An important update has been made to the "Looking Similar" feature in Recommend. Previously, this feature could only fetch a maximum of 500k items, which prevented the model from being trained if the catalog exceeded this limit. This posed a challenge for customers with larger indices. With this update, Algolia now accepts indices of all sizes, still fetching up to the maximum number of items. Records are browsed and ordered by Custom Ranking to prioritize the most relevant items based on your business criteria. For example, training on a catalog of 3 million book items now completes in 5 hours using the top 500k records. Next, the limit will be tested at 1 million items, enhancing flexibility and usability for larger datasets.

An option to customize the events used by the Frequently Bought Together (FBT) model has been released. Previously, customers reported that the FBT model did not recommend items that they expected to go together, partly because the model considered all conversion events, such as purchases, add-to-cart, and add-to-wishlist, which diluted the accuracy of recommendations. To address this, a new setting has been introduced in the FBT model configuration that allows users to specify which event types should be considered for training. This customization increases the relevancy of FBT recommendations and improves adoption rates. Customers can update their existing FBT configurations or set it up during the initial model setup to ensure the model is trained with the most accurate conversion data. This enhancement is part of a broader initiative to detect sufficient event data availability before training, reducing the likelihood of training failures.

Enhanced explainability for NeuralSearch results has been implemented, enabling customers to better understand why search results are presented and ordered as they are. Previously, the combination of multiple factors such as keyword matching, semantic understanding, and ranking factors made it challenging to discern the reasoning behind search result rankings. This update addresses that issue by introducing a new "Why were these results returned?" section on the search results page, providing a clear breakdown of the composition of the search results feed. Additionally, visual aids have been added to help understand the relative ranking of results. Enhanced explainability not only boosts confidence in search results but also provides actionable insights, ensuring a better user experience. 

Frontend support for content search has been added to Salesforce B2C (SFCC) cartridge, version 24.2.0. Customers have expressed the need for a unified search experience that includes both product data and content such as FAQs. SFCC content assets and Page Designer content indexing is now available with added frontend support for Autocomplete and InstantSearch. This enhancement allows customers to achieve a truly unified search experience, improving discoverability and user satisfaction by seamlessly integrating product and content search capabilities.

Learn more: Indexing | SFRA Frontend Support

Version 24.2.0 of the Salesforce B2C (SFCC) cartridge has been released, featuring enhanced variant handling through a new record-per-base product option and out-of-the-box frontend support. This update addresses the need for e-commerce customers to merchandise and optimize relevance at the base product level rather than on a per-variant level. Merchandisers can now manage relevance for products with different colors and sizes as a single entity, simplifying the process and aligning with their view that variants do not significantly influence relevance. This flexibility allows customers to display products in a combined result tile with variant-specific swatches and images, enhancing the user experience with minimal coding effort. The new option provides customers the ability to choose between managing products at the base level or variant level, depending on their preferences. 

Learn more: Record Model Considerations | SFRA Front End Support of Color Swatches 

The new “Recommended for You” model is now in private beta. It enables the creation of personalized product carousels by leveraging individual affinity profiles. This feature delivers more relevant content to your users, boosting user engagement and conversion. 

Merchandising Studio performance has been significantly improved, addressing previous slowness issues. Enhancements in caching, API efficiency, and request management have reduced loading times by 60%, particularly on the Visualize page. These optimizations enhance user experience, ensuring smoother and faster interactions. Further enhancements are in progress to ensure ongoing performance gains.

Revenue Analytics is now out of beta, providing new revenue metrics for improved business decision-making. This highly requested feature closes a competitive gap and enhances the ability to demonstrate the value Algolia brings to your bottom line. Available to all customers on Build, Premium, Elevate, and Elevate Ecomm plans. Click here to learn more.

The new Comparison Mode in Merchandising Studio is designed to enhance transparency and control over ranking factors such as NS, QCat, DRR, and Rules. This unified view allows merchandisers to easily simulate changes and understand the impact of each ranking layer, improving decision-making and identifying potential misconfigurations. By providing clear insights into the AI and ranking system, this feature fosters trust and enables more effective management of result sets. Future updates will continue to refine this tool based on user feedback.

”Looking Similar,” a new visual recommendations model, has been launched in Recommend. This innovative feature analyzes images to find visually similar items, addressing the cold start problem by generating recommendations without needing events. Perfect for eCommerce companies aiming to create visually engaging shopping experiences. Available to all Recommend users starting today, this model marks our first step into image-based recommendations and search.