Share product ideas and upvotes with our product team
We'd like reviewed conversations (in Monitors) actually feed back into the AI agent's training.Right now review scores and reviewer comments just sit in the record with no way to use them to improve the AI agent's future responses other than updating our content (which is a manual process).The review data, the score AND the comments should be used as a training signal so the AI agent gets better based on real feedback from our team instead of having us manually update our content and snippets, which is very time-consuming. Comments are an important part because they could tell the model what’s wrong. Most of Fin’s mistakes are small but impacful nuances that don’t require an article or snippet update of its own.There should be a way to directly tell the model what it got wrong and what should have said instead during the review process rather than having our team update our content.
Description:During QA reviews in Monitors, reviewers often identify improvements needed after evaluating a conversation. Today there’s no structured way to capture follow‑up actions.Request: introduce a dedicated action / to‑do step at the end of the QA review where reviewers can define the next step (e.g., coaching, training, process update, content change). This would help teams track improvement tasks directly from the QA workflow instead of relying on notes or external tools.
We only use Fin for Service, but when managing which roles content should be used for, Sales, Service, and Ecommerce are all shown as options. I can turn Sales off, but Service and Ecommerce appear to be linked, meaning Ecommerce cannot be turned off without also turning off Service. Likewise, turning Service on also turns Ecommerce on.From a user perspective, this is confusing because it looks like these roles can be managed independently, even though they cannot. Since we don’t use Fin for Sales or Ecommerce, I’d ideally like the option to hide or disable those roles entirely from our workspace, or at least have clearer UI messaging explaining that Service and Ecommerce are linked.The current setup creates unnecessary friction when managing content because I have to keep seeing and accounting for roles that are not relevant to our use case. It also makes the settings feel more complex than they need to be for teams that only use Fin for support/service.Suggested improvements: Allow admins to hide or disable unused Fin roles at workspace level. Allow Service and Ecommerce content settings to be managed independently. If they must remain linked, make this explicit in the UI before users try to toggle them. Consider simplifying the content management experience for teams that only use Fin for Service. This is not blocking us from using Fin, but it adds avoidable confusion and makes content governance harder than it needs to be.
Workspace members should be able to view and edit the shared memory that Operator uses (e.g. tone of voice, style guides). At the moment, this context isn't visible or editable by the team, which makes it harder to keep it accurate and up to date, and leads to duplicated effort when different team members try to recreate or adjust the same guidance.Giving teams visibility into this shared memory, and the ability to update it, would reduce repetitive work and give everyone clearer insight into how Operator's output is shaped, supporting better collaboration across team members who share responsibility for its performance.
Current behaviour: Monitors can only be archived. Once archived, the only available action is to restore them.Requested change: From the archive, allow users to either restore or permanently delete a Monitor.Why: Without a delete option, the archive accumulates Monitors indefinitely, including test items and other monitors that are no longer needed. Over time this creates clutter and makes the archive harder to manage.A delete option would let us clear out items we genuinely don't need, keeping the archive clean and relevant.
Hello Fin team,It's easy to go a little blind on resolution rates, since what really matters is what the customer believes is the truth. Fin will calculate an assumed resolution if a customer stops replying to the thread and leaves. Although this is quite common in the SaaS industry, it is also not a very safe indicator that the issue was in fact resolved. Even for confirmed resolutions, it may be that the customer thought it was perfect and confirmed, but later found that this was actually not the resolution anyway. To get a better picture of this, please consider this feature request for reporting:Any lead or user who does not only reopen a previous conversation (this can already be done in reporting), but opens a new conversation within X time (a variable that we should be able to select), is considered a recontact. The combination of AI resolution together with low recontact rate is dynamite, and far stronger than just showing off with a high AI resolution alone. Thanks,Christian Osmundsen, Deliverect
We are building multiple procedures that need to share data with each other during the same conversation. Today, when a procedure ends, the context it built (amounts, decisions, variables) is not reliably accessible by the next procedure or by Fin in a general conversation turn.Our specific use case:We need to pass the refund amount calculated in one procedure to a second procedure (e.g. a refund/credit dispute procedure) so Fin can tell the rider exactly how much they were compensated without asking again or losing that information.What we are asking for:The ability to write a conversation attribute inside one procedure and read it reliably in any other procedure or general conversation turn within the same conversation for example:compensation_minutes set in a procedure → readable in the refund dispute procedure refund_amount set during a credit step → displayable to the rider later if they askWhy this matters:Without this, we are forced to either cram all logic into one mega-procedure (which causes the re-triggering and idempotency issues we already flagged) or lose context between procedures and give the rider a worse, disconnected experience.
Currently, website source re-sync requires a full crawl of the entire website, even when only a single new page or a small number of URLs have been added or updated.For organizations with large knowledge bases, this can be inefficient and time-consuming. In our case, we have thousands of indexed articles, and triggering a full re-sync just to make one new page available to Fin is not practical.It would be valuable to support one or more of the following capabilities:• Sync a specific URL or a set of URLs on demand• Incremental sync that only processes new or modified pages since the last crawl• Ability to submit a URL or a set of URLs for immediate indexing without triggering a full website re-syncBenefits:• Faster availability of new content in Fin• Reduced crawl time and resource consumption• Better experience for teams managing large documentation sites• Improved agility for time-sensitive content updatesThis would significantly improve knowledge management workflows for customers with large and frequently updated websites.
