Train Fin

We've all spent time improving Fin's knowledge, testing answers and looking at resolution rates.But I'm curious about what happens after Fin actually sends the answer.Say Fin tells a customer:“Yes, you can get a full refund.”…but your actual policy doesn't allow it.Or it gives an answer that sounds completely normal, but contains something your team shouldn't be promising.How are you catching those cases today?Manual QA? Monitors? Sampling conversations? Customer complaints? Or do they mostly get missed?I'm particularly interested in teams running Fin at meaningful volume. How are you handling this in practice?

Hi community 👋Has anyone found an effective way to handle customers who open a conversation by simply typing “agent”, without giving Fin / the AI Agent a chance to help?I’ve tried:general guidance encouraging self-serve first (minimal impact), and enabling “Set expectations for human support”, which unlocks “Ask for information before handover.” However, in our case the bot does ask a follow-up question, but then hands off to a human agent before waiting for the customer’s reply.What I’m looking for is a smarter approach (likely via guidance/behavior rules) where, when someone just types “agent” up front, the bot:probes for key details first, attempts to answer appropriately, and only escalates if it can’t be resolved.Any best practices, example setups, or wording that’s worked well for you would be hugely appreciated 🙏🏼

Hi everyone,We’re using Fin in Hebrew, but we keep seeing cases where Fin adds random English words inside Hebrew replies, even when those words do not appear in our Help Center articles or snippets.We also had a more serious case where Fin inserted Japanese/Chinese-looking characters into a Hebrew customer-facing reply.We already added clear Guidance telling Fin to reply fully in Hebrew and avoid English unless it is a brand name, URL, product name, or a required term. The issue still happens.Intercom Support suggested using Guidance and mentioned the multilingual glossary, but also clarified that the glossary does not apply to AI-generated replies.Has anyone using Fin in Hebrew or another non-English language found a reliable way to prevent this?Is this something Guidance can actually solve, or is this a current limitation of Fin?This is affecting customer trust, so any practical advice would be appreciated.

Hey Everyone, I just wondering some intercom users, I am working on Compliance Monitoring for Intercom users.

Fin doesn't read your knowledge base the way a person does. It reads structure, tags, and outcomes. Get those wrong during migration, and resolution rate suffers from day one — even if every article technically "moved."Across Intercom's customer base, Fin averages a 67% resolution rate. Top-performing teams hit 80% or higher. The gap between those two numbers usually comes down to migration quality, not the model.How Fin actually reads your knowledge baseFin doesn't treat your help center as a flat pile of articles. It uses hierarchy — Collection, Section, Article — to understand scope and context before it answers.Most help desks only support a three-level structure. Intercom allows a simplified Collection > Article path with no Section in between. That's fine natively. It's a problem in migration: if your source platform's categories don't map cleanly onto that structure, a migration tool has to guess, usually by dropping orphaned articles into a default folder. That default bucke

Hi everyone,We’re trying to build an Intercom Data Connector that retrieves customer-specific data from our backend.Our backend API requires a user-specific access token. This token is generated only after the customer completes an OTP/authentication flow, and each customer receives a different token.The flow we want to support is: The customer completes OTP authentication. The OTP/auth connector returns a response containing: { "data": { "access_token": "..." }} A second connector should then call our recommendations API using that token in the Authorization header: Authorization: Bearer <data.access_token>The issue:When we try to pass the token dynamically in the connector header, for example:Authorization: Bearer {{accessToken}}or by using a value returned from a previous connector/action, our backend does not receive the Authorization header/token.The only setup that worked was using Intercom’s Authentication Token feature with a static Text token. However, this does n

When Fin replies or send a follow up for inactive customers, the outbound reply doesn't include the quoted thread below the message. Recipients only see the reply itself with no prior context.Is there a setting in Intercom to enable quoted/threaded replies so the previous email history is included in outbound replies the way traditional email clients handle it? If not, what's the recommended workaround for managing external partner email threads through Intercom without losing context for recipients?

Hi everyone,We're exploring how to measure content readiness before syncing our website pages to Fin. We want to ensure our content is structured so Fin can parse it cleanly and pull accurate answers without losing context.We know about the optimization guidelines,Rather than testing AI search results after the fact, I want to focus on scoring the quality and structure of the content itself before it gets synced. A few areas I'm looking to refine: Structure & Noise Removal: Verifying that HTML clutter (navbars, footers, sidebars) is stripped so only clean body text is processed. Chunk Self-Containment: Testing whether standalone sections or paragraphs make sense on their own without missing key context or relying on surrounding text. If you've built any checks or used tools to score content readiness before pushing to Fin, I'd love to hear what's working for your team!

Hi all,Looking for guidance on customizing Fin's default human handoff behavior (We use Fin over API).Right now, when a user asks for a human (Right from the first message), Fin responds with:"I understand you would like human support. Would you like me to connect you to a human agent? You can also continue working with me and provide more details if you prefer."We're building a multi-stage escalation flow (a short ladder of clarifying prompts before handoff) to collect 3 key infos for our support team (What customers were trying to do in the product, where are they in the product, what happened), but we need the final stage to first check whether the user already gave issue details earlier in the conversation, rather than asking for information a third time if they already provided it. I understand that Escalation Guidances are stateless so we can’t store anything in there. How can we build a robust support detail collection flow in case customer asked directly for human agents? i.e (

Hi, we have our workflow set up to allow 5 minutes to have a team member answer incoming chats and emails during office hours,  Fin kicks in before the 5 minutes and answers the chat or email. When one our team members replies to a chat or email Fin will then reply to the team member. Has anyone experienced this? I feel like its an easy fix , not sure if we are missing a step in our workflow. Any advice would be greatly appreciated

With Fin for Sales, we aren't using the meeting booker integrations for qualified sales meetings. Instead, we route prospects with the gCal app to schedule meetings with sales reps based on territory. One limitation we've encountered is that we haven't been able to set the meeting owner to a field that syncs with the Salesforce record owner. How can this be made possible?

