Notes
Research on product data and discovery, with every figure attributed. Where a number is ours it says so, and the command that reproduces it is on the method page.
Completeness measures whether the fields you defined are filled. It cannot tell you whether you defined the right fields, and the difference is where discovery quietly fails.
6 min readAdobe scored US retail product pages at 66% machine readability, the lowest of any page type. Product pages are exactly where the comparison detail an agent needs is supposed to live.
8 min readAdobe's figures show the reversal clearly, and Salesforce put AI agents behind more than a fifth of holiday retail sales. Both numbers raise the same question about your catalogue.
7 min read76% of retail search ad spend goes to Shopping formats. Those formats match on feed attributes, which makes a missing field a bidding problem you cannot fix by bidding.
7 min readOnboard four thousand listings into a million-listing marketplace and coverage barely moves. The category those listings landed in can become unshoppable and nothing in the reporting says so.
6 min readIndustrial catalogues were built for buyers who know the part number. Everyone else is filtering on thread size and voltage, and those live in a supplier datasheet nobody mapped.
6 min readA traveller who needs step-free access and a cot is reading forty descriptions looking for two sentences. The facts are usually written down. They are just not fields.
6 min readA bit is one halving of the shelf. That fixed meaning is the only reason two attributes as different as screen size and processor can be compared at all.
8 min readWe have run no live experiment on merchant traffic, so we have no uplift number. Here is the experiment we would run instead, and why publishing the design is worth more than publishing a percentage.
6 min readWe write when we have measured something worth writing about, which is not on a schedule.