Nobody finishes a necklace with one component

A single piece of finished jewelry needs components from four or five different listings, and the shopper assembles that basket in one sitting. Whoever is visible at each step gets that part of the order.

The brand had 190 SKUs and no relationship between any of them. A customer landing on its clasps would be shown competitors' cord in Amazon's carousel, competitors' crimps in the sponsored slots, and nothing at all from the brand — which stocked all of it. It was paying to acquire a customer and then handing three quarters of that customer's basket to whoever was bidding on the adjacent step.

We made the catalog behave like a project rather than a warehouse.

• Component compatibility mapped — which cord gauges take which crimps, which clasps suit which weights, expressed as combinations a maker recognizes.
• Self-targeting product ads on the brand's own detail pages, claiming the sponsored slots that were showing competitors' next steps.
• A+ comparison modules built as a project sequence, so each page shows what comes before and after it in the build.
• Search language extended to the piece, not just the part — the necklace and bracelet terms that lead a maker to the first component.

Orders now contain 3.6 of the brand's SKUs on average, against 1.4, and revenue grew 96% in six months on essentially the same number of customers.

ACoS fell fourteen points for the obvious reason: the acquisition cost is paid once and now returns a basket instead of a part. The single highest-leverage change was the least glamorous — buying the sponsored slots on the brand's own pages, which had been quietly advertising the competition to a customer the brand had already paid for.