Search and merchandising
E-commerce Search & Product Discovery Architecture
A vendor-neutral cornerstone guide to catalog indexing, facets, relevance controls, merchandising workflows, feedback loops, and system boundaries.
Sources checked: October 5, 2026
Start with an index contract, not a search box
Search quality begins before a query is entered. Define which system creates each product and variant identity, which records are eligible for indexing, which fields are searchable, which attributes are filterable, which availability or catalog-assignment signals must be current, and what makes a record safe to publish. Keep stable identifiers in the index so operators can trace a result to the catalog source, the storefront representation, and any downstream buyer or order context. An index is a derived representation: it needs an accountable owner, a documented refresh or event path, and a repair procedure when a record is incomplete or stale.
Design facets as product-data decisions
Facets turn catalog attributes into navigational choices, so their quality depends on vocabulary, normalization, coverage, and buyer meaning. A useful facet has a clear label, a governed value set, enough products behind each value to be meaningful, and a known source of truth. Decide whether each attribute is global, market-specific, account-specific, or only appropriate after a category is chosen. Test combinations that a buyer would reasonably make, including empty-result states and conflicting selections. Do not assume that every attribute in a PIM or ERP should become a storefront filter; some fields are operational, sensitive, too sparse, or too volatile for discovery.
Make relevance controls explainable and bounded
Relevance should be treated as a collection of explicit decisions rather than an opaque vendor setting. Document the searchable fields, field importance, availability treatment, stock or catalog constraints, sort behavior, and any boosts, suppressions, or redirects. For every manual control, record the purpose, owner, effective period, affected queries or categories, review date, and rollback path. This makes a seasonal change or a correction to poor results inspectable after the fact. A control may be appropriate for a defined business condition, but it should not silently override the catalog or become permanent merely because no one revisits it.
Govern synonyms as a shared vocabulary
Synonyms can help when buyers use different terms for the same product concept, but they can also broaden results beyond a buyer’s intent. Maintain a small, reviewable vocabulary that distinguishes equivalent terms, directional expansions, abbreviations, model-family language, and terms that should remain separate. Source candidates from search queries, support language, sales terminology, and catalog gaps; then test the results against representative catalog records before publishing. Assign an owner for additions, removals, locale variants, and expiry. Treat a synonym change as a relevance release with observable effects, not as a one-time dictionary upload.
Run merchandising as an accountable release workflow
Merchandising controls—such as pinning, hiding, redirecting, or changing a result set for a context—need the same discipline as other customer-facing changes. Begin with a written intent: the audience, the query or browse context, the product set, the start and end conditions, the catalog dependencies, and the expected support impact. Review the experience on representative devices and buyer contexts, confirm that promoted records remain eligible, and schedule a removal check. Keep a change log that distinguishes a deliberate editorial decision from an indexing or relevance defect, so the team can correct the right layer.
Close the loop with query and interaction evidence
Instrument the discovery journey so a team can connect a query or browse context to the returned set, visible filters, selected refinement, item interaction, and any later outcome that is meaningful to the operating model. Preserve the identifiers needed to join those events without inventing a single universal success metric. Review zero-result queries, low-engagement queries, refinements that repeatedly narrow to nothing, unexpected result changes after catalog updates, and manual overrides that outlive their purpose. Pair quantitative signals with catalog and support review: a low-interaction query can indicate vocabulary mismatch, unavailable inventory, incomplete product data, an unsuitable facet, or a buyer intent that belongs in a different workflow.
Keep PIM, ERP, portal, platform, and integration boundaries explicit
The discovery layer should consume governed product and buyer-context data rather than becoming the hidden authority for it. Link product enrichment and taxonomy to PIM ownership; availability and operational item status to the ERP or inventory boundary; company-specific eligibility to the B2B buyer portal model; and index updates to a versioned integration contract. The storefront platform defines which documented search surfaces and rendering patterns are available, while integration architecture determines how changes are delivered, retried, reconciled, and explained. Review the related PIM, ERP, B2B buyer portal, platform architecture, B2B commerce integration, tech-stack decision, and vendor-evaluation guides before deciding where a discovery requirement should be implemented.
Use vendor documentation as capability evidence, not an architecture substitute
Current first-party documentation offers useful implementation references without prescribing a universal stack. Shopify documents search and discovery capabilities that include storefront filtering, product boosts, and search analytics. Algolia documents configuration for faceting, synonym management, rules with conditions and consequences, and click or conversion events that can be associated with a search request through a query ID. Elastic documents synonym sets and search-time synonym analysis. These capabilities differ in configuration model and operating implications; use the first-party source register below to verify the current surface, then test the specific workflow, catalog scale, buyer context, change controls, and recovery needs that apply to your organization.
Editorial note
This guide is education-only technical research for B2B e-commerce teams. It does not provide legal, tax, accounting, financial, security, performance, pricing, compliance, or individualized implementation advice. Verify current vendor documentation, your own policies, and your operating constraints before making an architecture or tooling decision.
First-party source register
First-party documentation and standards referenced during editorial review.
Continue reading
Commerce platforms
Product Master Migration and Catalog Cutover: A Runbook for Commerce Teams
Read guideAutomation and integrations
Commerce Data Synchronization Observability: Freshness, Ownership, and Repair
Read guideCustomer operations
B2B Customer-Specific Catalogs and Contract-Pricing Operations
Read guideStorefront architecture
Shopify App Architecture: Themes, Apps, APIs, and Extension Points
Read guideAutomation and integrations
B2B Inbound EDI Advance Shipment Notices / DESADV: PO Matching and Receipt Reconciliation
Read guideStorefront architecture
The E-Commerce Tech Stack Decision Guide
Read guideData and developer workflows
Vendor Evaluation Scorecard for an E-commerce Technology Stack
Read guideData and developer workflows