We currently use Fin Service for our e-commerce site, and it handles our 2,000+ SKU catalog well. That said, Fin for E-Commerce looks far more capable than Service when it comes to actually selling products. The catch is that we run on Magento, not Shopify, so E-Commerce isn't available to us. I'd love to see E-Commerce support extended to platforms beyond Shopify.
ProblemFor website-based knowledge sources, the current Re-sync operation performs a full crawl of the entire website.In large deployments, this can involve thousands of pages. In our case, the source contains approximately 3,600 articles. When a single new article is published, there is currently no way to make only that article available to Fin immediately without triggering a full website crawl.Requested CapabilityProvide a way to selectively sync website content instead of requiring a full re-crawl.Possible options:Sync a single URL on demand. Sync a selected set of URLs. Incremental sync that discovers only new or modified pages since the previous crawl.BenefitsFaster Content AvailabilityNewly published articles can become available to Fin immediately without waiting for the next scheduled crawl.Reduced Processing TimeAvoid re-processing thousands of unchanged pages when only one or two pages have changed.Lower Resource ConsumptionReduces crawler workload, indexing time, and system resource usage for both customers and Intercom.Better Operational EfficiencyAllows support teams and knowledge managers to quickly publish urgent content updates and make them available to Fin.Example Use CaseA website source contains 3,600+ articles.A new troubleshooting article is published for an ongoing customer issue.The administrator wants Fin to learn only that newly published page immediately, rather than initiating a full crawl of all 3,600 articles.Current WorkaroundsWait for the weekly automatic website sync. Create a temporary Knowledge Hub Snippet. Trigger a full website re-sync.None of these provide true URL-level synchronization.Expected OutcomeAllow administrators to selectively re-index specific pages or run incremental website syncs so that newly published content can be made available to Fin quickly and efficiently.
I think we can all come to an agreement that it would be great if Fin knew where and what you had just been doing in the product before you ask for help. It could better clarify your needs, no need to re-explain where you’re stuck or what you’re looking for.Adding this context is such a no-brainer and no one is doing it.
The problemWe manage Fin with a large, frequently updated ruleset: Context & clarification Guidance plus Escalation Guidance (separate objects in Train → Guidance and Train → Escalation).Today this content is UI-only. There is no public REST API to list, read, create, update, enable/disable, or delete Guidance entries — unlike Help Center articles, which we already manage programmatically via the Articles API.This creates real operational pain:No single source of truth — teams maintain a parallel wiki/repo and manually copy-paste into Intercom after every change. No programmatic audit — conversation audit events expose fin_guidance_applied with guidance id and title, but we cannot fetch that object via API to verify live text, diff against git, or detect drift. “Ghost” guidance risk — we have seen guidance applied in production (visible in conversation audit) that does not appear in the editable Train list, making it impossible to fix or retire bad rules without Intercom support. High iteration cost — when routing logic changes (e.g. buy/ramp vs gift card vs swap), updating 15–20 guidance bodies by hand is slow and error-prone (character limits, audience, channels, paused vs live). Multi-workspace sync — other customers already ask for export; we need import + update, not just export.We already automate Fin training for Help Center and Data Connectors; Guidance is the missing piece.Proposed solutionAdd REST API resources for Fin Guidance and Escalation Guidance, modeled after Articles:Suggested endpoints (names illustrative): GET /fin/guidance — list all guidance entries (paginated) GET /fin/guidance/{id} — retrieve one entry by id (same id as fin_guidance_applied) POST /fin/guidance — create PUT /fin/guidance/{id} — update body/metadata DELETE /fin/guidance/{id} — delete or archive GET /fin/escalation_guidance (+ CRUD) — separate resource for Train → Escalation tab Fields to expose (minimum):id, title, body (plain text / markdown) type or category (e.g. context vs escalation) audience, channels state (enabled | paused | draft) created_at, updated_at Optional: usage stats (used, resolved, escalated) for observabilityNice-to-haves:Deep link from Inbox audit → specific guidance id in API response Webhook / event when guidance is created, updated, or deleted MCP server support (same pattern as articles) Preview vs production workspace targeting (if applicable)Why this mattersFaster, safer Fin iteration — git-reviewed changes deployed via CI, not manual paste. Drift detection — compare live Intercom state to repo; catch ghost or stale rules. Incident response — pause or patch one guidance entry in seconds when misrouting is detected (e.g. Phaze gift-card rule matching crypto buy conversations). Consistency with existing APIs — we already manage articles and snippets are requested separately; Guidance is the core Fin routing layer and should be first-class in the API too.Related requests (complementary, not duplicate)Ability to Export Guidance — export-only; we need full CRUD + escalation Add Snippets to the Intercom API Direct API Mapping for Fin Data Connectors — our team’s related Fin automation requestHappy to share our workflow (repo mirror, deploy checklist, conversation-driven QA) if useful for prioritization.
