What is topic detection?
Topic detection is analysis that finds the subjects customers are talking about across many conversations, including subjects nobody thought to look for in advance.
Why it matters
A fixed list of tags only counts what a team already expects. New problems, such as a fault in a recent update or confusion about a changed policy, do not fit any existing label and can go unnoticed for weeks. Topic detection brings these to the surface from the conversations themselves, so a team can act while an issue is still small.
How it works
The system takes a large set of transcripts or messages and groups those that are about similar things, usually by comparing embeddings. Each group is given a short description of what its conversations have in common. Analysts review the groups, merge or rename them and track how their size changes over time. A group that grows suddenly is flagged.
Example
A streaming service reviews a week of chats. Alongside the usual groups about billing and passwords, topic detection shows a new cluster of several hundred conversations mentioning subtitles out of sync on one type of television. No tag existed for it. The support team passes the finding to engineering and adds a help article, and a tag is created to track it.
