WooCommerce exposes a large amount of product data, but a personalization system does not benefit from receiving every available field. For agencies building product recommendations with Amazon Personalize, the important task is to identify fields that describe a product consistently, help models distinguish products, and remain useful when a new item has little or no interaction history.
Start with a stable product identifier
The most important field is the product identifier. In WooCommerce, this is usually the product id, although some stores use the SKU as the external identifier.
Whichever identifier you select must be stable across catalog synchronizations and must match the identifier sent in interaction events. If an item is imported into Amazon Personalize as 12345, an add-to-cart or purchase event must use that same value as its itemId.
{
"itemId": "12345",
"eventType": "purchase",
"sessionId": "session-abc",
"timestamp": 1710000000
}
Do not casually switch between a WooCommerce post ID, SKU, variation ID, and a custom ERP code. If the store changes its identifier strategy, the catalog and interaction history must be migrated together or the model will treat the same product as different items.
Fields that usually matter most
For a typical WooCommerce catalog, the following fields are strong candidates for item metadata:
- Product type: Simple, variable, grouped, or another product type can help distinguish the structure of the item.
- Categories: Category assignments are often among the most useful categorical features, particularly for new or infrequently purchased products.
- Tags: Tags can capture merchandising concepts that are not represented by the category hierarchy, but only when the store applies them consistently.
- Brand: Brand is valuable for stores where customers show strong brand preferences. It may come from a global attribute, taxonomy, or custom field.
- Attributes: Color, size, material, style, compatibility, and similar attributes can describe product differences that categories cannot capture.
- Price band: A normalized price range can be useful when customers have clear budget preferences. Use a controlled band rather than a constantly changing raw price when possible.
- Availability: Stock status and purchasability are useful for filtering recommendations, although they are often better handled by the serving application than by the model itself.
- Text fields: Product names, short descriptions, and carefully cleaned descriptions can provide semantic information, subject to the selected Amazon Personalize schema and recipe.
WooCommerce fields and their role
A WooCommerce REST API product commonly contains fields such as name, sku, price, regular_price, sale_price, stock_status, categories, tags, images, attributes, variations, type, status, and catalog_visibility. These fields do not all have the same modeling value.
name is usually a reasonable text field. Categories, brands, and controlled attributes are usually more reliable as categorical metadata. permalink, image URLs, menu order, and display settings are generally presentation data rather than useful behavioral signals.
date_created and date_modified can help an agency monitor synchronization and identify newly launched products, but they should not automatically be treated as customer preference features. A recently edited product is not necessarily a new or popular product.
Parent products versus variations
Variable products require an explicit modeling decision. A store can recommend the parent product, recommend each variation, or use both levels for different purposes.
Recommend the parent product when customers primarily choose a product family and select size or color after opening the product page. This avoids splitting interaction history across many variation IDs. Recommend variations when the differences affect buying intent, inventory, price, or compatibility. For example, a customer choosing a specific laptop configuration may have a meaningfully different preference from a customer viewing another configuration.
If variations are included as items, propagate useful parent metadata to each variation. A variation record should normally retain the parent category, brand, and product family, while adding variation-specific attributes such as size, color, capacity, or material.
Do not use volatile fields carelessly
Price, sale status, stock quantity, and visibility can change frequently. Including them in a catalog feed is not inherently wrong, but stale values can make recommendations misleading.
A practical architecture separates model metadata from real-time eligibility:
- Use relatively stable fields such as category, brand, and product attributes as item metadata.
- Use the WooCommerce catalog or inventory service at request time to remove products that are out of stock, unpublished, restricted, or unavailable in the shopper’s region.
- Refresh price-related metadata when it is genuinely relevant to the recommendation strategy.
For example, a recommendation response may contain a product that was valid when the model produced it. The application should still verify stock_status, purchase restrictions, and current pricing before rendering or adding it to a cart.
Normalize fields before exporting them
Raw WooCommerce values often need transformation before they become item metadata. Category names should use a consistent identifier or taxonomy path. Brand values should not appear as separate values because of capitalization or punctuation differences. Attributes should be mapped to a predictable representation.
{
"itemId": "12345",
"category": "outdoor > hiking > footwear",
"brand": "example-brand",
"product_type": "variable",
"color": "black",
"material": "leather",
"price_band": "100-199",
"title": "Waterproof hiking boot"
}
Use controlled categorical values where possible. A field containing thousands of nearly unique strings, such as an unprocessed description or an internal export label, may add noise without improving recommendations.
Fields that should usually be excluded
Many WooCommerce fields are important to the storefront but provide little value as product metadata. Examples include edit links, administrative author information, image URLs, HTML fragments, internal notes, shipping configuration, tax settings, and layout-specific menu positions.
Exclude customer-specific or sensitive information from item metadata. A product feed should describe the product, not contain a customer’s name, address, order notes, or account data. Customer behavior belongs in the interactions and users datasets according to the applicable Amazon Personalize schema and privacy requirements.
Use interaction events to capture behavior
Product fields describe what an item is. They do not describe how shoppers respond to it. Views, clicks, add-to-cart events, purchases, wishlist actions, and dismissals should be sent as interaction events with consistent item identifiers.
For example, a product’s category should come from the item catalog, while the fact that a shopper purchased that product belongs in the interaction dataset. Avoid encoding every behavior as a product attribute; doing so loses the relationship between a user, an item, and an event.
A practical field-selection checklist
- Does the field remain consistent across catalog exports?
- Does it describe a meaningful product distinction?
- Is the value normalized across products and languages?
- Will the field be available for newly launched products?
- Can the value be refreshed before it becomes misleading?
- Does the selected Amazon Personalize schema support the intended data type?
- Can the storefront still enforce real-time availability and merchandising rules?
A strong initial catalog for many WooCommerce stores contains a stable item ID, category, brand, product type, selected attributes, a cleaned title, and an optional price band. Agencies can then measure recommendation quality and add fields based on evidence rather than exporting the entire WooCommerce product object by default.