Go to zenloop

Support Centre

Search for any help questions or topics.

zenDictionary: AI-Powered Feedback Categorization

Avatar
Virna Harri
Updated 10 hours ago

zenDictionary automatically categorizes customer feedback using AI.

zenDictionary is zenloop's intelligent feedback categorization feature that automatically labels and organizes customer feedback using artificial intelligence. It applies relevant topics to incoming comments, helping you quickly identify patterns and prioritize improvements.


How zenDictionary works

AI evaluates each customer response and assigns up to 5 relevant topics. Each topic receives its own sentiment classification (positive, neutral, or negative).

The AI learns from:

  1. Topic titles — Your primary signal. Clear, descriptive names guide classification.

  2. Topic hierarchy — Parent-child relationships (up to 4 levels deep)

  3. Descriptions — Context about what the topic covers

  4. Examples — Real customer phrases that match the topic

  5. Sentiment indicators — Keywords signaling positive, neutral, or negative feedback



Setting up your topics

Step 1: Create clear topic titles

For most topics, a descriptive title is all you need:

  • ✅ "Lieferung & logistik" (Delivery & logistics)

  • ✅ "Produktqualität & produkterfahrung" (Product quality & product experience)

  • ✅ "Preis & erwartung" (Price & expectations)

  • ✅ "Kundenservice > Reaktionszeit" (Customer service > Response time)

  • ❌ "DL" (too abbreviated)

  • ❌ "Misc" (too vague)

Tip: Include unique business terminology in titles. If you call your installation service "Montage ABC," use that name.

Step 2: Organize with hierarchy (optional)

Group related topics for more precise categorization:

Service (Level 1)

├── Response Time (Level 2)

   ├── Email Response (Level 3)

   └── Phone Response (Level 3)

└── Staff Friendliness (Level 2)

When feedback matches a child topic, parent topics are automatically included.

Format option: Use > separators in names: "Service > Response Time > Email"

⚠️ Limit: Maximum 4 hierarchy levels supported.

Step 3: Test your setup

Before adding descriptions or examples, test with real comments:

  1. Open the "Test with comment" feature

  2. Paste a sample customer comment

  3. Review which topics the AI assigns

  4. Check the sentiment detected for each topic

If classification is accurate, you're done. Only proceed to Step 4 if you see issues.

Step 4: Add guidance when needed

When titles alone aren't enough—typically for company-specific terminology, ambiguous topics, or specialized subject areas—enhance your topics:

Descriptions

Explain the scope and boundaries of a topic.

Topic: Montage ABC
Description: "Feedback about our solar panel installation process, including scheduling, technician visits, and installation timeline"

Examples

Your most powerful tool. Add real customer phrases (one per line):

Installation took too long

Installer was very professional

Schedule was changed last minute

Monteur war pünktlich

Pro tip: Include examples in multiple languages if customers provide multilingual feedback.



Language support

zenDictionary supports 200+ languages automatically:

  • Auto-detection — The AI identifies each comment's language

  • Translation — Comments can be translated to your target languages

  • Multilingual topics — Add examples in multiple languages for better matching

Supported languages include English, German, Spanish, French, Italian, Portuguese, Dutch, Polish, Japanese, Chinese, Arabic, and many more.



Best practices

Do

  • Start simple — Let the AI work with just titles first

  • Test with real comments — Verify accuracy before and after changes

  • Use customer language — Include phrases customers actually use, not internal jargon

  • Iterate gradually — Add descriptions and examples as needed, not all at once

  • Include multilingual examples — If you receive feedback in multiple languages

Don't

  • Over-complicate — Only add details when titles alone aren't sufficient

  • Use unexplained abbreviations — Spell them out or include both versions

  • Create overlapping topics — Keep boundaries clear between related categories

  • Skip testing — Always verify changes improve accuracy

  • Exceed 4 hierarchy levels — Deeper hierarchies aren't supported



Automatic detection

zenDictionary automatically flags:

  • Email addresses — Comments containing email addresses (may need direct follow-up)

  • Questions — Comments containing "?" (may need a response)

Use these flags to filter and prioritize feedback requiring action.

Need help?

If you're experiencing persistent categorization issues or need guidance on structuring your topics, reach out to the zenloop team.

💡 Remember: The AI performs best when your topic structure mirrors how customers naturally talk about their experiences. Keep it simple, use their language, and let the AI do the heavy lifting.

😞 😐 😃