How to spot churn signals in Intercom conversations

Most customers who leave a subscription say something first. They ask how to export their data, complain about a charge, rate a conversation badly, or write in three times about the same bug. In Intercom, those messages sit in the Inbox, get closed, and are rarely looked at again as a group.

This guide shows how to turn Intercom conversations into churn signals you can count and act on: which signals to look for, how to capture them with tags and conversation data attributes, how to read conversation ratings, and how to get the data out. It sticks to Intercom’s own documentation: conversation ratings, CSAT reporting, the conversation tags report, conversation data, Topics Explorer, data exports and the Conversation API reference (all accessed November 2026).

The signals worth capturing

Not every conversation says something about churn. These are the patterns that usually do. Treat them as hypotheses to check against your own cancellations, not as rules.

Signal	What it looks like in Intercom	How to capture it
Explicit cancellation intent	“How do I cancel”, “close my account”, “stop billing me”	Tag, or a conversation data attribute set by a Workflow
Billing friction	Disputed charge, unexpected renewal, refund request	Tag billing-complaint
Data exit	Questions about exporting data or deleting the workspace	Tag data-export
Low satisfaction	A bad conversation rating, often with a remark	Conversation rating (native)
Repeated contact	The same customer writing about the same issue several times	Attribute such as “Issue” plus reporting by company or user
Missing feature	“Does it do X?” followed by “we’ll look elsewhere”	Tag feature-gap with a list attribute naming the feature

The goal is a small, stable set. A list of forty tags that nobody applies consistently is worse than five that everyone uses.

Tags vs conversation data attributes

Intercom gives you two ways to label a conversation, and they suit different jobs.

Tags are quick and free-form. Intercom’s conversation tags report shows new conversations, tagged conversations, tagged conversations over time and the most used tags. Two caveats from that page matter for churn analysis:

• Conversations are dated by their start date, not the date the tag was added.
• Tag reporting is updated with a delay of 24 to 36 hours, so it is not a real-time alert.

Conversation data attributes are closer to ticket fields. Intercom’s conversation data guide explains that you create them under Settings > Data > Conversations, with a data type of text, number, true/false or a list of values. They can be set manually in the conversation, with macros, or automatically by Workflows.

For churn signals, a list attribute such as “Churn signal” with fixed values (Cancellation request, Billing complaint, Data export, Feature gap, None) is usually more reliable than tags: teammates pick from a list instead of typing, and the values are consistent in reports.

Routing cancellation intent with Workflows

Intercom’s help center on assigning conversations notes that Workflows can automatically tag conversations, mark them as priority, set conversation attributes and route to a team. A simple setup for cancellation intent:

1. Create a list attribute “Churn signal” (Settings > Data > Conversations).
2. Build a Workflow that runs when a customer sends a message. Add a branch for messages that contain phrases such as “cancel”, “unsubscribe”, “close my account” or “refund”.
3. In that branch, set “Churn signal” to Cancellation request and add a tag if you also want it in the tags report.
4. Assign the conversation to the team that handles billing or retention.
5. Review the branch every few weeks. Keyword rules catch “cancel my order” and “cancel the meeting” too, so expect false positives and let teammates correct the attribute.

One rule that should never change: if the customer asks to cancel, the team that receives the conversation cancels. Routing is there so the request is handled fast and recorded properly, not to put more steps in front of it. See our guide on click-to-cancel rules.

If your workspace runs Fin, Intercom also documents Topics Explorer, which uses AI to group conversations into topics and subtopics. Admins can create, merge, rename and delete topics, so you can shape a “Cancellation” topic by hand. Intercom lists requirements: it needs the Pro add-on, it is available only while Fin is live, and the initial model draws on inbound conversations from the last 90 days. It is separate from the older keyword-based conversation topics report.

Reading conversation ratings

Conversation ratings are Intercom’s built-in CSAT. According to Intercom’s conversation ratings article:

• After a teammate closes a conversation, the customer sees five emoji reactions (in the Messenger) or five emoji with text labels (in email), and is asked for an optional comment, called a remark.
• The rating applies to the teammate who last sent a message in the conversation.
• Group conversations, with more than one user or lead, are not surveyed.

The CSAT reporting article describes a Surveyed CSAT report template available to teammates with reporting access, including conversation ratings and remarks broken out by sentiment. Intercom suggests filtering by conversation rating to find negative conversations and work on those first. Custom CSAT charts, and using conversation rating attributes in custom reports, require the Advanced or Expert plan.

For churn, the remark is often more useful than the score. A two-out-of-five with “third time I’m asking about this invoice” tells you more than any average. Read the negative remarks weekly and label each conversation with your “Churn signal” attribute.

Two cautions:

1. A rating describes one conversation, not the customer’s whole relationship. A happy rating on a cancellation handled quickly is a good outcome, not a sign the customer is staying.
2. Response bias is real. Only some customers rate. Do not compare small samples across teammates or months as if they were precise.

Getting the data out

To link support signals to who actually cancelled, you need the data outside Intercom. Intercom’s export article and related help pages describe several paths:

Path	What you get	Limits Intercom documents
Reports > Dataset export	CSV of conversation metadata: IDs, timestamps, attributes, metrics	No full message content; browser downloads capped at 10,000 rows, larger ones emailed
Inbox export	Transcripts as text or PDF	A few hundred conversations per operation
Conversations API	Full conversation objects, including tags, custom attributes and the rating	Rate limits and pagination apply

The Conversation object in the API carries tags, custom_attributes (your conversation data attributes) and conversation_rating, which holds a rating from 1 to 5 and an optional remark. That is enough to join Intercom data to your billing data by customer email or user ID.

A weekly routine

1. Monday: open the Surveyed CSAT report, filter on negative ratings from the past week, read every remark.
2. Set the “Churn signal” attribute on any conversation that shows one and that the Workflow missed.
3. Export the week’s conversations with a churn signal (Dataset export or API).
4. Match them to your subscription data: who cancelled, who downgraded, who failed a payment in the following weeks.
5. Once a month: count which signals came before cancellations most often, and check whether the reasons customers give in support match the reasons they give when cancelling. Our guide on finding what predicts churn explains how to read that without confusing correlation with cause.

Keep the customer’s words as they wrote them. A quote is evidence; a summary written later is an opinion.

Doing this with NoChurn

NoChurn’s Intercom connector is read-only: it reads conversations and never replies, tags or changes anything in Intercom. Conversations are matched to your customers by email; doubtful matches are suggested for you to confirm and never merged automatically. Matched conversations feed the churn score (0 to 100) and the churn drivers view, and they are kept as dated, verbatim evidence that can be attached to a chargeback response on Stripe or Braintree, which is drafted by AI and sent only after you review it. Tagging, routing, ratings and replies stay in Intercom. See NoChurn + Intercom.