Product Recommendations for E-commerce: Algorithms, Placement, and ROI

Product Recommendations for E-commerce: Algorithms, Placement, and ROI

Product recommendations are everywhere in ecommerce: "You might also like," "Frequently bought together," "Customers who viewed this also viewed." They have become so ubiquitous that shoppers barely notice them. Yet these widgets drive an outsized share of revenue relative to the page real estate they occupy.

Shoppers who engage with recommendations convert at two to five times the rate of those who do not. The question is not whether to use product recommendations: it is how to do them well.

How Recommendation Algorithms Work

Rule-Based Recommendations

Manually defined rules: "On moisturizer pages, show other moisturizers from the same brand." "In cart, show accessories if the cart contains shoes." Rule-based recommendations require no machine learning and work well for catalogs under 500 SKUs. Beauty brands with defined skincare routines can manually curate "complete your routine" recommendations that outperform algorithms because brand expertise adds value.

The limitation is scalability. As your catalog grows, maintaining rules becomes unwieldy.

Collaborative Filtering

The algorithm behind "Customers who bought this also bought." It finds patterns in aggregate purchase and browsing behavior to predict what individual shoppers want. Item-based collaborative filtering, finding products frequently purchased together, is more common in ecommerce because it handles new users better.

The strength is surfacing non-obvious relationships. You might never think to pair a specific hair treatment with a particular shampoo, but purchase data reveals the connection.

Content-Based Filtering

Recommends products similar to what the shopper viewed based on attributes: category, brand, price, color, material. If a shopper browses three blue dresses under $100, it recommends more blue dresses in that range. Works well for attribute-rich catalogs like fashion and solves the cold-start problem for new products with no purchase history.

Hybrid Approaches

Most production systems combine methods: collaborative filtering as the primary signal, content-based for new products, and rule-based overrides for merchandising priorities.

Placement: Where Recommendations Appear

The algorithm determines what to recommend. Placement determines whether anyone sees it. Optimizing placement is often more impactful than optimizing the algorithm.

Product Detail Pages

The highest-impact placement. Key widgets include:

• "Similar products": Alternatives for shoppers unconvinced by the current item. Keeps them on-site instead of bouncing to search.
• "Frequently bought together": Complementary products that increase average order value: a dress paired with shoes, a face wash paired with toner.
• "Recently viewed": Simple navigation aid that reduces friction.

Cart Page

Cart recommendations target shoppers at maximum purchase intent. "Don't forget" add-ons and "complete your set" bundles work because the incremental decision is small relative to the committed purchase. Avoid mid-checkout recommendations: anything distracting from completion is counterproductive.

Homepage

For new visitors, bestsellers and trending products serve as social proof. For returning visitors, "recommended for you" based on history drives engagement. For post-purchase visitors, complementary products capitalize on the engagement window.

Email

Email recommendations reach shoppers outside the session. Post-purchase recommendation emails timed two to five days after delivery, abandoned cart recovery with alternatives, and re-engagement campaigns for lapsed customers all drive significant revenue. Integration with your marketing automation platform is essential.

Testing and Measuring Impact

A/B Test Everything

Every recommendation strategy should be validated with A/B testing:

• Algorithm performance: Does collaborative filtering beat rule-based for your catalog?
• Placement: Does a cart widget increase AOV or distract from checkout?
• Design: Carousel versus grid. Number of products shown. With or without prices.
• Quantity: Four products versus eight versus twelve.

Key Metrics

• Widget engagement rate: Percentage of visitors who interact.
• Revenue per session (influenced): Revenue from sessions with recommendation clicks versus without.
• AOV lift: Shoppers engaging with recommendations nearly always have higher AOVs.
• Incremental revenue: The critical question: are recommendations driving additional purchases, or are engaged shoppers simply higher-intent visitors who would have bought anyway? Run holdout tests to measure.

The Attribution Challenge

Product recommendations create an interesting marketing attribution challenge. A shopper arrives via Google Ads, clicks a recommendation widget, and buys the recommended product. Did the ad or the recommendation drive that revenue?

Both contributed. The ad brought the shopper; the recommendation guided them to the right product. Multi-touch attribution that accounts for on-site touchpoints provides a more complete picture than models focused solely on acquisition channels.

Common Mistakes

Recommending what they already own. Suppress recently purchased products. This requires clean purchase history data.

Ignoring margins. Recommending low-margin bestsellers increases conversion but may decrease profitability. Factor margin into your algorithm: optimize for profitability, not just revenue.

Generic fallbacks. When data is insufficient, many systems default to sitewide bestsellers. Use category-specific popular items or content-based filtering instead.

Too many widgets. More widgets does not equal more revenue. Clutter creates decision fatigue. Test adding and removing widgets to find optimal density.

Connecting Recommendations to Marketing Strategy

Product recommendations interact with your entire funnel:

• Meta Ads dynamic retargeting uses the same product data as on-site recommendations. Ensure consistency.
• Customer lifetime value increases when recommendations successfully cross-sell over multiple purchases.
• Your marketing analytics platform should track recommendation-influenced revenue alongside channel-attributed revenue.

The brands extracting the most value treat recommendations as core growth strategy, not a bolted-on widget.

Getting Started

If you are not running recommendations, start with rule-based "frequently bought together" on product pages and "you might also like" in post-purchase emails. Measure with holdout tests. Layer in algorithmic approaches as data and capabilities mature.

If you are already running recommendations, audit: Are you measuring incrementality? Testing placement? Using clean data? Most brands leave significant revenue on the table through suboptimal configuration.

See pricing for plans supporting full-funnel measurement, or book a demo to see how connecting recommendation data to your attribution model reveals the true ROI of your personalization investments.