Catalogue search assumes the shopper knows the right words. For a catalogue of over one hundred thousand products, that assumption hides most of the inventory from most of the people who would have bought it. We added vector-based visual search — find equivalents from a photograph — alongside normalised cross-brand comparison tables.
Shoppers comparing products across brands run into two walls. First, specifications are published in incompatible formats, so putting two models side by side means manually reconciling different units, different attribute names and different levels of detail. Second, and more often, the shopper cannot name what they are looking for at all — they have a photograph of something they saw, and no text query that reliably finds its equivalents. Catalogue search is built entirely around the assumption that the user knows the right words. For a catalogue of over one hundred thousand products, that assumption quietly hides most of the inventory from most of the people who would have bought it.
Product specifications are reconciled into a common attribute model so any two models can be compared row by row, with differences visible at a glance instead of reconstructed by the shopper.
Products are embedded as vectors and searched by nearest neighbour, so a photograph returns visually equivalent items across brands. This serves the shopper who knows what they want but not what it is called — the majority case in categories where model names are meaningless.
Equivalences between competing model lines are surfaced directly, turning 'what is the alternative to this' from a research task into a single click.
Over one hundred thousand products became reachable by photograph rather than only by keyword, which opens the catalogue to shoppers who could not previously describe what they wanted. Combined with normalised comparison tables, the service shortens the path from 'I have seen something like this' to a specific purchasable model. It is in production use.
Tell us what the process looks like today and we will tell you what can be automated — and what should not be.
LET'S TALK
