Claude reads every ticket
The Theylore pipeline runs on the Claude API in production today. About 1.2M tokens a day go through it, across 2,300 teams and 38M processed items.
The pipeline
- Ingest. Items arrive from your sources and are cleaned, de-duplicated and split by language.
- Classify. Claude returns structured JSON for each item: theme, product area, sentiment, urgency (1-5), churn-risk signal and a one-line reason.
- Look up. Using tool use, Claude calls Theylore functions to fetch the account's plan, recent items and open flags before it decides on a churn-risk label.
- Cluster. Items are grouped into themes and ranked by volume and week-over-week change.
- Write. Claude drafts the weekly digest from the ranked themes and cites the items behind every claim.
Structured output and confidence
Every label is validated against a fixed schema. Items below the confidence threshold (90% by default) go to a review queue. Average confidence over the last 30 days: 96.1%.
Your data
Items are sent to the Claude API for analysis only. We do not train models on your data. See the Privacy Policy.
Related: Themes · Churn risks · Weekly digest