How Modern Recommender Systems Work (And Companies That Provide Them)

The four core approaches

Collaborative filtering

The classic approach. "Customers who bought X also bought Y." Works well when you have lots of behavioural data. Falls apart for new products and new users (the cold-start problem).

Content-based filtering

Recommends products with similar attributes to what the customer engaged with. Good for cold-start. Less personalised because it doesn't learn from collective behaviour.

Hybrid models

Combine collaborative and content-based, often with machine learning models that weigh different signals based on context. Most production systems today are hybrid.

Sequence and session-based models

Newer approach. Looks at the order of clicks and browses in the current session to predict the next thing the customer wants. Strong for late-night browsing and gift discovery.

The signals that feed them

• Click and view history. What the customer browsed.
• Purchase history. What they bought, when, and at what price.
• Cart abandonment. What they considered but didn't complete.
• Search queries. Explicit intent expressed in their own words.
• Product attributes. Category, price, brand, variants.
• Real-time stock and price. What's available right now.

Companies that provide recommendation engines as a service

Clerk.io

Hybrid model combining collaborative, content-based and session-based recommendations. Marketer-operable. Optional built-in AI agent handles tuning. Native integrations with major ecommerce platforms.

Algolia Recommend

API-first hybrid recommender. Strong for teams with engineering capacity. Trade-offs on the Algolia alternative page.

Bloomreach

Enterprise recommender with CDP-aware signals. Trade-offs on the Bloomreach alternative page.

Nosto

Onsite-led hybrid recommender. Trade-offs on the Nosto alternative page.

Constructor.io

Discovery platform with strong learning-from-click ranking.

TL;DR

• Modern recommenders are hybrid: collaborative + content-based + session-based, with machine learning weighting the signals.
• The signals that matter: clicks, purchases, cart abandonment, search queries, product attributes, real-time stock and price.
• Clerk.io, Algolia Recommend, Bloomreach, Nosto and Constructor.io are commonly evaluated providers in 2026.