Warning: The AI Search RNN (legacy) version has been succeeded by AI Semantic Search, which represents the latest generation of the feature.
Note: We strongly recommend implementing the Front-End Tracker before beginning AI Search RNN (legacy) preparations and implementation. The tracker captures the search interaction data required to support potential future re-trainings of the AI Search model, helping maintain and improve its relevance over time, as well as tracking performance of the feature against business KPIs. Moreover, the Front-End Tracker enables A/B-Testing for AI Search RNN (legacy).
To activate AI Search RNN (legacy) for your eCommerce environment, we require specific data to fine-tune and configure your personalized AI model. This ensures the model is accurately aligned with your product catalog and customer search behavior.
Once you provide the required data in the appropriate quality and format, your Crownpeak Technical Consultant will handle the setup of AI Search RNN (legacy) and notify you once it is ready to be used as part of your live search configuration.
Required Data
Product Data
Your product catalog forms the basis for interpreting search queries and returning relevant results. It is used to fine-tune the model underlying the AI Search RNN (legacy) functionality.
| Property | Requirement |
|---|---|
| Number of products | Minimum 5,000 |
| Format | CSV or JSON |
| Quality | Products should be clearly labeled with accurate categories, attributes, titles, and relevant descriptions |
| Images | Each product must include a URL to a corresponding image |
Analytics Data (Search History)
This data powers the AI model by establishing links between customer search terms and the products they interacted with.
| Property | Requirement |
|---|---|
| Columns | - Search term - Product clicked (product ID, product URL, or other matchable identifier) - Search frequency (click count per search term-product pair) - Language (optional; needed if multiple languages are present) |
| Format | CSV (can be split across multiple files) |
| Encoding | UTF-8 |
| Number of rows | Minimum 1 million rows per language |
| Quality | - Raw, unprocessed search terms (no spelling corrections or cleanup) - Natural language phrases only (not product IDs) - Spaces must be preserved between words - Bots and non-human traffic should be filtered out as much as possible - Products must be matchable to items in the product data |
Image Data
Hhigh-quality image data improves the effectiveness of AI Search RNN (legacy), particularly in features like visually similar recommendations.
| Property | Requirement |
|---|---|
| Resolution | Minimum 1000 pixels on the longer side (width or height) |
| Number of images | Recommended average of 3 images per item |
| Format | PNG or JPG |
| Access | Images must be accessible for download (whitelisted IP or header-based access if needed) |
Next Steps
Once the required data is provided:
The Crownpeak team will validate the data and set up your AI model.
You will be notified when AI Search RNN (legacy) has been successfully integrated into your search configuration.
No manual configuration is required on your side.
If you have any questions about the data, please reach out to your Crownpeak Technical Consultant.
Comments
0 comments
Please sign in to leave a comment.