Major

Intercom Conversation Tagging Agent for Weekly Reporting and Bug Filing

Learn how an Intercom conversation tagging agent automates tagging, reporting, and bug filing across Linear, Slack, and Notion.

Ritvij Gautam
Major x Intercom logo lockup

An Intercom conversation tagging agent helps teams categorize closed conversations by topic and product area, rolls up tags for weekly voice-of-customer reports, and files repeat bug reports in Linear with conversation context. This guide explains how to design and build this workflow, connecting Intercom, Linear, Notion, and Slack using Major.

This guide covers one Intercom process end to end. For the Intercom API guide, connection setup and the full list of Intercom use cases, see More Intercom use cases.

Why This Process Breaks Today

Tagging conversations in Intercom, building voice-of-customer reports, and ensuring bug patterns are documented usually break down for three reasons:

  • Manual tagging is inconsistent: Support reps may not tag, or may apply tags unevenly
  • Reporting is time-consuming: Exporting tags and compiling reports across product areas sucks up hours each week
  • Bug filing gets skipped: Repeated issues can go unnoticed, and even when caught, evidence is often missing or incomplete

Without better tooling, useful feedback gets trapped in Intercom, or falls between the cracks.

The Systems and Data Involved

The process touches several specific Intercom API entities and events. Knowing where information lives is critical:

Intercom

  • Objects:
  • conversation, The core message object. Contains all messages, participants, timestamps, custom attributes
  • tag, Tags applied to conversations with POST /conversations/{id}/tags
  • Events:
  • conversation.closed, Event fired when a conversation in Intercom is closed
  • tag.created / tag.deleted, Events for tag lifecycle (optional)

Linear

  • Issue, Each bug report is a Linear issue, created with its specific API (e.g. POST /issues)

Slack

  • Pushes notification to a channel (using webhook URL) when bugs are filed or new weekly reports are posted

Notion

  • Stores the weekly voice-of-customer report as a new page in a database via the Notion API (e.g. POST /pages)

Code, Agent Judgment, Human Approval

It helps to split the work into what can run in deterministic code, versus what needs judgment or manual approval:

Code Handles

  • Listening for closed conversation events
  • Pulling full conversation history and existing tags
  • Storing tag tallies
  • Posting summaries to Slack and Notion
  • Pushing new Linear issues

Agent Judgment Handles

  • Reading conversation transcripts
  • Deciding topic and product area tags based on content
  • Spotting repeated bug descriptions (matching against existing known bugs)
  • Drafting Linear bug reports from customer descriptions

Human Approval Handles

  • Reviewing weekly voice-of-customer reports before they are pushed to Notion or Slack
  • Approving drafted Linear bug issues for submission, especially to check for duplicates

Worked Example: One Conversation End-to-End

Let’s walk through a single record to see how each component works together:

Example Intercom Conversation

  • Conversation ID: conv_7482
  • Content: Customer reports that “the Chrome extension crashes when clicking Save in dark mode.” This is the third report this week referencing Chrome extension crash.
  • Tags (before): None

Process Walkthrough

  1. User closes the conversation in Intercom.
  2. Workflow triggers on the conversation.closed event.
  3. The agent reads the transcript.
  • Decides this conversation relates to the topic extension-crash and product area chrome-extension
  • It applies both tags to the closed conversation via Intercom’s tagging endpoint.
  1. Agent notes this is a repeated report.
  • There are two prior conversations tagged extension-crash this week with similar context.
  1. Agent drafts a bug report:
  • Title: “Chrome extension crashes on Save in dark mode”
  • Description includes snippets from all three similar customer conversations, with Intercom URLs for context
  • Agent submits this draft for human approval
  1. Manager reviews in Major.
  • Approves the draft, providing any edits
  1. Major workflow creates a new bug in Linear using project/team mappings
  2. Workflow updates Slack.
  • Posts a message to #product-bugs: “New bug filed: Chrome extension crashes on Save in dark mode” with a link to the Linear issue and related Intercom conversations
  1. Weekly, the workflow tallies tags.
  • Groups topics by product area, counts occurrences, and composes a Notion page report for review
  • With manager sign-off, page goes to Notion, and Slack gets a summary link

How to Build This Process on Major

Major is designed to bridge tools, coordinate code, agents, and human reviewers for processes like this.

