Top questions

What it shows

A ranked list of conversation clusters in your window. Each cluster is a group of conversations sharing the same underlying topic. Top of the list = highest volume.

Topics are detected from conversation content, not from tags or intents declared in Intercom. You don't need to maintain a taxonomy.

How clustering works

Each new conversation is summarized to a topic vector. Vectors are grouped into clusters and re-balanced daily. A cluster needs at least 3 conversations on the same topic in the window to surface.

The cluster title is generated from the dominant phrasing customers use, not from internal jargon. "Why am I being charged twice?" beats "Double billing edge case".

Reading a cluster

Each row shows:

• Volume — number of conversations in the cluster over the window.
• Sentiment mix — share of positive, neutral, negative. A high-volume cluster with mostly negative sentiment is your priority signal.
• Trend — direction over the last 30 days. Sparkline arrow up = rising pain.
• Doc match — link to the related Help Center article, or a gap badge if missing.

Click a cluster to see all the conversations grouped into it, the average quality score on the topic, and the suggested article draft if there's a gap.

Excluding clusters

Some clusters are noise — billing disputes routed to Finance, onboarding calls scheduled via support, opportunistic sales questions. Mark them as Off-topicfrom the cluster detail view and they drop out of the ranking. The conversations stay analyzed, they just don't count toward your operational top-N.

Workflows

Monthly product review

Open the last 30 days, filter to clusters with rising trends, and export the top 10 with sample conversations. Hand it to product. A rising cluster is feedback, not a coincidence.

Quarterly Fin tuning

Sort clusters by volume and look at the average quality score per cluster. Topics where Fin scores below 70 are candidates for a scripted answer, a new article, or removal from Fin's scope.