The Topics Explorer uses AI to automatically group your support conversations into topics and subtopics. It shows what customers are asking, how those conversations are performing, and where to focus.
See what’s driving volume—no tagging required: AI topics and subtopics are generated automatically, giving you a live view of what customers are asking without any manual effort.
Track performance by topic, not just team: Each topic includes key metrics like CX Score, resolution rate, and handling time—so you can see which issues are handled well and which need attention.
Catch problems early, before they escalate: Monitor changes in volume and sentiment over time to spot emerging issues and act before they grow.
Focus where it matters most: Identify high-volume, poor customer experience topics and make targeted improvements that move the needle.
Note:
Topics Explorer requires the Pro add-on and is available for workspaces hosted in the US, EU, and AU regions.
The Topics Explorer is unable to support the following languages for and names will continue to display in English: "Swahili" "Bengali", "Bosnian", "Azerbaijani", "Persian", "Sinhala" and "Tamil".
Pro tip: You can now tailor your AI Topics to better reflect your business terminology and reporting structure with Topic Curation. This allows you to rename, merge, and move topics and subtopics. You can access this feature by clicking Manage topics in the top-right corner of the Topics Explorer page. Learn more about Topic Curation.
Understanding AI-powered topics and subtopics
AI topics use machine learning to group conversations into topics (broad themes) and subtopics (highly specific, recurring issues).
How AI topics and subtopics are discovered
When Topics is first enabled for your workspace, the system analyzes historical conversation data from the past 90 days to build the initial topic model. It looks for patterns in the questions customers ask, then groups similar conversations together.
After initial setup, a daily pipeline runs and assigns topics to all newly closed conversations and this runs indefinitely. So customers who have had Topics enabled for more than 90 days will have data going further back than that initial window.
Subtopics are discovered first by clustering similar questions from past conversations.
These subtopics are then grouped into broader topics.
Finally, the system automatically generates clear topic titles to help you quickly understand each topic and subtopic.
Note: Topics and subtopics are not based on predefined keywords. Any keywords shown in the product are only there to help explain what each topic is about.
How conversations are assigned to AI topics
Once topics and subtopics are discovered, the AI analyzes the entire conversation transcript—including both customer messages and teammate replies (such as macros or saved replies)—and takes two steps:
Backfilling: This step involves identifying and organizing subtopics from past conversations (the last 90 days). Conversations are retroactively assigned to one or more discovered topics and subtopics.
Inference: Each day, the system reviews tickets/cases that have recently closed and assigns them to the relevant topics.
Important: The AI analyzes the complete conversation transcript when assigning topics. This means that teammate replies, macros, and saved replies can influence which topics a conversation is assigned to. If your team uses standard responses that mention specific topics or keywords, be aware that these may affect topic categorization.
Conversation criteria for generating AI topics
To build accurate topics and subtopics, the system uses conversations that meet certain criteria:
Conversations must not be marked as spam.
Conversations must have at least two participants (for example, a customer and Fin or a teammate).
Each conversation is summarized into up to three key questions, which are used to identify patterns and assign to a subtopic.
At least 15 questions or conversations are needed to form a meaningful subtopic.
Note: If your conversations are too varied, or if there isn’t enough volume around a single theme, no topics may appear—even if there are many conversations.
Ongoing updates to AI topics and subtopics
Topics/subtopics are built to adapt:
Daily updates ensure new conversations are categorized promptly.
New topics and subtopics are added as they emerge, without removing or changing the ones already discovered.
Some conversations may not get assigned to any topic if they are too different, low quality (like spam), or don’t meet the criteria.
Tip: You can also customize AI Topics for more control over how these are defined and applied to conversations.
Why you may have many smaller topics
It’s common to see a few large topics with lots of subtopics and conversations within them, and many smaller topics with only a few subtopics and conversations. That’s because:
Some topics come up frequently across customers, while others are highly specific or niche and don’t fit well with existing clusters.
The system avoids combining unrelated subtopics just to form larger topics—it focuses on natural groupings.
Note: AI topics and subtopics do not:
Detect spam
Analyze sentiment
Determine if a query is informational or requires action
How to use AI-powered topics and subtopics
Support leaders and teams can use AI topics to understand what’s driving volume and how to prioritize efforts to optimize their support.
Spotting topic trends
To view the Topics Explorer, go to Fin AI Agent > Analyze > Topics Explorer. Here, you’ll see two main sections:
The left side has a tree map of topics:
The size of the box signals the volume of conversations in that topic.
The color of the box is related to the metric selected.
In light mode, the darker colors signal areas needing attention. In dark mode, lighter colors signal areas needing attention.
The right side has a series of ridge line charts: These take the same topics from the tree map and show their performance over time.
Select how many topics you want to display and choose which metric to use:
Fin involvement rate
Fin resolution rate
Median handling time
Median first response time
Focus where it matters most by identifying high-volume, poor customer experience topics and click on them to see the tree map and line charts broken down by subtopics. This enables you to make targeted improvements to the most impactful subtopics by addressing the root cause of the volume and poor CX.
Catch problems early, before they escalate by monitoring changes in volume and key metrics over time to spot emerging issues and act before they grow. For example, the chart below shows a sudden spike in volume with negative CX Scores for the “Account locked” topic. This could highlight a bug or unexpected issue that’s preventing customers from accessing their account.