We were using the Review Statuses in Monitors to track conversations where we needed Fin to improve its responses, and then update the status once we’d implemented changes and tested that Fin had improved its responses.It’s helpful when the reviewer and the conversation designer are different people, or simply when the reviewer doesn’t have time to make improvements while reviewing conversations.Now we had to create a tracker using another app. Bringing the Review Status will keep us working within Intercom.
Hello Intercom community, When a customer clicks 😞 or 😐 on a Help Center article, Intercom automatically opens a conversation and sends a built-in reply:German:"Schade, dass du die benötigte Antwort nicht finden konntest. Möchtest du das Team noch einmal um Hilfe bitten?"English:‘It's a shame you couldn't find the answer you were looking for. Would you like to ask the team for help again?’There is currently no way to: Edit or replace this default message Trigger a Workflow at the moment of the reaction The "customer sends first message" workflow trigger never fires in this context because the built-in reaction flow opens the conversation automatically and takes over before any workflow can run.Who is affected Any team that: Uses Fin AI Agent as their primary support agent Wants to collect structured feedback on why articles fail Cares about brand tone consistency across all customer touchpoints What we want Option A - Editable reaction reply text Allow admins to customize the default message that appears when a customer reacts negatively to an article - either globally or per Help Center.Option B - Dedicated Workflow trigger for article reactions Add a new Workflow trigger: "Customer reacts to article" (with filter options for 😞 / 😐 / 😃) that fires immediately when the reaction happens - before any conversation is opened.Ideal Workflow this would unlock Customer clicks 😞 or 😐 on an article A customizable message appears asking why the article didn't help Structured reply buttons are offered (e.g. "Too technical", "Missing information", "Didn't match my situation", "Something else") A short free-text follow-up is collected based on the button selected A closing message confirms the team will review the article The conversation is assigned to a teammate internally for action Why this matters Catches missing or unclear content early - before it turns into a support ticket Provides structured data on why articles fail, rather than just a reaction count with no context Enables teams to close the feedback loop and improve content proactively Ensures a consistent brand tone across all Fin and Messenger interactions Current workaroundNone available. The built-in reaction flow cannot be overridden or customized.Thank you for your cooperation. Best regardsMatej - reevSenior Customer Support Manager
It would be helpful if Intercom had an AI feature that can rephrase a support agent’s message to sound more friendly, natural, and professional before sending it to the customer.For example, agents could write a quick rough reply, then use AI to improve the tone so it feels more like a real customer support conversation. The AI could make the message warmer, clearer, and more empathetic while keeping the original meaning.This would help support teams respond faster while still maintaining a friendly and high-quality customer experience.
When configuring Fin under Settings → Channels → Messenger → Conversations tab → "With Fin," selecting the recommended option "Set AI Agent expectations" automatically inserts the subtitle "The team can also help" beneath the workspace name in the Messenger conversation header.This string is currently hardcoded — there is no way to edit, replace, or remove it anywhere in the Messenger or Fin settings.We'd like to see several customization options for this text:1. Custom text — the most important one. Let us provide our own wording so it matches our brand voice and support model (e.g., "Our support team is here if you need a person" or "Ask to speak with our team anytime").2. Keep the default — workspaces that are happy with "The team can also help" should be able to leave it as-is with zero extra setup.3. Hide it entirely — a toggle to remove the subtitle for teams that don't want a second line in the header at all.4. Per-language customization — the custom text should support Intercom's multilingual settings so each supported language can have its own version, just like other Messenger strings.5. (Nice to have) Availability-aware text — the ability to vary the message based on office hours or team status, so we're not promising "the team can also help" at 2 AM when no one is online.Suggested implementation: an option set within the "Choose how Fin sets expectations" section — Default / Custom text / Hidden — with a multilingual text field when Custom is selected. This would mirror how "Start conversation button text" is already editable directly below this setting, so there's an existing pattern to follow.Thanks for considering!
We would like to be able to pull ticket data via the Intercom MCP, ticket number etc. Currently this isn’t possible. Requesting ticket data to be added to your MCP
Currently, FIN Voice supports keypad input (DTMF) only for limited use cases such as call ratings (1–5) and simple reply options (press 1, 2, etc.).We would like support for free-form multi-digit keypad input during voice conversations, for example:Entering an identity number Entering a store number Entering customer/account IDsUse case:When callers are asked for numeric identifiers, many users prefer entering the numbers on their phone keypad instead of speaking them aloud. This would improve:Accuracy User experience Accessibility Handling of noisy environments Recognition reliability for long numeric sequencesExample:“Please enter your 8-digit store number using your phone keypad.”Requested capability:Capture arbitrary multi-digit DTMF input Store the entered value as a variable in the FIN Voice flow Optionally support validation rules (length, numeric only, regex, etc.) Optional confirmation step (“You entered 12345678, is that correct?”)This would significantly improve voice automation flows that rely on customer identifiers.
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.