Einstein Product Recommendations
Einstein Product Recommendations is the Salesforce Commerce Cloud feature that powers product carousel recommendations across the storefront.
Definition
Einstein Product Recommendations is the Salesforce Commerce Cloud feature that powers product carousel recommendations across the storefront. It analyzes shopper behavior (page views, searches, cart activity, purchase history) plus catalog metadata, and selects products to surface in carousels on product detail pages, cart pages, category pages, homepages, and order confirmation pages. Each carousel uses one of several recommender strategies (Customers Also Bought, Recently Viewed, Top Sellers, Complete the Set, Recommended for You), and merchandisers pick the strategy per placement without writing code.
Product Recommendations is the workhorse personalization feature inside the Commerce Cloud Einstein bundle. It pays back fastest of any Einstein commerce feature because the carousels are visible to every shopper on every visit, and the recommendations refresh as shopper behavior changes. The biggest leverage admins have is choosing the right strategy per placement and respecting business rules (out-of-stock filtering, merchandiser-pinned items) that the algorithm should not override.
Why one feature powers half the personalization on most storefronts
The recommender strategies and what each does
Customers Also Bought looks at co-purchase patterns: shoppers who bought this product also bought these others. Recently Viewed shows products the current shopper visited in their session or recent sessions. Top Sellers is a simple ranked list of best-selling products in the current category or site-wide. Complete the Set recommends accessories or complementary items based on the current product. Recommended for You is the most personalized strategy, blending the shopper history, similar shopper behavior, and category context. Each strategy has different data needs and different best-fit placements; matching strategy to placement is the highest-leverage decision in the rollout.
Placement and strategy matching
Product detail pages benefit most from Customers Also Bought and Complete the Set (the shopper is looking at a product; recommend related ones). Cart pages benefit from Complete the Set and Recently Viewed (the shopper has committed to buying; suggest items to add). Category landing pages benefit from Top Sellers and Recommended for You (the shopper is browsing; surface the highest-converting products). Homepages benefit from Recommended for You and Top Sellers (the shopper has not signaled intent yet; show what is likely relevant). Order confirmation benefits from Customers Also Bought (the shopper just bought; suggest next-time additions). Wrong strategy on wrong placement converts at half the rate of the right pairing.
How the recommendation models train
Each strategy runs a different model under the hood. Customers Also Bought uses collaborative filtering over purchase transactions. Recently Viewed is a simple per-shopper session memory. Top Sellers is a windowed ranking over recent sales. Complete the Set uses content-based similarity over product attributes. Recommended for You uses a personalized blend that combines collaborative filtering, content-based, and recent behavior. Training is Salesforce-managed; admins do not pick algorithm parameters. The models retrain on a schedule (hourly for behavior-driven strategies, daily for catalog-driven ones), so the recommendations stay fresh as both the catalog and the shopper base shift.
Business rules that override the algorithm
Five business rules apply across strategies. Out-of-stock items are excluded by default (overridable per placement). Merchandiser-pinned items take priority over algorithmic picks. Restricted products (age-gated, regulated) are filtered. Sale items can be boosted in the ranking on placements where the merchandiser wants to push them. Recently purchased items can be filtered (do not recommend a coat the shopper bought last week). These rules live on the placement, not on the strategy; the same Customers Also Bought strategy can apply different rules on PDP vs cart depending on the merchandising intent.
Acceptance rate and the placement-level metric
Each recommendation surfaces with click-through and add-to-cart tracking. The acceptance rate per placement (clicks divided by impressions, then add-to-cart conversion among clicks) is the diagnostic metric. Good carousels hit 15 to 25 percent click-through and 4 to 8 percent add-to-cart conversion. Below those numbers, the strategy is wrong for the placement, the products being recommended are wrong, or the carousel placement itself is poorly positioned on the page. Three diagnostic levers: change strategy, adjust business rules, or move the carousel.
