Share product ideas and upvotes with our product team
Hey everyone! 👋 I’ve been diving deep into the new multi-role capabilities for Fin, and while the "one agent for everything" approach is a great start, I believe the next evolution of Fin should be Persona-Based. Right now, we have one "Fin" with one name and one tone across the whole journey. But in the real world, a customer's needs change as they move from the storefront to the support queue. Here’s why I think Intercom should develop the ability to define distinct Personas (e.g., "Mike" for Sales vs. "Thomas" for Support): 1. Trust is Built on Context 🤝A "Sales" persona needs to be proactive, persuasive, and high-energy 🚀. A "Support" persona needs to be empathetic, cautious, and technical 🛠️. Using a single "Humorous" or "Professional" tone for both feels like a compromise. 2. Visual Continuity & Human-Centric Design 🖼️By allowing us to set different Names and Avatars for different services: For Sales: A friendly face (Mike) that feels like an account executive. For Support: A specialist identity (Thomas) that signals expert assistance. This helps customers mentally "switch gears" and sets the right expectation for the conversation. 3. Specialized "Tool-Belts" for Agentic AI 🧰If we treat these as separate personas, we can define specific Permissions and Data Connectors for each: Mike (Sales) could have access to HubSpot or Salesforce to check lead status 📈. Thomas (Support) could have access to internal dev docs or API logs to troubleshoot bugs 🐛. 4. Cleaner Analytics & Handoffs 📊Separating personas allows for role-specific KPIs. We could measure "Mike" on Conversion/Booking Rate and "Thomas" on Resolution/Deflection Rate. It also makes the "AI-to-AI" handoff possible—imagine Thomas realizing a query is actually a sales lead and "transferring" the user to Mike! The Vision: Moving from a "Generalist Bot" to a "Specialized AI Team." 🏛️🇬🇧 I’d love to hear from the product team and the rest of the community—would you prefer one consistent Fin, or do you see the value in a squad of specialized AI personas? #FinAI #ProductFeedback #AgenticAI #CustomerEngagementCheers, Roy 🇬🇧
We’re looking for a way to route conversations to a human team after Fin AI Agent has resolved the user’s questions — specifically through Workflows, not Procedures.Important context:We need this workflow to be triggered based on conditions like specific URLs (e.g. contact buttons or pop-up proactive messaging on a specific page)And at this time, that requires Workflow-level control. This cannot be handled in a Procedure currently, as far as I understand. Use case:There are certain topics where Fin can fully answer questions, but a manual action is still required afterward.Example:A user lands on a migrations page and clicks “Contact us” Fin answers all their migration-related questions The conversation is resolved But we still need it routed to our team to process the migration requestCurrent limitation:If we force routing early, we lose the benefit of Fin answering questions first If we let Fin resolve, the conversation never reaches the team Procedures don’t solve this because they can’t control post-resolution routing based on entry conditions like URLWhat we need:A Workflow-based solution that can:Trigger from specific entry points (e.g. URL, button click, referral source) Allow Fin to fully handle and resolve the conversation Then automatically route the conversation to a specific team Optionally mark it for internal follow-up without reopening it to the userWithout this, we’re forced to choose between automation or operational workflows, instead of combining both.Ideal behavior:Workflow triggers → Fin handles conversation → Fin resolves → Workflow routes to team for internal processing
Assign X random conversations per teammate for review (including phone calls and Fin not involved)I need some kind of tool that assigns an specific amount of conversations from my teammates for my review on a weekly basis. Some of these conversations are phone calls, some of them do not have 2 messages from Fin nor 2 responses from the customer.Monitors is a great tool, but it only works for conversations with at least 2 messages from Fin and 2 messages from the customer, and I need a sample of all conversations, not only those with the previous criteria.
Edit: seems like despite it saying it’s making an update, it does ask for confirmation once it’s finished the proposed update. The thought process as it loads give the impression it’s going to make the update without confirmation,Just like with Claude, the ask feature is too eager to make changes to things. I want guardrails in place that it always asks before updating procedures, guidance, etc rather than just updating itself and having to click “stop” before its too late.
At the moment we’re unable to add Albania as a supported language within the help centre.Adding it would be much appreciated. Thanks!
It would be very helpful to allow category change within the guidance editor.Often, I will find my guidance changes over time and needs to be moved category, but this is a manual process of creating a new guidance and copy/pasting content over. Not only for speed, this removes usage stats. While in alot of cases this will be necessary as the guidance scope will have changed you may not want to lose those stats if you have simply miscategorised a piece of guidance or slight change in wording reuqires a move to a more accurate category.
Pls add Uzbek language for AI detection, as many Uzbek customers are not speaking russian, or don’t fully understand it. You have whole bunth of localization languages, but not Uzbek, and we have link 250k customers speaking Uzbek
Currently, Fin can’t tell if my customer is logged in or not during a conversation. This would be nice so that I can apply different Procedure steps based on that (sometimes I need the customer to log in, especially if I’d need to come back later and they would have lost their web session).
We're building a FIN AI Attribute to automatically detect and tag which law firm a user belongs to, so FIN can route and respond appropriately across both firm users and their end clients.We have 526 active law firm clients on the Legl platform, each requiring their own value with a distinct name and description so FIN can correctly identify users by email domain and firm name mention.The current 250-row CSV upload limit means we cannot load our full firm list in a single attribute — we have to split it across multiple uploads, which creates a fragmented and harder-to-maintain setup. Ideally we'd be able to upload all values in one go.Request: Please increase the CSV upload limit for AI Attribute values to at least 600 rows, or allow multiple CSV uploads to append to an existing attribute's value list rather than replacing it.This would be particularly valuable for any Intercom customer operating in a B2B2C model where they need to tag by client organisation at scale.
Most teams treat QA, documentation, and Fin as separate problems.They’re not.In this episode with Thomas Hils (Head of Support @ Wingspan), we break down how to connect all three into a single system that: Uses real conversations to find gaps in your docs Improves processes automatically Makes Fin significantly more effective The key idea: Fin can’t answer what doesn’t exist.If your resolution rate is stuck, this is likely why.Watch here
I noticed fin took it up on himself to use English terms even though our German help centre is fully translated.1. It be nice that fin relies on the terms in the translated article2. Fin to use the uploaded Glossary in the Multi-language section,
At the moment, Copilot relies on the original attribute assigned to the ticket. Manually changing the attribute afterward does not have any effect.In Snippets, we assign a specific audience based on an attribute, and this attribute depends on where the customer contacts us from. When customers reach out through the in-app chat, everything works as expected. However, when they contact us via email or our company’s website chat, we’re unable to identify which brand they belong to. As a result, the correct audience is not applied, and the relevant Help Center articles are not used for their questions. I also tried manually updating the attribute after the ticket was created, but Copilot still does not recognize the change.
We need a way to include only conversations that received a CSAT rating during the selected period, so we can evaluate only the conversations that were actually rated. Right now, Monitors can include conversations that have no CSAT in the period (because they’re being included based on the customer’s most recent rating), which makes the Monitor and any attached scorecard unreliable for period-based evaluation.Justification: Monitor reports are already sensitive to how metrics are aggregated, so adding conversations that weren’t actually rated in the period makes it harder to trust the totals you’re reviewing.
Description :We operate with multiple support tiers (Level 1, 2, and 3), where query complexity increases with each level. We'd like to deploy Fin on our email support channel, but only for queries where we're confident it can deliver accurate answers.For Level 1 and Level 2 requests, Fin handles responses reliably. However, for Level 3 requests, the answers are significantly less accurate, and we need a human agent to respond instead.What we'd like:The ability to configure a workflow where, upon receiving an inbound email, Fin first detects the support level via a custom attribute, and then routes accordingly: Fin answers Level 1–2 queries automatically, while Level 3 queries are routed directly to a human agent — without Fin attempting a response.Today, Fin either answers or doesn't, but there's no way to conditionally gate its response based on a pre-classified attribute before it decides to engage. This would give us much more control over where Fin adds value versus where it could hurt the customer experience.
Summary:When a scorecard's attributes are updated, unreviewed conversations already captured by a Monitor should reflect the new attributes, either as incomplete/pending fields, or via a manual "re-apply scorecard" option.The problem:Scorecards evolve. When you refine attributes to better capture quality signals, only future conversations benefit from those changes. Existing conversations in the Monitor, especially unreviewed ones, are locked to the old scorecard structure, making it impossible to apply your improved rubric retroactively without manual workarounds.This is particularly frustrating when a Monitor has been running for weeks and you haven't yet reviewed the backlog. Updating the scorecard mid-cycle effectively creates two incompatible datasets within the same Monitor.What I'd expect instead:At minimum, unreviewed conversations in a Monitor should have the new or updated attributes added as empty/pending fields when a scorecard is changed, preserving any existing scores while surfacing the new ones for review. Reviewed conversations could reasonably remain unchanged.Ideally, there would also be a "Re-apply scorecard to unreviewed conversations" option, a reset or refresh button scoped to conversations that haven't been manually evaluated yet.Impact:Without this, teams have to either (a) accept inconsistent data across a Monitor's history, or (b) or manually re-process old conversations. All of these options undermine the value of Monitors as a continuous QA tool.
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.