision, Fin writes and sends the final draft, and if that draft becomes a Fin resolution, it's charged the same as any other Fin resolution.
Fin's draft goes out unreviewed after a timeout: this is billed the same way. It's still Fin's own reply becoming a resolution.
Teammate selects Other options › Take over from Fin: this is never billed. The conversation is assigned to the teammate, Fin stops replying, and the teammate — not Fin — sends whatever comes next. Because Fin never sends the resolution, there's no resolution to charge for.
In short: "helping" Fin (answering its question) can lead to a billed resolution once Fin sends that reply. "Taking over" removes Fin from the conversation entirely, so it's never billed.
FAQs
Can Fin check with my team before replying?
Can Fin check with my team before replying?
Yes, Fin can check with your team before replying. With Human in the Loop, Fin asks a teammate for an approval, information or a decision, then finishes the conversation itself, on email and chat.
Does Human in the Loop lower my Automation Rate or Resolution Rate?
Does Human in the Loop lower my Automation Rate or Resolution Rate?
Human in the Loop doesn't lower your Automation Rate or Resolution Rate when Fin finishes the conversation. A teammate reviewing or answering Fin's question doesn't stop the conversation counting as a Fin resolution:
Approve original solution or Give Fin an answer: Fin sends the reply and owns the resolution. If the conversation becomes a Fin resolution, it counts toward Resolution Rate and Automation Rate, the same as any other Fin resolution.
Timeout runs out: Fin sends its original draft, and the conversation is counted the same way.
Other options › Take over from Fin: the conversation is reported as escalated, with the reason Teammate took over, and it doesn't count as a Fin resolution.
The Human in the loop report shows how many of your loops ended in a Fin resolution.
Why isn't my Human in the Loop rule firing?
Why isn't my Human in the Loop rule firing?
A Human in the Loop rule most often doesn't fire because Fin offered to hand over instead of answering (for example, because the customer was upset or asked for a person). Human in the Loop only applies when Fin is about to send an answer. See the troubleshooting section.
Is Human in the Loop available with Fin on other helpdesks?
Is Human in the Loop available with Fin on other helpdesks?
No, Human in the Loop isn't available with Fin on other helpdesks. It's only available with the Intercom helpdesk.
Can teammates get Fin's questions in Slack?
Can teammates get Fin's questions in Slack?
No, teammates can't get Fin's questions in Slack. When using Human in the Loop, Fin's questions appear in the Intercom Inbox only.
Is Ask for input available in Australia (AU) workspaces?
Is Ask for input available in Australia (AU) workspaces?
No, Ask for input on escalation rules and escalation guidance isn't available in workspaces hosted in Australia (AU), because confidence scoring isn't supported in that region. This is expected behavior, not a permissions or setup issue.
In AU workspaces:
The Fin will section (Ask for input or Hand over) doesn't appear in the escalation rule editor.
Escalation rules that were set to Ask for input before this change no longer apply. Rules set to Hand over keep working as before.
As a workaround, create an escalation rule that hands over, and use a Fin workflow for routing and any steps after the escalation. This isn't the same as Ask for input: Fin hands the whole conversation over instead of pausing for a teammate to review a draft and then finishing the reply itself.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts



















