Free tool · Feedback Classifier
What is your feedback actually telling you?
Upload app reviews, support tickets or survey answers. The classifier finds the friction, the feature requests, and the jobs and needs behind them — sentence by sentence.
Runs entirely in your browser. Your text is never uploaded, stored or seen by anyone.
1 · Your data
Drop a CSV file here, or choose a file
One row per review or answer · up to 2,000 rows · German and English
No file at hand?
EnglishDeutsch FrançaisEspañol ItalianoNederlands
2 · Check the setup
3 · Analysing
4 · Results
Results
Row by row
The first 100 rows. The download has every row, the scores and the sentence behind each label.
Beyond 2,000 rows
Need this on your full data, with themes and drivers?
This free version stops at 2,000 rows and four feedback types. For larger volumes I build classifiers trained on your own categories, add themes and what drives your ratings, and set it up so your team can rerun it every week.
How it works
What the labels mean
- Friction Ready
- A bug, an interruption, or something that needs a usability fix. Asking for a bug to be fixed counts here.
- Request Ready
- A new feature, or a change to how the product is designed to work.
- Job Beta
- What the product is hired to do: the task the person is getting done (“I use it to plan weekend rides”).
- Need Beta
- The person's own situation or goal that is currently unmet; often the reason behind a request.
- Sentiment
- Positive, negative or mixed for the whole text. Neutral is rare in reviews and shown with low confidence.
How good is it?
Tested on reviews it never saw, against labels coded by hand. The score is F1, from 0 to 1 (1 = every label found and none wrong). Labels below 0.6 are marked beta: useful for exploring, worth a second look.
| Label | German | English |
|---|---|---|
| Friction | 0.89 | 0.93 |
| Request | 0.63 | 0.64 |
| Job | 0.56 | 0.71 |
| Need | 0.52 | 0.50 |
| Sentiment (positive / negative) | 0.83 / 0.78 | 0.87 / 0.82 |
Model: multilingual-e5-small (open source), with classifiers trained on about 4,900 public app reviews in German and English. Each review is split into sentences, and the page shows the sentence behind each label.