Intercom Guide9 min read
How to Analyze Intercom Conversations for Product Feedback and Feature Requests
Last updated: · By Matt Teixeira, Founder and CEO
On this page
To analyze Intercom conversations for product feedback, export the full threads, label each one as a bug, feature request, how-to question, or account issue, group them into specific problem themes, and rank each theme by how many companies raised it. Include the conversations Fin resolved alone, because no agent tagged them.
To find feature requests in Intercom, search what your agents write, not what customers write. Customers describe one missing feature fifty ways. Agents say "that isn't possible right now" in about five.
Key Takeaways
- The Reports CSV has no message text. Use the inbox export, the Conversations API, or an integration that reads full threads.
- Search agent replies for phrases like "not possible at the moment," then add the Fin-only conversations no agent saw.
- Fin Attributes, free on every plan, can label feature requests and bugs when Fin is in your workflows.
- Topics Explorer measures support volume, not what to build, and small, high-value requests never become a topic.
- Rank themes by unique companies and their plan. One frustrated account can open a dozen conversations in a bad week.
What Can Intercom's Native Tools Tell You?
Intercom's AI tools are built to run support, not to decide what to build. Use them to label and size volume, then make roadmap calls on company counts.
| Tool | What it does | What it needs | The gap |
|---|---|---|---|
| Tags | Agents tag a message with Shift + T; tags are reportable (source) | An agent who remembers | Misses conversations Fin handles alone |
| Fin Attributes | AI sorts conversations into values you define, such as "Feature request," for reports (source) | Any plan, no added cost; Fin in your workflows | Conversations only, not tickets |
| Topics Explorer | AI groups conversations into topics and subtopics, sized by volume (source) | Pro add-on; Fin live on a channel | Inbound conversations only; no sentiment |
Topics Explorer and its weekly Trends report need the Pro add-on: $0.20 per conversation, a 500-conversation monthly minimum, and Intercom's latest pricing model (Intercom Help). Two limits matter more than the price:
- Volume isn't value. A big topic caused by a confusing help article needs a writer, not a roadmap slot.
- Small requests stay invisible. A subtopic needs at least 15 similar conversations (Intercom Help). Three enterprise accounts asking for Okta SSO is a roadmap conversation that will never be a topic.
How Do You Export Intercom Conversations with the Message Text?
Only the inbox export and the Conversations API include message text. The other exports carry metadata (Intercom Help).
| Method | Message text? | Limits | Use it for |
|---|---|---|---|
| Reports dataset CSV | No | 10,000 rows in the browser; more by email | Counts by tag, team, or attribute |
| Inbox export (PDF or text) | Yes, minus internal notes | 200 to 300 conversations at a time | A one-off batch for an LLM |
| Conversations API | Yes, full threads | Needs a developer | Recurring analysis at volume |
As a rule of thumb, under about 300 conversations a month, the inbox export and an LLM are enough. Above that, script the API or connect a tool. For the labeling prompt and spot checks, see how to use AI to analyze support tickets and sales calls.
Step 1: Pull Full Threads, Including Fin's
Analyze the whole conversation, not the opening message. The real request often shows up in the fourth message, after the agent offers a workaround and the customer explains why it won't do.
Include the conversations Fin resolved alone. The more Fin handles, the less agent tags see, so ask any tool which conversations it reads before you trust its counts. Keep each conversation's company, plan, and tags attached.
Step 2: Find Feature Requests by Searching What Your Team Says
Your agents' replies are the best feature-request detector you already have. Inbox search matches conversation content and takes exact phrases in double quotes, but it has no OR, so run one search per phrase (Intercom Help). Start with a phrase you know an agent and Fin each used, to confirm your search finds their replies.
Agent phrases that usually mean a feature request
"not possible at the moment"
"isn't something we support"
"passed this on to our product team"
"on our roadmap"
"workaround"
"for now, you can"
Calibrate the list first: open 20 conversations you know were requests and copy the sentence the agent used. Do the same with 20 Fin-only conversations to learn how your Fin says it can't help. Then search "cancel," "downgrade," and competitor names, which surface the reasons accounts leave.
Going forward, give agents one "not supported yet" macro that also applies a tag, so new requests label themselves. A macro can add a tag only once the conversation has a reply (Intercom Help).
Step 3: Label Each Conversation by Type and Problem
Give every conversation one type and one problem theme. Four types cover nearly every Intercom inbox: bug, feature request, how-to (the feature exists, but the customer couldn't find it), and account issue, such as billing or logins, which stays out of product counts.
Name themes after a specific problem, such as "Bulk edit contacts," not an area like "Contacts." To label inside Intercom, create a Fin Attribute called "Product feedback type" with these four values, describe each plainly, and test it in the preview first. Intercom lists Fin Attributes on all plans at no additional cost (Intercom Help).
Step 4: Count Companies, Not Conversations
A theme's weight is the number of distinct companies behind it, split by plan. Twelve conversations from one account in a bad week are one signal, not twelve.