A/B testing carousels with Einstein Commerce Insights
Einstein Commerce Insights includes A/B test scaffolding for recommendations. Pick a placement, configure two strategies (or two rule sets), and the platform splits traffic and reports conversion lift. Most teams underuse this; running one A/B test per quarter on the highest-traffic placement (usually homepage Recommended for You or PDP Customers Also Bought) consistently produces meaningful uplift. Without testing, decisions about strategy or rules become opinion battles between merchandising and engineering rather than data-driven calls.
Cold-start: what happens when behavior data is sparse
New storefronts and newly added products have a cold-start problem. Without shopper behavior data, behavior-driven strategies (Customers Also Bought, Recommended for You) cannot produce useful recommendations. The platform falls back to content-based similarity (Complete the Set) and global popularity (Top Sellers) until enough behavior data accumulates. New storefronts should plan for 30 days of Top Sellers as the dominant strategy before the personalization features become genuinely personal. New products should be flagged manually for inclusion in Top Sellers and Complete the Set during their first few weeks until they appear in enough behavior data to surface algorithmically.
How to deploy Product Recommendations and measure the lift
The deployment is straightforward; the discipline of matching strategy to placement and measuring conversion is where teams either win or fail. Spend the first week getting the placement decisions right and the next month tuning based on what shoppers actually do.
- Confirm Commerce Cloud Einstein is enabled
Business Manager, Administration, Site Development, Einstein Configuration. Without Einstein enabled, the recommender carousels do not work. Most editions include the bundle.
- Pick the placements for recommendation carousels
Product detail, cart, category landing, homepage, order confirmation. Most storefronts have all five active. Decide before you write the strategy assignment.
- Match strategy to placement deliberately
Customers Also Bought on PDP. Complete the Set on cart. Top Sellers on category landing. Recommended for You on homepage. Customers Also Bought on order confirmation. Document the rationale per placement.
- Add the recommender components to your storefront templates
The standard storefront templates ship with recommender components. Drop one per placement and pick the strategy in the component config. No code changes needed.
- Configure business rules per placement
Exclude out-of-stock, exclude restricted products, exclude recently purchased on PDP and cart. Pin merchandiser items where strategic. Document the rules so the merchandising team knows what the algorithm will and will not surface.
- Measure for 30 days then run one A/B test
Let the carousels run for 30 days to gather baseline data. Then pick the highest-traffic placement and A/B test an alternative strategy or rule set. Repeat quarterly.
- Build a monthly review with merchandising
Pull click-through and add-to-cart conversion per placement monthly. Review with merchandising. Adjust strategies on placements that underperform; the iteration loop is what sustains lift.
Customers Also Bought, Recently Viewed, Top Sellers, Complete the Set, Recommended for You. Per placement; not site-wide.
Out-of-stock filter, pinned items, restricted product filter, recently-purchased filter, sale boost. Per placement.
Which storefront page the carousel renders on (PDP, cart, category, homepage, order confirmation).
Which placement is currently under A/B test, with which two strategies or rule sets. Drives the conversion data the next decision is based on.
How often the underlying model retrains. Hourly for behavior-driven, daily for catalog-driven; tunable in advanced configurations.
- The same strategy on different placements converts differently. Picking one strategy site-wide leaves significant lift on the table.
- New storefronts and new products have a cold-start problem. Plan for Top Sellers as the dominant strategy until shopper behavior data accumulates.
- Business rules sit on the placement, not the strategy. The same Customers Also Bought can apply different rules on PDP vs cart; configure deliberately.
- Without A/B testing, strategy decisions become opinion battles. Run one A/B test per quarter on the highest-traffic placement to keep decisions data-driven.
- Recommender carousels without monitoring quietly underperform. The monthly merchandising review is the discipline that sustains the lift.
Trust & references
Cross-checked against the following references.
- Salesforce Commerce CloudSalesforce
- Product Recommendations referenceSalesforce
Straight from the source - Salesforce's reference material on Einstein Product Recommendations.
- Einstein Product RecommendationsSalesforce Help
- Einstein for Commerce CloudSalesforce Help
- Einstein Commerce InsightsSalesforce Help
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. What does Einstein Product Recommendations do?
Q2. Why are product recommendations high-leverage in e-commerce?
Q3. Where can recommendations appear?
Discussion
Loading discussion…