Intercom Conversation Summarizer and Tagger

Overview

This flow turns raw Intercom support conversations into searchable, categorized knowledge using Azure OpenAI. On a schedule it finds recently closed conversations, sends each transcript to a chat-completions deployment that returns a concise summary, topic category, resolution type, and sentiment as JSON, writes that summary back onto the conversation as an internal note, and stores the structured result in a Dataverse table for reporting.

Why it matters: Transcripts are long and unstructured. An AI summary plus consistent tagging makes conversations skimmable, reportable, and minable for trends - without agents writing wrap-ups by hand. Ships Off; going live requires only connection authorization and filling the env var values.

Use Case

A support team wants every closed Intercom conversation auto-summarized and categorized, with the summary visible on the conversation and structured data stored for analytics (volumes by category, resolution mix, sentiment trends).

Flow Architecture

Environment Variables

Connectors & Connections

Connector	API name	Actions used
Intercom	shared_intercom	POST /conversations/search GET /conversations/{id} POST /conversations/{id}/reply
Azure OpenAI	shared_azureopenai	ChatCompletions_Create_2024Feb15Preview
Microsoft Dataverse	shared_commondataserviceforapps	ListRecords CreateRecord

Note — All connections are referenced as solution connection references; the flow is portable between environments as long as a connection is mapped at import time.

Customization Guide

Almost every realistic variant of this flow can be implemented by changing environment variable values. A few cases require small edits inside the flow definition — those are called out explicitly below.

Cadence / window

Match Initialize_Lookback_Hours to the Recurrence interval (both default ~hourly/24h) to avoid gaps or rework.

Event-driven option

If you only need new conversations, the connector's TrigNewConversation trigger can replace the Recurrence poll; the close-state poll is required for on-close semantics.

Auto-tag

Extend the note step (or add a tag call) to push the category back to Intercom as a native tag for filtering.

Category set

Tighten the allowed categories inside the system prompt variable.

Trend reporting

Point Power BI at the summary table for category/resolution/sentiment dashboards.

PII

Add a redaction step before the model call if policy requires.

Key Expressions

The flow is intentionally light on Power Fx / WDL gymnastics — the heaviest expressions are the branch-name concatenation and the approval outcome check. They are listed below in the order they appear in the flow.

@sub(div(sub(ticks(utcNow()),ticks('1970-01-01T00:00:00Z')),10000000), mul(variables('varLookbackHours'),3600))

@coalesce(body('Search_Closed_Conversations')?['conversations'], json('[]'))

@body('Generate_Summary_With_Azure_OpenAI')?['choices']?[0]?['message']?['content']

@addSeconds('1970-01-01T00:00:00Z', int(coalesce(items('For_Each_Conversation')?['updated_at'],0)))

Customize & download