Pull company and plan from Intercom's company data, or join them from your CRM. As a rule of thumb, a theme raised by five or more companies in a month, or by two on your top plan, earns a roadmap look.
Step 5: Review Themes Weekly and Hand Them Off with Quotes
Hold a 30-minute review each Monday on new themes and the top themes by company count, with three quotes per theme linked to their Intercom conversations.
Send bugs to engineering with those links, and requests to roadmap review with the company list. To tell a rising theme from a one-week spike, see how to tell which support ticket pain points are trending.
Worked Example: A Month of Intercom Conversations
Illustrative numbers, not customer data. A B2B SaaS company has 600 conversations in a month, and Fin resolves 40% alone. Agents tagged 18 as feature requests, agent-phrase searches found 47 more, and Fin's "can't help" wording found 29: 94 request conversations from 58 companies.
| Theme | Conversations | Companies | Top-plan companies | Companies an agent saw |
|---|---|---|---|---|
| Bulk edit contacts | 26 | 4 | 1 | 4 |
| Export to Google Sheets | 19 | 15 | 5 | 4 |
| Okta SSO | 6 | 4 | 4 | 4 |
Bulk editing leads on conversations, but one account sent 12 of the 26, so it gets a help article for the workaround and a call with that account.
The Google Sheets export goes to roadmap review: 15 companies, five on the top plan, and 11 that only ever talked to Fin, so agent tags alone showed four. Okta SSO would never form a topic at six conversations, but every requester is on the top plan, so it goes to the enterprise discussion.
Tools That Help
| Option | How it reads Intercom | Watch out for | Price |
|---|---|---|---|
| Spreadsheet plus an LLM | Inbox export or an API script | You rerun and maintain it | Free, plus your time |
| Deck | 90-day backfill, then scheduled syncs into quoted insights and Themes (source) | Starter plan or above; read-only | $99/month, or $79 billed annually |
| Enterpret | Polls every 15 minutes by default and can pull conversations before they're resolved (source) | Enterprise pricing | AWS lists $120,000 a year for 300,000 records (source) |
| Productboard | All or tagged conversations, or a manual search by date, status, or tag (source) | Tags register only after a reply or note; exports drop Intercom text (source) | Business $59 per maker a month, billed annually (source) |
| Canny | Agents vote or create posts from an inbox app; Autopilot suggests posts (source) | Autopilot needs an agent reply and skips Fin-only conversations | Free for 25 tracked users; Pro from $79 a month, billed yearly (source) |
Choose Canny if agents should log requests to a public board from Intercom, and Productboard if you already plan features there. Choose Deck or Enterpret to have every conversation read and grouped without tagging, then decide on volume and budget. To weigh them against tools that also sync Zendesk and Slack, see the best tools to centralize customer feedback.
Where Deck Fits
Deck turns each synced conversation into insights in the customer's own words and groups them into Themes alongside feedback from your other connected tools.
- Sync: the last 90 days on connect, then a daily sync, with tags and custom attributes attached (Deck Intercom docs). The connect your feedback stack playbook shows how to confirm data is arriving.
- Analysis: relevance filtering, analysis, insight extraction, and a quality review that checks every quote. Fin-only conversations count too, as long as the customer wrote at least one message. Insights get a category such as Feature Request (Deck categories docs), and the Opportunity Backlog ranks problems by impact, sentiment, and trend (Deck Opportunity Backlog docs).
- Segments: rules can match Intercom tags or custom attributes, such as
customer_tier=Gold, so any view can filter to your top plan (Deck segments docs). - Credits: at Deck's planning figure of about 3 credits per support ticket, 600 conversations a month is about 1,800 credits, inside Starter's 2,500. By our estimate, the 90-day backfill adds about 5,400 up front, so the first month comes to roughly $334 on Starter with overage, or $249 on Business, which includes 7,500 credits.
- Limitations: Intercom needs Starter or above, so Free's one integration can't be Intercom. History older than 90 days comes in only by CSV, which you'd build from the API. Contact sync needs Business or Enterprise. Access is read-only, so Deck adds no tags or notes in Intercom, and agents get no Deck app in the inbox.
To see a month of your own Intercom conversations sorted into Themes, compare Deck's plans.
Frequently Asked Questions
Sources
- Intercom Help: Use the Topics Explorer to see what's driving volume
- Intercom Help: How to use Trends to spot shifts in your support data
- Intercom Help: Pro add-on
- Intercom Help: Fin Attributes
- Intercom Help: Keep track of support requests and bugs by tagging conversations
- Intercom Help: Inbox search and filter
- Intercom Help: Creating and managing macros
- Intercom Help: Export your conversation data
- Intercom developers: MCP server
- Enterpret Help Center: Intercom
- Enterpret on AWS Marketplace
- Productboard Help: Intercom integration
- Productboard Support: Export guide
- Productboard Support: Spark pricing and billing FAQ
- Canny Help: Intercom integration
- Canny pricing
- Deck Intercom integration docs
- Deck categories docs
- Deck Opportunity Backlog docs
- Deck customer segments docs
- Deck pricing
Sources checked November 2, 2026.