Search Analytics: The Queries That Tell You What's Wrong With Your Catalog

Every query is a customer telling you what they want

Search logs are the most honest feedback you have — shoppers describing their intent in their own words, at scale. Ecommerce search analytics is the practice of reading those logs as a to-do list for the catalogue and the relevance config, not just watching a dashboard.

The reports that matter, and what each one means

• Zero-results queries. Unmet demand — a stocking gap, a missing synonym, or a content hole.
• High-exit queries (results returned, shopper left). A relevance problem — the results looked wrong.
• Low-conversion top queries (lots of searches, few purchases). Either the results are off, or the products themselves have thin content, bad images, or poor pricing.
• Refinement rate (how often shoppers add filters or re-search after the first result set). High means the first set isn't good enough.
• Query phrasing trends. Shoppers shifting to longer natural-language queries is a signal your search needs to handle intent, not just keywords.

How the merchandising team uses it

The boost-and-bury console is downstream of these reports. A cluster of boosts around one supplier's products usually means that supplier's product data is poor. A pile of zero-results for a category means a stocking or synonym gap. On our semantic search project the console and the analytics live together, so a merchandiser can see the failing queries and act on the same screen.

The feedback loop

Review weekly. Act on the top items. Watch whether the metric moves. Search analytics that nobody follows up on is a dashboard; search analytics with an owner and a weekly action list is a compounding improvement to relevance and catalogue quality.

The weekly review, concretely

Fifteen minutes, the merchandising lead plus whoever owns the relevance config. Pull three lists: the top 20 zero-results queries, the top 20 high-exit queries, and any query whose conversion dropped week-on-week. For each item, pick one action: add a synonym, add a boost or bury rule, flag a stocking gap to buying, or flag a product-data gap to the catalogue team. Note who owns each action, and next week check whether the metric moved. That's the whole loop — small, repeatable, and it compounds.

The trap of vanity search metrics

"Search usage is up 10%" is not a win on its own. It can mean navigation is broken and shoppers are searching because they can't browse. Pair every volume metric with a quality one — usage with search-conversion, query count with zero-results rate — or you'll celebrate a number that's telling you something's wrong.

What not to over-read

A single weird query is noise. Look at clusters and trends. Chasing every individual low-conversion query is a treadmill — fix the systemic causes (data quality, relevance config), not the symptoms one by one.

The reports you can mostly ignore

Average results per query, total number of searches, and "search coverage" percentages are largely vanity — they go up and down without telling you what to do. The four that reliably drive an action are zero-results queries, high-exit queries, high-volume low-conversion queries, and refinement rate. Build the weekly review around those and skip the rest of the dashboard.

Where this stops being right

• Analytics without an owner is a dashboard nobody reads — assign it to the merchandising team with authority to act.
• Small query volume makes the reports noisy — aggregate over longer windows before drawing conclusions.
• Acting on every query individually is unsustainable — group the failures by root cause.

FAQ

What's the single most useful report? Zero-results queries — a direct list of demand you're not meeting.

What does a high-exit query tell me? The results looked wrong to the shopper — a relevance problem, not a stocking one.

Who should own search analytics? The merchandising team, reviewed weekly, with authority to act on catalogue content and relevance rules. If it sits with engineering or analytics as a reporting task, the insights get produced but not acted on — the value is in the follow-up, and that needs someone who can change the catalogue and the ranking config.

How far back should the reports look? A rolling four to eight weeks for trends, plus the last week for what needs action now. Longer windows smooth out noise; shorter ones catch a regression from a recent change.

ISTRALLEN builds search with an analytics view the merchandising team acts on weekly; see AI for Retail.