Can Fin read customer order forms sent by email as PDF attachments, including ones with complex or multi-column layouts, and convert the data into a standard JSON format? If not, can we use a workflow or integration in Fin to reliably process these documents and ensure the final output is in JSON format?

We're trying to get Fin to behave differently depending on when the customer asks for a human. We've tested two separate guidance rules but it's not sticking. Anyone solved this?Here’s what we want to do:First message = "offer" escalationCustomer says "representative"? Fin should hit them with: “Would you like me to connect you with a human agent? Or if you tell me what you're looking for help with, I'd be happy to try assisting you first.”(Reason: Let Fin actually try to help before bouncing them out)Mid-conversation = "escalate immediately"Customer says "representative" after you've been chatting? Skip the offer—just connect them. No questions asked.(Reason: They're probably frustrated and have already interacted with Fin. Don't want to make them explain twice.) 

I'm looking for suggestions on how to organize Help Center articles and knowledge base content so users can quickly find the information they need. Do you prefer organizing content by categories, user intent, or product features?While researching content structure and user engagement, I came across a resource about baby care products that does a good job of presenting information in a clear and easy-to-navigate format. I'm interested in learning what strategies others use to improve content discoverability and user experience.

Hey all!I’d been exploring this using Fin but had hit a road block and wanted to see if anyone had created anything like this:We want every negative Fin CSAT to create a ticket to our team to review to see if we could make any changes to Fin.We use a reusable workflow to route negative Fin CSATs to the team - and we want that to continue - but in the background we want this ticket to route to a different team to review. Any thoughts? My first post here, so let me know if not the right format.

I find the in app Fin experience on intercom to be incredibly helpful and I use it all the time. However, I get so many “rate your conversation” emails that it is starting to get annoying. Is there a way that I can mute these? To be clear, this is my as an intercom user using the in-app Fin to ask intercom related questions. Thanks!

Ability to Export Guidance. Manually gathering guidance from 7 workspaces is too time consuming. 

We have customers that answer no multiple times to Fins questions of ‘Did that answer your question’ or similar. They will say ‘No’ and rephrase their question but Fin still can’t answer it correctly. Does anyone have any recommendations on how I could get it to ask if they want to speak to a human after the 2nd time of saying ‘No it hasn’t helped’

Can we better contextualise Help Centre content for Fin?For example, having internal-only guidance/instructions embedded directly within articles that end users cannot see. While content/source guidance and snippets help, they become difficult to scale when managing thousands of articles and hundreds of supporting snippets.In our case, Fin’s errors are often less about incorrect content and more about misunderstanding nuance or scenario context. Being able to add contextual instructions, edge cases, or operational guidance directly alongside the article content would make maintenance significantly easier and likely improve response quality.

We use the “Other” guidance heavily for our platform and voice isn’t supported yet. Wondering how other people have worked around this? I tried to update our FAQs as much as possible but there has been issues in which our Fin voice keeps repeating the wrong thing that would have worked in chat or email because of Other guidances set in place. 

Something is going wrong with my anti-repetition guidance. No matter what, Fin will not stop repeating/rephrasing the same answer multiple times. I have guidance on asking clarifying questions and am wondering if that’s part of the issue. My adjustments to clarification guidance also aren’t making things better. Using optimize and other ai tools isn’t helping.I’m stuck and am a team of one. If you have any suggestions or examples that work well for you, please list them here!

We have a number of products with a lot of overlapping terminology, and a lot of audiences that overlap with other audiences. This has made it difficult to target the information Fin uses very specifically. We have been considering setting audiences up using conversation data instead of just user data, but I’m having a hard time understanding if/how that would work. We have a new conversation attribute that would be selected and help define the audience for that conversation. But is that effective? If the customer asked another question in the same conversation, or had chosen something incorrectly, could the audience for that conversation ever be changed? Open to any suggestions here on how this could work, or what better options might be. 

Fin AI will always give suggestions when a customer asks a question. Let’s say the customer asks about a device (we have physical products that require a little technical knowledge), Fin AI always responds something along the lines “What do you want to know? (for example: this thing, that thing, many other things) You might be able to do this or that”I would like to to answer the customer’s question and not give extra information unless absolutely relevant. However, no matter what guidance I give, it will not stop this behavior.

Hi All, How do you all work around the guidance limitation of 100 pieces of total guidance? We have the opportunity to consolidate some guidance, but because the bot is used across multiple teams with different workflows and channels, this limitation will hamper us. Is there a way to increase the limit?  What alternatives do y’all use to guidance to achieve the same outcome?  How many pieces of guidance do y’all have? Thanks!Afton

I have a very simple use case. I want Fin to ask for the Project and the Organization before it escalates. For some reason, Fin continues to frame it like this:“Would you like me to connect you with a human agent to assist further? Or if you provide more details about your project and organization, I can continue to help you.”No amount of escalation guidance refining is working. Fin should be able to handle this, no?