Magento
Adobe Commerce Product Recommendations on production catalogues
Adobe Commerce Product Recommendations is the SaaS layer I use when a Magento catalogue is too large for hand-picked related products but still needs intelligent upsells on PDP, category, and home. I am Alan Vo, a Gold Coast web developer with 18 years on Adobe Commerce, Shopify, and WooCommerce storefronts. Adobe's Product Recommendations overview (updated September 2026) describes units powered by aggregated shopper behaviour and catalog data, not only static related SKUs. This note is for teams who already run cron and indexers but have never wired the Commerce Services Connector for PREX.
Product Recommendations requires Adobe Commerce entitlements and the magento/product-recommendations metapackage. Magento Open Source does not ship the Admin workspace, SaaS sync, or storefront SDK. If your licence is Open Source, treat this as an upgrade planning doc, not a Composer line you can add quietly.
Why Adobe Commerce Product Recommendations matter in 2026
Adobe Commerce Product Recommendations matter in 2026 because manual related-product rules do not scale when you carry tens of thousands of SKUs, seasonal drops, and multiple store views. Adobe documents nine intelligent recommendation types spanning shopper-based, item-based, popularity, trending, and similarity signals. You still keep Magento product schema and merchandising rules, but PREX fills the long tail where no editor will ever curate every PDP.
Adobe raised operational limits that change how merchandisers plan slots. Release notes for November 2025 state you can create up to 50 active recommendation units per page type, up from five. That sounds generous until marketing wants a unique unit per category landing page. You still need naming discipline and a kill list for underperforming units.
The 6.8.0 metapackage (September 2026) adds optional Fastly Image Optimization for recommendation image URLs via magento/module-fastly-recommendations. On Adobe Commerce on Cloud, that matters when recommendation carousels suddenly pull full-size cache misses and hurt LCP on home and category templates.
Adobe also documents hard constraints: Product Recommendations is not HIPAA-ready. Do not enable PREX on implementations that process protected health information under Adobe's HIPAA-ready offering. Privacy and cookie restriction mode got tighter in 2025 releases: PREX now respects cookie restriction mode by not collecting behavioural data when restrictions are enabled, which is correct for EU and Australian consent postures but can make units look empty until shoppers accept cookies.
PREX is complementary to native Product Relationships and related rules. Adobe's overview recommends using AI recommendations at scale and manual rules when you must pin a specific SKU for a segment. I use both on spirits and confectionery catalogues where brand partnerships override algorithmic suggestions.
How Adobe Commerce Product Recommendations actually work
Adobe Commerce Product Recommendations actually work as a three-part loop: catalog export from your instance, behavioural events from the storefront SDK, and recommendation units you create in Admin that the SDK renders into placeholders.
Install and connector. Adobe's install and configure guide has you run composer require magento/product-recommendations, then configure the Commerce Services Connector with API keys and a SaaS Data Space. The SaaS Data Export extension handles synchronization. After linking, catalog sync begins; you verify behavioural data on the storefront before declaring victory.
Indexers and cron. Adobe requires cron jobs and indexers to run, with the Product Feed indexer set to Update by Schedule. If Product Feed sits in Update on Save on a huge catalogue, you will stall exports. I treat Product Feed like Live Search feeds: monitor the Data Feed Sync Status dashboard Adobe added in 6.6.0 (January 2026) for real-time export health.
Admin units. In Admin you create a recommendation unit under Marketing > Promotions > Product Recommendations (exact menu labels match your Adobe Commerce version). Adobe's create recommendation workflow asks for an internal name, page type (Home, Category, Product Detail, Cart, Confirmation, or Page Builder), recommendation type, storefront label, product count (default 5, max 20), and placement on the page.
Page type limits matter: Adobe notes recommendations are not supported on the Cart page when the store sends shoppers straight to the cart after add-to-cart. That configuration is common on B2B sites; do not promise cart upsells without checking checkout behaviour first.
Page Builder path. Optional module magento/module-page-builder-product-recommendations lets you embed units in CMS content. Adobe requires Page Builder recommendation units to be created in the default store view, then deployed elsewhere. Non-default store view metrics for Page Builder units improved in 2025 releases, but creation scope rules still trip multi-store teams.
