Tune and merchandise

Working the zero-results report

The weekly loop.

The list of searches that returned nothing is the most valuable document your shop produces. Every line is a buyer who wanted something specific, did not get it, and either phoned or left. This article is the weekly routine for working that list.

Cockpit screens: planned. The Search app screen that lists no-hit queries is not yet on your tenant. The method on this page does not depend on it: today the list comes from your storefront's search log or an export of it, and every fix below is a data fix you can make now. Decide who owns the routine before you start. It is a thirty-minute job once a week and it belongs to whoever owns the catalog, not to IT; almost none of the fixes are search-engine settings.

Where the list comes from

Until the Search app ships, ask whoever runs your storefront for a weekly export of searches with zero hits. You need four columns:

  • The query text, exactly as typed.
  • How often it was searched in the period.
  • Which channel and which locale it came from.
  • Whether the buyer was signed in.

The planned zero-results list in the Search app is meant to show the same four columns, per channel and locale. Whichever source you use, sort by frequency, never alphabetically. The top twenty is the whole job; the tail is typos and noise.

Classify each line into one of four buckets

This is the entire method. Do not fix lines as you read them. Classify first, then fix in batches.

BucketHow to recognise itFixOwner
Vocabulary gapYou sell it, under a different wordAdd the word to the product, or a synonymCatalog
Data gapYou sell it, but the searched fact is not in the dataFill the attribute; check completenessCatalog
Assortment gapYou sell it, but not in this channel or to this customerAssign the product, or accept itSales
Genuinely not soldYou do not stock itNothing, or a sales leadPurchasing

Two more kinds of line are not gaps at all and should be filtered out before you start:

  • Not a product query. AGB, Rechnung, Ansprechpartner, Retoure. These belong in redirect rules once they ship, and in most B2B shops they are a surprising share of all searches.
  • Typos and pastes. A single occurrence of verschraubng needs no action; the engine's typo tolerance already catches most of it. A repeated typo is worth a synonym.

The unit-and-format bucket hides inside the first two

A fifth pattern deserves its own pass because it looks like a vocabulary gap and is a data gap. Buyers type 1/2" where the data says 1/2 Zoll, M8x40 where the attributes hold M8 and 40 mm separately, 10bar where the rating is a number in a measurement family. The engine's prefix search and typo tolerance do not bridge these. The fix is a searchable, normalised value on the product, in name or description, where the index reads it today. A synonym per format is the wrong tool; you would need one per size.

The assortment gap is the ambush

Of the four buckets, this is the one that costs you orders silently.

A channel set to Hidden until assigned (unassigned_visibility = assigned_only) hides unassigned products from search in that channel too. Correctly, and invisibly. The article exists in the Cockpit. You can see it. The buyer cannot.

The signature is a query that returns nothing for the buyer and returns results for you, or for a buyer on a different channel. Before you conclude the data is missing, open Configuration › Channels, read the Unassigned rows setting of the buyer's channel, and check the product's assignment to it. See Product not visible.

Fix at the source

For every line, ask which layer the fix belongs in. The order matters, because the cheapest fix is often the wrong one.

  1. Is this a missing product attribute? Fill it. The fix then also improves facets, exports, punchout catalogs and classification. A synonym fixes only search.
  2. Is this a word missing from the product? Add it to the name or the description. Better than a synonym, for the same reason, and it works today: those are the fields the index reads.
  3. Is this genuinely a different word for the same thing? Now a synonym is right.
  4. Is this not a product at all? Redirect.
Resist the urge to synonym your way out of a data problem. A synonym list that has grown past a few hundred entries is usually a catalog that was never enriched, patched one query at a time. It is unmaintainable, it does not export, and it does not help the buyer who filters rather than searches.

The weekly routine

  1. Pull the last seven days of no-hit queries, sorted by frequency.
  2. Remove the non-product queries and the one-off typos.
  3. Classify the top twenty into the four buckets. Ten minutes.
  4. Batch the fixes: all attribute fills as one bulk edit, all assortment questions in one message to Sales, all synonyms in one sitting once the screen exists.
  5. Write the "genuinely not sold" lines into a list for Purchasing. Over a quarter this list is real demand data, and it is free.
  6. Re-run the same queries after the index has picked up your changes and confirm they now return something sensible.

What to check

  • The same query does not appear in next week's list. If it does, the fix did not take, usually because the change went to the wrong channel or locale, or the index had not caught up when you tested.
  • Your zero-result rate is falling. As a working target, a maintained B2B catalog sits under three percent of searches; a rate above ten percent means the vocabulary work has never been done.

Next