AI Enricher Properties (text-based)

Transform your product data with AI-powered enrichment that automatically generates new product attributes based on existing information.

AI Enricher Properties use intelligent prompts to analyze your products and create valuable new data points that enhance your catalog.

Before you begin

• Set up your product feeds and ensure data is flowing into Crobox

• Create the product properties you'll use as inputs and outputs

• Ensure you have properties with the "AI Enricher" value type available in your property list

Create your AI Enricher property

Navigate to Product Data > Properties and either create a new property or edit an existing one.

1. Set the Value Type to "AI Enricher"

2. Choose your Target Level (Product or Variant)

3. Configure basic property settings:

   • Lock being edited in the app - Prevents manual editing of AI-generated values

   • Mark as internal - Keeps the property for internal use only

   • Set validation rules if needed for your outputs

AI Enricher properties can be best organized when you have clear, descriptive property names that indicate their purpose, like "AI Generated Description" or "Style Category AI."

Configure AI Enrichment Settings

In the AI Enrichment Settings section:

1

Enable background enrichment

Toggle "Enable automated background enrichment" to activate the AI processing. When enabled, Crobox will automatically process your products in the background based on your configuration.

2

Set product filters (optional)

Use "Only for these products" to target specific product subsets:

• Apply enrichment only to new products

• Focus on specific categories or brands

• Target products missing certain attributes

• Limit processing to high-priority inventory

3

Write your prompt

Create a clear, specific prompt that tells the AI exactly what you want to extract or generate. Your prompt should include clear instructions, for example:

Extract the following properties from the title:
- the player name, never include the shirt_nr (e.g. Lionel Messi)
- team (e.g. Barcelona)
- league (e.g. La Liga)

Prompt best practices:

• Use clear, actionable language

• Provide specific examples

• Define the expected output format (including formatting requirements)

• Include constraints to avoid unwanted results

4

Configure input properties

Select the Inputs that the AI will analyze:

• Choose properties containing the source data for enrichment

• Multiple inputs can provide richer context

• Common inputs include titles, descriptions, categories, or specifications

5

Define output properties

Select the Outputs where enriched data will be stored:

• These properties will receive the AI-generated values

• Can include multiple outputs for different extracted attributes

• Output properties should already exist in your property setup

Make sure your output properties can accommodate the type of data the AI will generate (text, numbers, categories, etc.).

Test with preview

Before enabling full enrichment, use the AI Enrichment Preview to validate your setup:

1. Click "Fetch Preview Values" to test your prompt on sample products

2. Review the preview table to see:

   • Input data being analyzed

   • AI-generated output values

   • AI explanations showing the reasoning

3. Use "Add Filter" to test different product subsets

4. Refine your prompt based on preview results

The preview shows up to 20 products and includes an "AI Explanation" column that reveals how the AI interpreted your prompt and source data.

Save and monitor

1. Click Save to store your configuration

2. Monitor enrichment progress in the Insights tab:

   • View enrichment coverage statistics

   • See value distribution across your catalog

   • Check for validation errors

   • Track which experiences use this property

The insights dashboard shows exactly how many products have been enriched and provides detailed breakdowns of the generated values.

Review and refine

Check enrichment quality

• Review the generated values for accuracy

• Look for patterns in the validation errors

• Verify that outputs match your expectations

Optimize your prompt

• Refine instructions based on actual results

• Add examples for edge cases you discovered

• Adjust constraints to improve consistency

Monitor performance

Use the Insights tab within output properties to track:

• Product Value Insights - See distribution of enriched values

• Validation Errors - Identify data quality issues

• Connections - Track where this property is used in experiences

Common use cases

Product categorization

Extract categories, styles, or attributes from product titles or descriptions:

From the product title, identify the clothing style category.
Options: Casual, Formal, Athletic, Vintage, Contemporary
Only return one category that best matches.

Attribute extraction

Pull specific product details from unstructured text:

Extract these details from the product description:
- Material (fabric type only)
- Care instructions (washing temperature)
- Country of origin
Format as: Material | Care | Origin

Content generation

Create marketing copy or enhanced descriptions:

Create a concise product highlight (max 15 words) that emphasizes:
- Key product benefits
- Unique selling points
Use an engaging, sales-focused tone.

Troubleshooting

AI outputs are inconsistent

• Refine your prompt with more specific instructions and examples

• Add constraints to limit possible outputs

• Test with preview using different product samples

Enrichment isn't running

• Check the toggle - Ensure "Enable automated background enrichment" is active

• Verify filters - Make sure your product filter isn't too restrictive

• Review inputs - Confirm input properties contain data for your target products

Preview shows errors

• Validate prompt syntax - Ensure instructions are clear and actionable

• Check input data - Verify input properties have values for preview products

• Simplify outputs - Start with fewer output properties to isolate issues

Values don't appear in experiences

• Check property connections - Use the Insights tab to see where properties are used

• Verify property settings - Ensure target level (Product/Variant) matches your needs

• Review validation rules - Make sure generated values pass validation

FAQs

How long does AI enrichment take to process my products?

Processing time depends on your catalog size and prompt complexity. Small catalogs (under 1,000 products) typically complete within minutes, while larger catalogs may take several hours. The system processes products in batches in the background.

Can I use the same property as both input and output?

While technically possible, it's not always recommended as it can cause data conflicts. Create separate properties for inputs and outputs to maintain data integrity and clear audit trails.

What happens if my prompt generates invalid data?

Invalid outputs are can be caught when previewing output. You can refine your prompt and re-run enrichment to fix these issues, or "Remove enriched data" to clear existing AI-generated values if needed.

Can I stop enrichment once it's started?

You can disable "Enable automated background enrichment" to prevent new products from being processed, but you can't cancel in-progress batch operations. Use "Remove enriched data" to clear existing AI-generated values if needed.

How do I know which products have been enriched?

The Insights tab in the output properties shows detailed statistics including total products, value distributions, and coverage percentages. You can also filter the product catalog to show only products that have a value by targeting your output property.

What's the difference between AI Enricher and manual enrichment?

AI Enricher Properties automatically generates values using prompts and existing data, while manual enrichment requires individually assigning values to products. AI Enricher scales better for large catalogs but requires careful prompt engineering for accuracy and user UAT.

What's next

• Create Product Finder experiences using your enriched data for better product discovery

• Set up Campaigns that leverage AI-generated attributes for targeted messaging

• Monitor performance through analytics to measure the impact of enriched data