Vertical
Direct-to-consumer swimwear and resortwear
A cold read of a catalog where nobody can try anything on — so fit has to live in a size run and a set of structured attribute tags, or it does not reach the shopper at all. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.
How EKOM read it. The engine took the live public catalog cold — no file, no credentials, no schema — profiled the vertical and the field semantics on its own, then read every record against its siblings and against how each field is used catalog-wide. That context is what turns a valid-looking string into a finding: a sibling record carrying a tag its twin is missing, an image filename that names a different style inside a product where every other file obeys the convention, a coverage value that is plausible until you see the one its sibling uses. Nothing was supplied about what these fields mean. A catalog with no conventions produces no findings; it produces confusion.
And then it went back. Before the client report shipped, the findings carrying the most weight were taken back to the live storefront and confirmed still present, and the ones raised from an inference were pushed at the field that would settle them. On the image finding that meant checking the storefront's own primary-image field — the one that actually decides which photograph renders — which returned the same mismatched file the filename had pointed to. Those findings ship marked verified live. It is the difference between a client receiving leads and a client receiving a list they can act on.
From diagnosis to channel-ready. The 15 findings collapse into six upstream mechanisms — more than one item code alive for one garment, one attribute with more than one live vocabulary, sibling records that diverge only in their tags, copy and assets attached to the wrong record, published pages with no availability check in evidence, and records live before their copy and classification. Fixing findings fixes findings; fixing mechanisms fixes every product that has already passed through them and every product that will. Seven further findings were withdrawn before the client report was written, and the case study groups and explains them rather than dropping them quietly. Separately, two of the fifteen were raised with a stated limit rather than withdrawn — because a finding that is right with a boundary named around it is worth more than a finding removed for being hard to prove. One of those two limits was then closed by the live re-check; the other is still declared in the delivered report, because the question behind it belongs to the client rather than to the data.