Storefront rendering. The recommendations SDK loads from Adobe CDN URLs on the adobe.io domain (updated in 6.4.0). Units render labelled carousels such as "Customers who viewed this product also viewed." Hyva and Luma both need theme integration paths that expose the SDK containers without duplicating jQuery widgets from legacy Luma overrides.
Catalog Service overlap. If you already run Adobe Commerce Catalog Service for PDP reads, PREX still maintains its own SaaS relationship to behavioural data. Coordinate feed timing so price and stock shown in recommendations match Catalog Service or core catalog truth.
Production checklist for Adobe Commerce Product Recommendations
A catalogue is ready for PREX when Commerce Services credentials work in every environment, Product Feed is healthy, and theme placeholders exist before marketing creates fifty units nobody can place.
- Confirm Adobe Commerce licence includes Product Recommendations entitlements before Composer work. Sales-led projects sometimes assume SaaS bundles that are not on the contract.
- Install
magento/product-recommendationson integration, runsetup:upgrade, compile, and deploy static content per your cloud pipeline. Pin versions using Adobe's product availability matrix for your Commerce core version (6.8.0 on latest 2.4.x as of late 2026).
- Configure Commerce Services Connector API keys and SaaS Data Space in Admin. Match data space between staging and production carefully; 6.5.0 improved cross-environment unit behaviour but you still avoid pointing staging traffic at production SaaS data.
- Set Product Feed indexer to Update by Schedule and verify cron groups
index,default, and SaaS export jobs are running on cloud. Watch the Data Feed Sync Status dashboard after first full export.
- Smoke-test behavioural collection on a storefront SKU path: view, add to cart, purchase in a test website. Empty units on day one often mean events never arrived, not that AI lacks data.
- Create one unit per page type first (home popularity, category trending, PDP viewed-this-viewed-that). Activate, wait for impressions in the Product Recommendations metrics table, then expand.
- Document cookie restriction mode impact for legal: when restrictions are on, behavioural personalization pauses. Plan copy for logged-out EU or AU shoppers.
- If you use Hyva, audit Alpine and Tailwind compatibility with the SDK mount points. Pair with Hyva theme and Magento Core Web Vitals checks so carousels do not destroy INP.
- Enable Fastly Image Optimization module only when Fastly is your edge and images are the bottleneck. Measure LCP before and after on category pages with 20-item units.
- For Page Builder campaigns, create units in default store view, then place blocks per Adobe's Page Builder guidance. Preview multi-store labels before campaign launch alongside Adobe Commerce content staging preview if prices change on schedule.
- Align with Adobe Commerce Live Search merchandising so search badges and PREX carousels do not contradict each other on the same PLP.
- Add PREX units to your release checklist: SDK version bumps ship outside core Commerce quarterly releases, so track
magento/product-recommendationsrelease notes independently.
What breaks when Adobe Commerce Product Recommendations are misconfigured
The first break is empty recommendation units with no Admin error. Usually Product Feed is stale, cron is disabled on a cloud environment, or Commerce Services keys belong to another data space. Fix exports before you tune recommendation types.
The second break is JavaScript errors removing entire units. Adobe fixed an intermittent issue in 6.4.0 where missing ds-view-history-time-decay local storage keys crashed PREX. Clear site data tests still belong in QA when privacy extensions wipe storage aggressively.
The third break is cart page expectations. Stores configured to jump to cart after add-to-cart cannot use cart page types Adobe documents as unsupported. Merchandising will still ask for "cart cross-sell AI"; you need a PDP or mini-cart strategy instead.
The fourth break is consent and empty carousels. Cookie restriction mode correctly stops tracking, but stakeholders interpret blank units as broken deploys. Communicate that personalisation requires consent where restrictions apply.
The fifth break is Page Builder default store view creation. Editors build a beautiful CMS row in a French store view, then cannot select the unit because it was never created at default scope. Train the workflow once in writing.
The sixth break is performance on image-heavy units. Twenty products with unoptimized images on home hurts LCP. Fastly Image Optimization module helps only when installed and when URLs pass through Fastly IO parameters Adobe documents for 6.8.0.
