Classify customer feedback with Jev

Can Jev classify customer feedback?

Yes. A Choice question asks Jev to select one of your categories and return probabilities across the options. Use that answer to suggest an initial support queue for a person to review. It does not verify an account, assess urgency, or approve a refund.

Open the editable feedback template. The full fictional input, question and six options appear above. Loading them creates a draft; a check starts only when you request it. This template asks one question, while the illustrated feedback scenario asks three.

Give each category a clear boundary

The template starts with labels and empty descriptions. You can add these suggested descriptions in the editor, then adapt them to your inbox; they are not automatically loaded:

If an app crash also causes a payment problem, decide whether the first destination follows the reported fault, the requested remedy, or the investigating team. Write that priority into the question and descriptions. The editor allows 2–6 options; this is a website limit.

Change one thing and compare

1. Load the template and save the input, question, options and any descriptions you add.
2. Check a non-sensitive message whose handling you can judge yourself. Record the selected category and the other probabilities.
3. Change only the message from a duplicate charge to a login crash. Keep the question and categories fixed, then compare the two responses with your own routing rule.

A concentrated probability distribution is not measured accuracy. This template does not assign a real ticket, and no expected answer is prefilled.

Two recorded checks from our English demo

On 18 September 2026, our original three-question English demo produced these live responses. The table preserves the exact English input and selected category:

The result text was saved at 03:05:24 and 03:07:12 UTC, respectively. These were two fictional messages, with one successful run each; they are not an accuracy benchmark. The exact model version was not recorded. These observations came from the illustrated scenario, not the editable one-question template above, and do not guarantee the same result in a new check. Published screenshots show the input changes and results.

What if the message fits two labels?

Try this new fictional input: “My subscription was charged twice, and the app crashes when I log in.” Decide the human handling before running it. If both issues need owners, one winning category cannot represent both assignments. Split the message into separate tasks or check each property with a separate yes/no question. Keep an “other” or human-review route for messages your policy cannot resolve.

See how to handle overlapping labels, then edit and test your own case. Account checks and any refund decision remain separate from this classification.