Search and Merchandising

When to use

• Selecting or tuning a product search engine (Elasticsearch, Algolia, Typesense, Meilisearch)
• Implementing or improving search relevance (field boosting, synonyms, function scoring)
• Building faceted navigation with dynamic facet counts and URL-preserving filters
• Implementing a recommendation engine (collaborative filtering, content-based, or hybrid)
• Running A/B tests on search ranking, listing layouts, or recommendation placements
• Adding personalization to search results, homepage, or email recommendations
• Implementing technical SEO for product pages (schema.org, canonical URLs, hreflang, sitemaps)

Core principles

1. Boost by behavior, not just text match — sales count and view count as function score signals outperform pure BM25 for commerce
2. Facet counts must stay accurate under multi-select — post-filter aggregations; users abandon nav when counts are wrong
3. Cold start kills collaborative filtering — always have a content-based fallback for new users and new products
4. Canonical every filter combination — faceted nav generates thousands of URL permutations; without canonicals, you split ranking signals
5. Hash user ID for A/B assignment — session-based assignment causes variant flicker on login; user ID gives consistent experience

Reference Files

• references/search-facets-recommendations.md — search engine selection, relevance tuning, autocomplete, facet types and implementation, collaborative/content-based/hybrid recommendation patterns and placement strategy
• references/abtesting-personalization-seo.md — what to A/B test in search and merchandising, feature flag implementation, personalization pipeline, schema.org markup, technical SEO for product pages