User Feedback Analysis Agent

Overview

A wealth-management fintech was running user discovery and survey calls, but the feedback lived in recordings nobody had time to process. The system turns every call into classified, prioritized, backlog-ready items with a human approval gate before anything hits Jira.

The challenge

User feedback was being collected and then lost.

1. Processing one call, transcribe, classify, create tickets, took 3 to 4 hours of manual work.
2. Classification was inconsistent between people, and insights slipped through.
3. Booking, recording, and the product backlog lived in three disconnected places.

The approach

1. Every booked call is recorded and transcribed the moment it happens.
2. Discovery and survey calls are identified by title and analyzed in full.
3. Each call returns a structured verdict in 4 fixed buckets: issues, feature requests, pain points with a priority level, and what users love.
4. On approval, every item is pushed into Jira under a matching label. Nothing enters the backlog unreviewed.
5. A query layer over the full transcript base lets the team interrogate raw calls directly, with a daily reporting cycle in Slack.

The results

1. Feedback processing reduced from 4 hours to about 20 minutes per call
2. Every call transcribed and categorized into 4 fixed buckets, no human effort
3. Approval-gated Jira routing with matching labels on every ticket
4. 24-hour reporting cycle via Slack
5. Zero missed discovery calls