Why tag feedback at all
A pile of feedback feels like noise until you count it. Tagging each item by topic, sentiment and urgency turns it into a list you can sort: which problems come up most, which ones are angry, and which ones mean someone is about to leave. It also stops the loudest recent message from setting the roadmap.
What each label means
- Topic. Bug, feature request, pricing or billing, UX or usability, performance, onboarding or docs, praise, or other. One per item, with how sure the model was.
- Sentiment. Very negative to very positive, on a five-step scale.
- Urgency. Low, medium or high. High means something is broken for the customer, money is at stake, or they sound ready to leave.
- Churn risk. The chance the item mentions leaving, cancelling or switching.
How to use the results
- Start with the churn-risk and most-urgent lists. Those are people to reply to today.
- Look at the topic counts. A topic that keeps coming up is a pattern, not a one-off.
- Read the top themes and their quotes, then check a handful of items in the table yourself.
- Export the CSV and keep it next to last month’s, so you can see whether a fix changed anything.
Counts only tell you what people wrote about, not how many customers feel the same. Five complaints from your biggest accounts can matter more than fifty from trial users. Treat the output as a starting point for reading, not a replacement for it.
Before launch, you can test the problem itself with the startup idea validator. Once you know the questions customers keep asking, turn them into an FAQ with the FAQ generator.
How it works
Jev 1.13, TypeSafe’s decision model, answers four questions about every item and returns calibrated probabilities rather than text, via OpenRouter. We send items in small batches so long lists stay fast. If you ask for themes, Claude Haiku 5.5 reads the whole list and points at the items that show each theme. Nothing is stored.