AI search for B2B catalogs Your catalog speaks product. Your buyers speak problem.

A buyer types what they need to solve. The catalog says what the part is. In between, the results list is often empty. We build search for Magento that connects both and still hits part numbers exactly.

Have your search reviewed
Example

food safe hose for hot caustic

0 results

In the catalog the matching item is called “EPDM cleaning hose, CIP-ready, up to 140 °C”. Not a single word matches.

Why search comes up empty

Keyword search finds words. Not intentions.

  • Different words

    The buyer writes “caustic”, the data sheet says “alkaline cleaning”. To Elasticsearch and OpenSearch those are two different things.

  • Half a part number

    Buyers copy the number from an old delivery note, often only half of it. A purely semantic search then “corrects” it to the wrong item.

  • Knowledge in PDFs

    Temperature range, resistance and standards often live only in the data sheet, not in the product attributes.

  • Not everyone may see everything

    Customer groups, shared catalogs, contract prices: search must not find anything this customer is not allowed to buy.

How we build it

Two searches. One result.

A vector search finds products by meaning, classic full-text search by exact terms. One shared score combines both with your rules.

  1. Prepare

    Descriptions, attributes and data sheets are turned into vectors. When a product changes, only that product is reindexed.

  2. Filter

    Before searching, everything the customer group may not see is removed. The language model only gets the permitted slice.

  3. Find

    Vector and full-text search run in parallel. If the query looks like a part number, the exact match counts more.

  4. Sort

    Stock, customer prices and order frequency feed into the ranking, following rules you define.

Measure first

Search logs first. Then the model.

We start with your real search queries. They become the baseline every change has to be measured against.

  • Searches without results

    How often does a search come up empty, and with which terms?

  • Ranking

    Are the right products at the top? Rated against a list of real queries (NDCG).

  • Clicks on the top results

    Do buyers click at the top, or do they scroll and give up?

Honestly

Not every shop needs AI. Sometimes good upkeep is enough.

Worth it if

  • your catalog has thousands of items with many technical attributes,
  • customers search by application rather than product name,
  • your search logs show many queries without results.

Probably not if

  • your customers order almost only by part number. A well-tuned OpenSearch will do more.
  • the product data itself has gaps. Then fix the data first, for example with a PIM.

Your customers search with binoculars? Search could simply find things.

We look at your search logs and show you which queries go nowhere today.

Have your search reviewed
Black binoculars