Hover over a topic/subtopic to see a description of what’s included in that topic.
From the conversations view, you can quickly click through conversations to identify issues and use the CX Score to understand how they were resolved. You can also open a conversation in the inbox to reply to the customer directly.
Tip: Don't want to manually monitor changes? Trends automatically scans your topics weekly and surfaces the biggest shifts in volume, resolution rate, and CX score, along with clear explanations. Learn more about How to use Trends to spot shifts in your support data.
Identifying areas to optimize
These topics also appear in the Recommendations dashboard to help you prioritize efforts in improving Fin across missing content, access to data, and ability to take action.
Tip: Review the recommendations weekly to improve Fin’s performance.
Filtering other reports
You can use the AI topics/subtopics to filter your other Intercom reports too. Simply add a filter for AI Topic or AI Subtopic to select specific topics you want to filter by.
FAQs
How do new AI topics get generated, and will it recategorize existing conversations when this happens?
How do new AI topics get generated, and will it recategorize existing conversations when this happens?
New topics are generated through machine learning analysis of historical conversation data from the past 90 days. Subtopics are identified first by clustering similar questions, then grouped into broader topics. Importantly, new topics and subtopics are added without removing or changing existing ones.
Why do some conversations not have topics?
Why do some conversations not have topics?
Some conversations might not appear under any topic if they:
Are too varied or don’t have enough volume around a single theme.
Are too different from existing topics.
Are low quality (e.g., spam).
Is there a way to search AI topics and subtopics?
Is there a way to search AI topics and subtopics?
Yes, you can filter other Intercom reports by AI Topic or AI Subtopic. This lets you search and narrow down data using specific topics identified by the AI.
When will I start seeing AI topics/subtopics?
When will I start seeing AI topics/subtopics?
You’ll begin to see topics/subtopics after deploying Fin. Your workspace needs to have more than one eligible conversation. However, even if your workspace meets these criteria, AI topics may not appear right away. Here’s why:
Topic generation is part of a pipeline that updates periodically. If your conversations qualify, they will be included in that pipeline.
Some customers start seeing topics after just 30–50 conversations, while others may need more to generate a related topic.
Once your workspace accumulates enough qualifying conversations, topics will begin to appear automatically as the pipeline processes new data.
Why do my AI topics/subtopics change over time?
Why do my AI topics/subtopics change over time?
Topics and subtopics are updated daily to include new conversations. As patterns evolve or new issues emerge, new topics are added, though existing ones remain unchanged. This ensures a live and accurate reflection of current support trends.
What does the color size/scheme mean on the Topics Explorer tree map?
What does the color size/scheme mean on the Topics Explorer tree map?
Size of each box = volume of conversations.
Color of each box = value of the selected performance metric (e.g., CX Score, resolution rate, etc).
In light mode: Darker colors indicate areas needing attention.
In dark mode: Lighter colors indicate areas needing attention.
How are AI topics/subtopics different from other topics in Intercom?
How are AI topics/subtopics different from other topics in Intercom?
AI topics/subtopics: Automatically group your support conversations (no manual tagging or set up) to show what customers are asking, how those issues are impacting KPIs, and how to fix them.
Conversation topics: Allow you to organize conversations through defining relevant keywords and phrases your customers use to talk about a topic, and then constantly iterating to narrow or broaden the keywords to capture all conversations in the relevant topic.
Fin Attributes: Enables Fin to classify conversations by topic, sentiment, or other chosen criteria that you define (not just topics). It does not automatically generate suggestions to improve Fin.
What are the minimum requirements for topics to start showing up?
What are the minimum requirements for topics to start showing up?
For topics to be generated, a few conditions must be met:
Timeframe: The system uses conversations from the last 90 days to build the initial topic model when Topics is first enabled. After that, the daily pipeline runs indefinitely so your Topics Explorer will show data beyond 90 days over time.
Conversation Type: It only includes inbound conversations that are not marked as spam and have at least two participants (e.g., a customer and a teammate/Fin).
Minimum Volume: There must be enough conversation volume to find meaningful patterns. If your inbound volume is low (e.g., only a few hundred conversations in 90 days), it may not be enough to create a topic cluster.
What happens if I reset my topics?
What happens if I reset my topics?
Resetting your topics will trigger a fresh backfill using only the last 90 days of conversations. Any topic assignments on conversations older than 90 days will be permanently lost and cannot be restored automatically.
If you want to preserve historic topic data, you should keep your currently assigned topics and instead use Topic Curation to delete or merge any duplicates.
Important: A full retroactive topic assignment for conversations older than 90 days is not possible without engineering intervention. There is no self-serve tool to force this.
Why is the Topics Explorer empty even though I have thousands of "migrated" conversations in my workspace?
Why is the Topics Explorer empty even though I have thousands of "migrated" conversations in my workspace?
The Topics Explorer only considers inbound conversations to build its topics model. It does not analyze conversations that were migrated or ingested from another source. Even if you have thousands of historical conversations, the explorer is looking for recent, inbound interactions (from a customer to Fin or a teammate) to find meaningful topics.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts