Cart recommendations should help shoppers complete an order, not distract them from checkout. For a WooCommerce store, the most useful suggestions are usually complementary products: accessories for a primary product, consumables that are often purchased with it, or a small add-on that increases order value without changing the shopper’s original intent.
Choose the recommendation strategy
Amazon Personalize supports several ways to produce recommendations. The right approach depends on whether the cart contains one product or several products and whether the store needs a simple related-items list or a user-specific ranking.
- Related items: Generate products related to a selected cart item. This is a practical starting point when the shopper is anonymous or when the cart contains one dominant product.
- User personalization: Use the shopper’s known user ID to request recommendations based on their interaction and purchase history.
- Personalized ranking: Create a candidate list of products and ask Amazon Personalize to rank that list for a specific user. This works well when the application already has a set of eligible cart add-ons.
For a multi-item cart, a common architecture is to use the cart contents to build a candidate set, remove products already in the cart, and then rank the remaining products. This avoids treating one cart item as the complete representation of the shopper’s current intent.
Design the cart recommendation flow
- The WooCommerce frontend requests recommendations from a WordPress endpoint.
- The endpoint reads the current cart and authenticated customer, if available.
- The server creates or retrieves a candidate product list.
- The server calls Amazon Personalize using a stable user ID, item ID, or both.
- The response is filtered against WooCommerce catalog and inventory rules.
- The frontend renders a small recommendation block without replacing the cart or checkout form.
Keep the Amazon Personalize call on the server. AWS credentials, campaign or recommender identifiers, and catalog rules should never be exposed to browser JavaScript.
Use WooCommerce as the source of truth
Amazon Personalize should recommend product IDs, but WooCommerce should decide whether those products can be displayed or purchased. Before returning a recommendation, verify that the product exists, is published, is purchasable, is in stock, and is appropriate for the current customer or market.
A WordPress endpoint can return only the fields required by the cart component:
[
{
"id": 481,
"name": "USB-C Charging Cable",
"url": "https://shop.example.com/product/usb-c-charging-cable/",
"price_html": "<span class=\"woocommerce-Price-amount\">£12.00</span>",
"image_url": "https://shop.example.com/uploads/charging-cable.jpg"
}
]Do not trust a product ID returned by Amazon Personalize without checking it against the current WooCommerce catalog. Products may have been deleted, unpublished, discontinued, or made unavailable in a particular region since the interaction data was imported.
Handling a cart with several products
There are two practical patterns for multi-item carts.
Aggregate related-item results
Request related items for each cart product, merge the returned IDs, remove duplicates, and score or limit the result. This approach is straightforward and can work well when the store has a strong product association model.
For example, a cart containing a camera, memory card, and tripod might produce candidate products such as spare batteries, a camera bag, and a cleaning kit. The integration should remove the memory card and tripod from the candidate list if they are already in the cart.
Rank a controlled candidate set
Build candidates from product relationships, a category rule, or a merchandising table, then send those candidates to a Personalized Ranking campaign. This gives the agency more control over eligibility while allowing Amazon Personalize to order the candidates for the shopper.
const input = {
campaignArn: process.env.PERSONALIZED_RANKING_CAMPAIGN_ARN,
userId: customerId || anonymousUserId,
inputList: candidateProductIds.map(String),
context: {
device: deviceType,
channel: "cart"
}
};
const result = await personalizeRuntime.send(
new GetPersonalizedRankingCommand(input)
);The exact AWS SDK operation and parameters depend on the Amazon Personalize resource being used. Keep the campaign or recommender configuration outside the WooCommerce theme and load it from environment-specific settings.
Anonymous shoppers and identity handling
Cart recommendations often appear before a shopper logs in. Generate a first-party anonymous identifier and retain it for the current browsing period. When the shopper authenticates, associate future events with the known customer ID according to the store’s identity policy.
Use the same identifier consistently for event ingestion and recommendation requests. Avoid using an email address directly as the user ID. A normalized internal customer key is easier to protect, migrate, and reconcile across systems.
For a new anonymous visitor with little or no history, use a fallback such as popular products, category-specific best sellers, or manually selected complements. Do not make an empty or low-confidence Amazon Personalize response look like a broken cart component.
Record the events that matter
Recommendations improve only when the interaction data reflects the cart experience. At minimum, consider recording:
- View: The shopper views a product or recommendation card.
- Click: The shopper opens a recommended product.
- Add to cart: The shopper adds the recommended product.
- Purchase: The order is paid or reaches the business’s selected conversion state.
- Impression: The recommendation was actually displayed, if the reporting model needs exposure data.
Include a placement or event value that identifies the cart placement, such as cart_recommendations. This allows the agency to distinguish cart behavior from product-page, search, email, and homepage interactions.
Send the item IDs in the format expected by the Amazon Personalize schema. WooCommerce product IDs and variation IDs should not be mixed accidentally. If customers purchase variations but recommendations operate at the parent-product level, define that mapping explicitly and apply it consistently during both event ingestion and response filtering.
Prevent performance and caching problems
A cart page is sensitive to latency. Do not block the entire cart render while waiting for Amazon Personalize. Load the recommendation area asynchronously and show nothing, or a small reserved placeholder, if the request exceeds the agreed timeout.
Cache carefully. A response based on a customer ID should not be stored in a public page cache. Anonymous recommendations can be cached briefly, but the cache key should include the relevant cart product IDs, currency, language, customer segment, and store context when those values affect the result.
Use short-lived object caching for repeated requests during cart refreshes. WooCommerce can update the cart through AJAX after quantity changes, shipping changes, coupons, or mini-cart actions, so the integration should invalidate or refresh recommendations when the cart contents change.
Apply business rules after personalization
Personalization ranks products; it does not replace commercial rules. A WooCommerce integration should apply rules for stock status, product visibility, restricted categories, minimum age, shipping destination, subscription compatibility, and promotional exclusions.
Limit the visible set to a manageable number, commonly three to six products. If the response contains fewer eligible products after filtering, display the remaining products rather than filling the block with unrelated items. Also avoid recommending a product that the shopper has just removed from the cart unless the placement is deliberately designed to recover that action.
Measure incremental value
Track the recommendation placement separately from the cart’s overall conversion rate. Useful measures include recommendation impressions, clicks, add-to-cart rate, attach rate, revenue from attached items, average order value, and cart abandonment.
Use an experiment or holdout group before declaring success. Compare shoppers who receive the personalized placement with shoppers who receive a suitable baseline, such as best sellers or no recommendation block. Measure incremental revenue and margin, not only click-through rate. A highly clicked product that replaces a more profitable item may not improve the business outcome.
For agency reporting, preserve the recommendation request ID, placement name, returned item IDs, and eventual order ID where permitted. This makes it possible to connect an impression or click to a later purchase without treating every cart conversion as recommendation revenue.