E-Commerce Search Feature Spec Template

For catalogs under 100,000 SKUs, a managed service (Algolia, Typesense, Meilisearch) is almost always the right choice. You get autocomplete, faceting, typo tolerance, and relevance tuning out of the box. Build in-house only if you have search-specific requirements that managed services cannot support (e.g., custom ML ranking models, multi-tenant isolation, data residency) and the engineering team to maintain it. The total cost of building and maintaining a custom search stack typically exceeds managed service fees until you reach millions of SKUs.

Show 4-6 facets on desktop and 3-4 on mobile. Display the most-used facets first (measure by facet click rate). Hide less common facets behind a "More filters" expansion. If a facet has only one value in the current result set (e.g., all results are the same color), hide it. Dynamic facet relevance based on the query type significantly improves the experience.

Index at the product level, not the variant level. A t-shirt with 12 size/color combinations should appear once in search results. Show available variants as swatches or badges on the result card. When a user clicks a color swatch, update the product image to match. Filter by size/color at the variant level so "blue XL shirt" only shows products with that specific variant in stock.

Show out-of-stock products but deprioritize them. Users searching for a specific product expect to find it even if it is temporarily unavailable. Show a "Notify me when back in stock" button instead of "Add to cart." For broad queries ("summer dresses"), push OOS products below all in-stock results. Track the percentage of search impressions going to OOS products and target keeping it below 5%.

Track five core metrics weekly: (1) Click-through rate on search results (target: 40-60%). (2) Add-to-cart rate from search (target: 8-15%). (3) Null result rate (target: < 5%). (4) Refinement rate, meaning how often users modify their query (target: < 30%). (5) Exit rate, meaning how often users leave after searching (target: < 20%). Combine these into a search quality score and review weekly with the merchandising team.