Recommendations+

Recommendations+ is an upgrade to the Personalized recommendations for you template. It helps improve click-through rates and user engagement. With this feature, you attract more visitors, get them to explore more products, and increase actions like adding items to the cart.

For a deeper understanding of our recommendations, explore the Recommendations section. You'll find detailed guidance to utilize and optimize this feature to enhance your user engagement.

Understanding the models

Sequential Two-tower neural network model

  • Uses two separate networks for users and items. It combines their outputs to make personalized recommendations based on user interactions and product data.
  • It's like having two sets of eyes, one focused on users and the other on items, working together to suggest what users might like.
  • Recommendations+ runs on this model and doesn't require specifying events. It uses View items directly from your project's data mapping.

Matrix-factorization model

  • Breaks down user and item data into smaller parts to predict user preferences.
  • Similar to taking a puzzle apart and using the pieces to figure out what's missing.
  • The standard Personalized recommendations for you template runs on this model.

Benefits of Recommendations plus

  • Personalized experiences: Sequential Two-Tower Neural Network model delivers personalized recommendations by analyzing each user's behavior in real-time.
  • Sequential recommender: This feature predicts the next best product by considering the order of user interactions, offering more accurate recommendations.
  • Higher click-through rates (CTR): Advanced AI techniques boost product CTR by aligning recommendations with the user's current journey.
  • Real-time updates: The system adapts instantly to user behavior, ensuring relevant recommendations for all visitors.
  • Scalable with data: The Sequential Two-Tower Neural Network model improves with more data, refining recommendations to better match user preferences.

Setup guide

  1. Navigate to Campaigns > Recommendations and click + New Recommendation.
  2. Select Personalized recommendations for you from the template list.
  1. In the Model type step, choose the Two-tower neural network, which is preselected by default.
    1. If you select the Matrix-factorization option, refer to the standard Personalized recommendations for you guide for detailed setup instructions.
  2. Fill in View items in data mapping to avoid the warning below.

Use cases

  • Reducing site abandonment

You can place personalized recommendations inside weblayers and trigger them before a visitor leaves the site. This will help reduce abandonment rates and increase the time users spend on the site.

  • Increasing product exploration

Convert homepage traffic into product detailed page (PDP) views by using personalized recommendations. This encourages users to explore more products and increases engagement.

  • Maximizing interaction on product detail pages

Provide multiple recommendation types on PDPs, such as "personalized" and "similar items," to maximize the chance of user interaction and increase the likelihood of conversions.

  • Driving traffic through email newsletters

Include personalized recommendations in regular email newsletters to drive traffic back to the website, increasing the potential for conversions.

  • Enhancing cart and site abandonment communications

Include personalized recommendations in cart or site abandonment emails and mobile messages. This helps re-engage users and encourages them to complete their purchases.

  • Improving product discovery

Deliver personalized recommendations after a product discovery quiz conducted over mobile messaging. This enhances the user experience and guides users toward products they are likely to be interested in.

Accessing Recommendations plus

Recommendations+ feature is available through two premium packages that must be enabled in account administration:

  • Loomi Web Personalization
    • Engage users directly on your website and through browser notifications.
    • Channels: Weblayers, Experiments, Browser push
  • Loomi Journey Personalization
    • Reach your customers through personalized messages and notifications.
    • Channels: Email, Mobile push, SMS, RCS, WhatsApp, App Inbox, In-app Personalization

Limitations

  • Real-time inference may have lower throughput than the Matrix-factorization model, especially for large campaigns.
  • Recommendations+ is not available in the advanced engine.
  • The engine can process up to 1 million active products. Exceeding this may cause training failures or slow response times.
  • Heavy dynamic filters may return fallback items.