Internal search for ecommerce & marketplaces

Internal search, diagnosed and improved.

Ranking, relevance, query understanding, and search analytics — evaluated and refined, independent of your search vendor.

Experience with large-scale ecommerce search systems — Algolia, Elasticsearch, OpenSearch, Typesense, Luigi's Box.

Most teams assume their search works well enough

Until you examine real queries closely, ranking issues, weak query handling, and dead ends stay invisible. The search engine returns results — just not the right ones, often enough to matter.

Ranking misses intent

Results look plausible at a glance but don't match what the customer meant — especially on your most important queries.

Query handling breaks quietly

Synonyms, typos, and compound queries return poor or empty results — and no one is alerted.

Dead ends cause silent exits

Zero-result pages and no-click searches push people away, but typical analytics don't flag them.

No way to measure quality

Without structured evaluation, teams can't tell whether search is improving, degrading, or standing still.

Diagnostic patterns

What we find when we look closely

These patterns appear in most ecommerce search systems. They're rarely visible in dashboards.

Relevance

Results look right. They're not.

Top queries return plausible products, but bestsellers are buried and weak matches surface first. The search appears functional — the ranking is wrong.

Query interpretation

Queries break without anyone noticing

Synonyms missing. Compound queries split wrong. Attribute searches like “red dress size 38” treated as free text. No alerts, no fallback.

Ranking

Ranking rules nobody owns

Boosting rules layered over months. Conflicting weights across categories. Result order changes and no one evaluates the impact.

Zero results

Silent dead ends

Queries return nothing — no redirect, no suggestion, no signal. Users leave. It happens most on long-tail and misspelled queries.

Quick diagnostic

Test your search in 5 minutes

Run three simple checks that often reveal hidden ranking or query interpretation problems.

Run the quick search check →

3 checks · 5 min · no setup

Diagnostic methodology

How we diagnose search

A structured process applied to every search system we evaluate.

01

Real queries

Analyze actual user queries — high-volume, high-intent, long-tail, and failure patterns.

02

Query interpretation

How the system reads the input. Tokenization, synonyms, compound handling, attribute mapping.

03

Result quality

Whether results match intent. Position accuracy, dead ends, coverage gaps.

04

Ranking logic

Boosting rules, attribute weights, merchandising overrides, and their cumulative effect.

05

Improvement roadmap

Prioritized, actionable changes your team can implement. No dependency on us.

Each step informs the next. The process repeats as the system evolves.

Search frameworks

Conceptual tools we use to analyze how search systems behave, where they fail, and how relevance should be evaluated.

Framework 01

Relevance evaluation

A structured method for measuring whether search results match user intent — beyond click-through rates and conversion proxies.

Inputs:

  • Representative query test sets
  • Human relevance judgments
  • Before/after result comparison

Framework 02

Query taxonomy

A classification system for query types. Each type has different failure modes and requires different ranking logic.

Types:

  • Navigational (exact product)
  • Attribute (color, size, material)
  • Exploratory (broad category)
  • Long-tail (rare or compound)

Framework 03

Ranking failure modes

A diagnostic checklist for identifying why result ordering breaks down. Used to trace ranking problems to their configuration root cause.

Common modes:

  • Boosting rule distortions
  • Synonym and tokenization gaps
  • Category or attribute bias

Deep, narrow expertise

We only do one thing: internal search for ecommerce. That focus means faster diagnosis, sharper recommendations, and less wasted scope.

Vendor-agnostic

Algolia, Elasticsearch, Luigi's Box, Doofinder, Coveo, Bloomreach, and others. The methodology applies regardless of your stack.

Experiment-driven

Every recommendation includes testable hypotheses and success metrics. Your team can prioritize and validate with controlled experiments.

Multilingual

Audits conducted in English, German, French, Spanish, Italian, Dutch, and Swedish. Query analysis in the language your customers actually use.

No lock-in

We deliver a diagnosis and a roadmap. Your team or vendor handles the build. No retainer, no ongoing dependency.

Search environments we've worked in

Catalog scale

10k–10M+ products

Query volume

High-traffic ecommerce

Ranking complexity

Boosting, weighting, rewriting

Platforms

Algolia, Elasticsearch, OpenSearch, Typesense

Languages

Multilingual European markets

Also Doofinder, Luigi's Box, Coveo, Bloomreach, and other platforms. The diagnostic methodology applies regardless of vendor.

What the diagnostic produces

A clear picture of where search is failing and a prioritized plan for what to fix.

Diagnosis

A clear account of where and how search is failing — broken down by query understanding, ranking behavior, coverage, and evaluation gaps.

Query analysis

Examples of problematic query patterns drawn from real traffic: misinterpreted compound queries, attribute failures, silent zero-result cases.

Ranking observations

An assessment of how ranking logic behaves across query types — where configuration is working and where it is distorting results.

Evaluation perspective

An honest view of what is and is not being measured, and what a minimal structured evaluation process would look like for the system.

Improvement roadmap

A prioritized list of changes — ordered by impact and feasibility — that your team or vendor can act on directly.

Example diagnosis

Ranking failure investigation

A simplified version of the diagnostic process used during a search audit. The query, the observed result behavior, and the likely root causes.

Query

black running shoes

Observed result behavior

Top-selling SKUs ranked below weak text matches. Position 1–3 occupied by low-conversion products with partial keyword overlap.

Identified root causes

  • Boosting rule on “new arrivals” overriding textual relevance score
  • Color attribute not indexed as a filterable field
  • No ranking validation against a representative query set

Different query. Same diagnostic structure.

From real audits

Failures we find repeatedly

Specific patterns that appear across search systems, regardless of vendor or catalog size.

  • Merchandising rules overriding textual relevance
  • Synonym lists masking deeper indexing gaps
  • Attribute queries parsed as free-text search
  • Compound queries returning partial or empty results
  • Zero-result queries with no fallback or redirect
  • Ranking configuration drifting without evaluation
  • Boosting rules conflicting across product categories

If search feels hard to evaluate, that's usually a signal

Results look acceptable. Confidence is low. Something feels off but there's no clear evidence yet.

That's exactly the kind of system we diagnose. We examine real queries, ranking behavior, and evaluation gaps — then give your team a clear picture of what's happening and what to fix first.