Skip to main content

Product Recommendations

Product recommendations are a fundamental part of almost every successful e-commerce experience. From "Top Picks for You" on the homepage, to "Customers Also Bought" on product detail pages (PDPs), or "Related Items" in the cart, these modules aim to help shoppers discover relevant products they might not have found otherwise. Effective recommendations drive significant business value by increasing product discovery, boosting click-through rates (CTR), improving conversion rates, and lifting average order value (AOV). However, building recommendation systems that go beyond simple "most popular" lists and deliver truly personalized, high-performing suggestions requires tackling substantial technical challenges.

Building Product Recommendations with Shaped​

Let's illustrate implementing homepage recommendations using Shaped.

Goal: Display a "Top Picks For You" carousel on the homepage for logged-in users.

1. Ensure Data is Connected: Assume you have connected:

  • user_interactions: Contains user_id, item_id (product ID), timestamp, event_type (view, add_to_cart, purchase).
  • product_catalog: Contains item_id, title, category, image_url, product_url, price.

2. Define Your Model (YAML): Focus the model on learning preferences from user interactions.

product_recommendation_model.yaml
model:
  name: homepage_product_recs
connectors:
  - type: Dataset
    name: user_interactions
    id: interactions
  - type: Dataset
    name: product_catalog
    id: products
fetch:
  events: |
    SELECT
      user_id,
      item_id,
      timestamp AS created_at,
      -- Assign weights/labels based on interaction type
      CASE
        WHEN event_type = 'purchase' THEN 1.0
        WHEN event_type = 'add_to_cart' THEN 0.7
        WHEN event_type = 'view' THEN 0.1
        ELSE 0.0
      END as label
    FROM interactions
  items: |
    SELECT
      item_id,
      title,
      category,
      brand,
      image_url,
      product_url,
      price
    FROM products

3. Create the Model:

shaped create-model --file product_recommendation_model.yaml

4. Monitor Training: Wait for the model to reach the ACTIVE state.

shaped view-model --model-name homepage_product_recs

5. Fetch Homepage Recommendations: Once ACTIVE, your website backend (when rendering the homepage for a logged-in user) calls Shaped's rank API.

const { Shaped } = require('@shaped/shaped');  

const shapedClient = new Shaped(); // Assumes SHAPED_API_KEY env var  
const response = await shapedClient.rank({  
    modelName: 'homepage_product_recs',  
    userId: 'USER_456',  
    limit: 10,  
    returnMetadata: true  
});  
console.log(`Rendering ${response.metadata.length} recs for USER_456`);

Your frontend receives the list of recommended products (with metadata) and renders the "Top Picks For You" carousel.

(Note: For PDP "Customers Also Bought" or "Related Items," you would typically use the similar_items endpoint, passing the user_id and the item_id of the product currently being viewed.)

Conclusion: Focus on Merchandising Strategy, Not ML Plumbing​

Effective product recommendations are crucial for e-commerce success, but building them from scratch requires overcoming substantial hurdles in data engineering, machine learning, infrastructure, and ongoing optimization. Shaped provides the specialized AI platform to handle this complexity.

By connecting your Shopify, database, warehouse, or event stream data to Shaped, you can leverage state-of-the-art models and scalable APIs (rank, similar_items) to power high-performing product recommendations across your site with significantly less internal effort. Let Shaped manage the complex AI engine so your team can focus on merchandising strategy, user experience, and driving growth.

Ready to implement product recommendations that truly convert?

Request a demo of Shaped today to see it in action with your specific use case. Or, start exploring immediately with our free trial sandbox.