Apps

  • Intercom App
  • Configure credentials with least privilege (read conversations, write tags)
  • Listens for conversation.closed events
  • Linear App
  • Credentials for issue creation
  • Slack App
  • Webhook or app credentials for posting to channels
  • Notion App
  • Configure for writing to your reporting database

Agent

  • Tagging and Bug Classifier Agent
  • Instructions: For each closed conversation, read the transcript. Assign one or more topic and product area tags based on the Intercom taxonomy doc (provided as a system prompt in Major). If the topic matches two or more recent closed conversations with the same keywords, draft a Linear bug report with quotes and links. Send all tagging and bug report drafts for human approval by default. Do not tag non-support conversations (e.g., marketing outreach).

Workflow

  • Trigger: On Intercom conversation.closed event
  • Sequence:
  1. Pass transcript to the Tagging and Bug Classifier agent
  2. Agent returns proposed tags and, if needed, a bug report draft
  3. Apply tags to the conversation via Intercom API
  4. Human reviews new bug drafts in Major’s review queue
  5. On approval, workflow creates a Linear bug (via Major app)
  6. Pushes evidence-linked message to the right Slack channel
  7. On a weekly basis, aggregates tag counts for the past week
  8. Drafts a Notion page for review
  9. Human signs off and Major posts the final summary to Notion and Slack

Review Queues and Approvals

  • All agent bug filings and reports are routed through Major’s review UI for responsible approval before they affect external systems

What to Measure

Tracking these metrics clarifies process health and accuracy:

  • Number of conversations tagged by the agent vs. left untagged
  • Precision of agent-assigned topics (human QA feedback score)
  • Number of repeated bugs identified and filed in Linear, per week
  • Average time from customer message to bug filing
  • Weekly report review time and accept/reject rate
  • Number of manager edits vs. agent drafts

Guardrails: What the Agent Must Never Do

To protect quality and avoid noisy or harmful output, instruct the agent to:

  • Never tag conversations that are not closed or resolved
  • Avoid over-tagging, stick to a maximum of two topics per conversation or defer to human reviewers
  • Never draft a bug report from a conversation with incomplete or ambiguous information
  • Never include customer PII in bug reports or summaries unless specifically allowed
  • Never file bugs or reports directly, always require human signoff
  • Never tag or process outbound marketing, sales, or non-support conversations

Frequently asked questions

What permissions does the Intercom app need on Major?

The Major Intercom app should have permission to read conversations, write tags, and access the conversation closed webhook event. Avoid granting broader admin or inbox management privileges unless necessary for other workflows.

How does the agent distinguish between real bugs and unrelated issues?

The agent is prompted with a taxonomy of valid topics and product areas. It looks for repeated patterns and clear evidence in the conversation transcript. If information is ambiguous, it marks the draft for human review rather than proceeding.

How are duplicate bug filings avoided?

Before drafting a new issue, the agent checks recent tags and prior bug filings for similar topics or matching conversation content. All bug drafts must pass human review in Major before submission to Linear.

Can this process handle conversations in languages other than English?

The agent’s language abilities depend on its underlying model settings. If support covers multiple languages, provide explicit instruction and sufficient taxonomy coverage for each relevant language. Always review non-English reports carefully.

Start building on Major

Connect Intercom, describe the process, and Major builds the app that holds its state, the agent that makes the judgment call and the workflow that runs it, with scoped access and an audit log from the first deploy.

Frequently asked questions

How do Intercom AI agents work with Intercom’s API?
AI agents connect to Intercom’s API endpoints like Conversations and Tags to monitor, analyze, and act on customer data. These agents can run stand-alone or as part of larger orchestrations built in Major.
Can I use Intercom’s built-in AI features with Major?
Yes, you can combine Intercom's built-in AI features with external agents orchestrated by Major. This setup lets you decide which tasks stay inside Intercom and which are handled by Major's agents for additional oversight or cross-system actions.
What authentication is required to connect to Intercom via API?
You need either a workspace Access Token or to authenticate through OAuth 2.0 with a Bearer token. Access is managed per Intercom workspace or app installation.
Where does Major hold state and approvals when automating Intercom?
Major holds all app state, credentials, approvals, and audit logs. Intercom remains unchanged as the source of customer messaging and data.
Which workflows should run in Intercom vs. Major?
Simple rules like message routing can run inside Intercom. Cross-system workflows, sensitive tasks requiring approvals, and anything needing state or audit logs should run in Major.