AI classification of return reasons

Optionally, an AI can automatically assign free-text return reasons to a category – e.g. "The jumper was way too small" → Wrong size. This makes even unstructured feedback analysable.

Setup

1. Open the Settings tab and the AI Pro features (AI classification) section.
2. Choose the AI provider: Anthropic Claude or OpenAI.
3. Add your API key:
   • Anthropic starts with sk-ant-
   • OpenAI starts with sk- or sk-proj-
4. For Anthropic you can additionally choose the Claude model (Haiku is fast & cheap, Sonnet more precise).
5. Save.

Test connection

Use Test connection to verify with a minimal call whether your key is valid. After saving, the key is displayed masked for security reasons (••••••••) – the test then automatically uses the stored key.

When is it classified?

• Automatically in the nightly scheduled task, right after the recalculation.
• Manually and immediately via the "Classify free-text reasons now" button in the settings – handy after a data import or for testing.

Only reasons with free text that have not yet been assigned an AI category are classified. Structured reasons (fixed codes) are left untouched.

Privacy & cost

The cost depends on the provider and model and is very low for small models (e.g. Claude Haiku, GPT-4o mini). A rough estimate is shown in the settings.

Result

After classification the reason is assigned to the category and visible in the Return reasons tab. Internally the confidence value and the model used are stored as well.

Next steps

• Return reasons
• Configuration