The seventh break is HIPAA and regulated data. Enabling PREX on a HIPAA-ready stack violates Adobe's service classification. Compliance reviews should catch this, but agency handoffs sometimes miss it.
The eighth break is environment clutter. Before 6.5.0, recommendation units interacted oddly across environments. Keep staging units named with an environment prefix and deactivate experiments you are not measuring.
How to measure Adobe Commerce Product Recommendations
You measure Adobe Commerce Product Recommendations with the metrics Adobe surfaces per unit in Admin, plus storefront performance and revenue sanity checks, not with gut feel from a single carousel.
Adobe documents impressions, views, clicks, and related metrics on the Product Recommendations grid once units are active. I export screenshots weekly during the first month, then monthly, and compare click-through rate and add-to-cart rate from recommendation clicks (via your analytics implementation) for top units versus bottom quartile units. Kill or retype units that never earn clicks after sufficient impressions.
Define minimum impression thresholds before decisions. New sites need weeks of traffic; launching PREX Monday and judging Tuesday wastes merchandiser time. Adobe's machine learning uses aggregated behaviour; cold-start periods are normal.
In GA4 or your analytics stack, tag recommendation clicks with a consistent event parameter for unit name and page type. Align with GA4 ecommerce events so purchase paths attributed to recommendation clicks are visible in explorations. I do not invent lift percentages for client work; I report directional trends and let A/B tests run where traffic allows.
Watch Core Web Vitals on templates that gained units. Use field data in Search Console or CrUX-style monitoring. If INP regresses, defer non-critical SDK loads or reduce product count from 20 to eight on mobile-first templates.
Compare search versus recommend revenue paths when Live Search is live. Sometimes search wins for intent-heavy shoppers while PREX wins for discovery. Both can coexist if you measure separately.
For catalog health, monitor Product Feed sync duration and error rows in the Data Feed Sync Status dashboard. Rising export times predict stale recommendations before shoppers notice wrong prices.
Related Magento work on this site
Product Recommendations sit on the same Adobe Commerce stacks I write about for large catalogues and seasonal drops. The retail conversion programme (+$2.5m YoY sales on a multi-store Adobe stack) needed merchandising automation that did not fight checkout or inventory truth. The KitKat Chocolatory Magento experience is the kind of campaign-heavy storefront where PREX trending types help gift seasons without hand-curating every SKU. The Spirited Gifts Magento shop mixes brand rules with discovery, which is where Adobe's guidance to combine Product Relationships with AI recommendations matches real merchandising politics.
For SaaS neighbours, read Adobe Commerce Live Search and Adobe Commerce Catalog Service so feeds, search, and recommendations share one operational calendar. For front-end performance after you embed SDK carousels, keep Hyva theme and Magento Core Web Vitals in the same release train.
FAQ
What are Adobe Commerce Product Recommendations?
Adobe Commerce Product Recommendations are SaaS-powered recommendation units that use catalog and shopper behaviour data to show personalized product carousels on home, category, product, cart, confirmation, and Page Builder pages. You install the magento/product-recommendations module, connect Commerce Services, sync the Product Feed, then create and activate units in Admin.
How do I install Adobe Commerce Product Recommendations on a live store?
Run composer require magento/product-recommendations, upgrade the database, configure Commerce Services API keys and SaaS Data Space, ensure cron and indexers run with Product Feed on Update by Schedule, then verify behavioural events on the storefront before activating units. Adobe's install and configure documentation is the canonical sequence.
Can Adobe Commerce Product Recommendations run on Magento Open Source?
No. Product Recommendations is an Adobe Commerce entitlement with SaaS connectors, Admin workspaces, and storefront SDK integration. Open Source projects need a different personalization approach or a licence upgrade.
Why are my Adobe Commerce Product Recommendations units empty?
Empty units usually mean catalog export or behavioural collection is not working: Product Feed indexer stale, cron disabled, wrong SaaS Data Space, cookie restriction mode blocking tracking, or insufficient traffic for cold-start models. Confirm exports in the Data Feed Sync Status dashboard and test events on a product page before changing recommendation types.
Keep reading
Contact if you want this kind of work on a live store.