Share product ideas and upvotes with our product team
Hi Intercom community! At the moment there's no way to stop a conversation closing for inactivity while a procedure is running. If a customer drops off halfway through, the conversation just closes. The procedure doesn't finish, nothing gets escalated, and we've got no reliable way of finding it again afterwards. That's fine for most topics but it becomes a problem for anything high-risk or high-priority, where we need the case seen through whether or not the customer comes back to us. The only workarounds are adding tags and creating extra workflows to reopen these conversations. It would be great to have functionality within the procedure itself to escalate if the user becomes inactive, or a global setting that always escalates to the default inbox if the conversation contains a certain tag & the user went inactive (this might be more future-proof, incase of new features or the same problem found in other existing features)
Currently, Fin stops responding when a teammate replies or an inactivity timer is reached. Manually closing a conversation does not interrupt a "Let Fin Handle" workflow.We'd love the ability for manually closing a conversation to immediately stop any active Fin workflow.This would allow us to use Fin's automatically generated follow-up responses while still giving teammates the flexibility to end a conversation when appropriate, rather than relying on a fixed "Customer has been unresponsive" workflow or waiting for inactivity timers.
It’d be great if we could trigger workflows based on customers clicking a button at the end of a tour or as part of a tooltip. The current workflow launch options are a bit limited for our use case - it’d help us automate more if we could (for example) trigger a workflow based on a customer pressing a “yes” or “no” button after: - doing a product tour - seeing a tooltip - reading a specific particular article - giving a survey response. These would help us make sure the right people see the right message before launching the workflow (as a pop-up not just in the chat window)
Problem:Fin AI Agent tags individual conversations with a "Fin referenced memory" event when Fin draws on stored memory to respond. That event is visible at the conversation level, but it's not exposed anywhere in reporting. The Fin AI Agent Analyze dashboard shows resolution rate, CX score, and involvement rate, but there's no way to filter or segment any of those metrics by whether memory was used.What we'd like to see:Turn the existing "Fin referenced memory" event into a reportable attribute, similar to how other conversation-level tags already feed into Analyze. Since the tagging already happens at the conversation level, this would mainly be a matter of surfacing that field for filtering and segmentation rather than building new tracking from scratch.Why it matters:Without this, teams have no way to measure whether Fin Memory is actually helping resolution rates, hurting them, or making no difference. Right now the only way to check is opening conversations one at a time and looking at the event log, which doesn't scale for anyone trying to evaluate memory as a feature. Being able to compare memory vs. non-memory conversations on resolution rate and CX score would let teams make an informed call on whether to keep leaning into it.
When Fin answers a question by drawing from a help article, it paraphrases and summarizes the article content well, but it strips out any hyperlinks that are embedded in that article. So if an article says something like "You can create one. Here's how" with "Here's how" linking to another resource, Fin reproduces the sentence in plain text with no link. The customer never sees it.We confirmed this is a documented limitation in Intercom's own support documentation. Specifically, the following two Intercom help articles both address it directly: Fin AI Agent FAQs (https://www.intercom.com/help/en/articles/7727118-fin-ai-agent-faqs), under "Customer Experience": "Currently, Fin AI Agent is not able to present links in consistently formatted hyperlinks. You can use Guidance to encourage Fin to hyperlink where possible, but this will still depend on which sources were initially retrieved by Fin when forming an AI answer." FAQs (https://fin.ai/help/en/articles/16257287-faqs), under "Fin Functionality": "Currently, Fin AI Agent is not able to consistently present links in formatted hyperlinks." We tried the Guidance approach referenced in the first article and tested it extensively. The result is that Fin does surface some URLs inline, but it does not carry through the actual hyperlinked text from the article body. The hyperlinks in our article content are simply dropped. Our customers never see them.The workaround we were pointed toward is writing out URLs as plain text in every article body rather than using formatted hyperlinks. For a help center of any meaningful size, that is not a realistic ask. Our articles are written the way any standard help center article would be, with descriptive anchor text linking to related resources, and the expectation that those links would carry through to Fin responses is a reasonable one.What we are hoping for is straightforward: when Fin references content from an article that contains a hyperlink, it should surface that link in its response rather than silently dropping it. This would make Fin significantly more useful for customers who need to take a next step or find related information, and it would eliminate the need to maintain two versions of article formatting.Thanks for considering this. Happy to share specific examples from our testing if that would be useful.
I’m using Operator when I need to improve a procedure. I’ve recently started using the human-in-the-loop step with some procedures and I’ve noticed that operator is not able to build/read/modify any procedure where this is enable which is quite frustrating. Since it can’t even read it it’s quite difficult to get what you need to be improved
Currently, when navigating the Procedures section, we have some great filtering options like State, Channels, Live first, Tags, and Date range. However, as our workspace scales and our team creates more targeted procedures, managing and finding specific ones is becoming a bit challenging. It would be incredibly helpful to add an Audience filter to the main Procedures view. The idea is to have a dropdown where we can easily switch between specific audiences—for example, quickly toggling between our "Peru" audience and our "Colombia" audience—to see exactly which procedures are active for each segment.
Currently with the slack integration, if Fin recommends any articles that could help for the time being, they are not hyperlinked. See example below. within Fin however, they do appear hyperlinked, just not within Slack.please let me know if there are any questions!
What has been tried: We attempted to address this by implementing and refining a configurable looping behavioural instruction rule, including expanding the frustration signal vocabulary and broadening the trigger criteria. Testing confirmed the rule fires reliably for verbatim-repetition loops. For progressive loops — where the customer advances the conversation rather than repeating it — the rule does not fire. This is a platform-level constraint, not a configuration gap: cross-turn state tracking is not available within the current behavioural instruction layer.We are requesting a conversation-state-aware loop detection mechanism built into Fin's core runtime, with the following properties:1. Turn counting with topic continuity tracking Fin should track, across turns, whether the customer's underlying issue has changed or whether they are persisting with the same root problem. A customer who addresses the same issue across 3+ turns — even from different angles — should be flagged as looping, not just customers who repeat the same sentence. Sometimes looping shows differently accross each conversation, it may appear as progressive instead of stagnant. 2. Frustration signal accumulation Rather than a single-turn evaluation ("does this message indicate frustration?"), Fin should accumulate frustration signals across the conversation. A neutral message following two rejected or corrected responses should be scored differently than the same message at the start of a conversation.3. Conversation re-open recognition If a customer re-opens a conversation (within, e.g., 24 hours) after it was closed as assumed resolved, Fin should treat this as a strong signal that the prior session did not resolve the issue and apply a lower escalation threshold in the re-opened session.4. Time awareness Fin should be able to tell how long a customer is involved in the conversation with FIN and make a call depending on the time spent chatting to offer human assistance or not All these should be confiurable on workspace level and useable in guidance to customise FIN messaging to customer.
Already have an account? Login
No account yet? Create an account
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.