Jewellery breaks most e-commerce assumptions: heavy variants, slow-moving hero pieces, photography that carries the sale. What we learned ordering collections on a real jewellery store.
The engine inside SellRank was not designed in the abstract. It was built for a live jewellery brand's collections, and jewellery turns out to be a usefully hostile place to learn merchandising, because it breaks several rules that hold elsewhere.
One product is twenty products
A ring is a metal, a stone shape, a carat weight and a size. The variant count per product runs into the dozens, so 'units sold' at the product level hides enormous variation underneath, and stock is never a single number. Any ranking that treats a product as one atom will mislead on a catalogue like this — which is why stock-aware rules have to reason about availability across variants rather than a single on-hand figure.
The expensive piece rarely sells, and still belongs on the page
In jewellery, the highest-value pieces anchor perception. Sorting purely by orders buries them, and the collection starts to look cheaper than the brand is. This is the clearest argument for percentage slots over a single ranking: reserve a share of the page for the pieces that do the positioning work, and let the rest of the grid be ruthlessly demand-driven.
Photography is a ranking signal nobody records
Two comparable pieces can perform completely differently because one photographs better. No metric in Shopify captures this, and no automatic order ever will. It is the single best argument for a permanent human override — a merchandiser who knows which pieces shoot well needs to pin them, and needs those pins to survive every scheduled re-sort.
Margin varies wildly across a page
Metal prices and stone costs mean two products at the same price can have very different margins. Sorting by revenue alone quietly favours whichever happens to be cheaper to sell. Blending profit into the score changes which pieces lead — as long as cost per item is actually filled in, which on most jewellery catalogues is the first data-hygiene job before any of this works.
Seasonality is sharper than you expect
- Gifting seasons move entire categories in days, not weeks — a monthly window is too slow to see it.
- Bridal and everyday pieces behave like two different stores sharing a catalogue.
- A short window catches the swing but is noisy on slow-moving pieces; blending a short and a long window is usually the right compromise.
None of this is jewellery-specific in the end. Any catalogue with heavy variants, wide margin spread and photography-led conversion — furniture, fashion, watches — hits the same walls. It is simply that jewellery hits all of them at once, which made it an unusually good place to build the engine.
