How to build a price comparison engine

A comparison engine is four problems stacked. Identity, so two listings that are the same product resolve to one record. Normalisation, so prices are comparable — landed cost including shipping, and unit price where pack sizes differ. History, so a discount claim can be defended. And freshness, so the number you show is still true when someone clicks.

Identity is where most projects fail. With a valid GTIN it is a join and essentially free; without one it is inference, and inference produces confident errors. Publish your match confidence on the page. It is a trust signal that costs nothing because the number already exists internally.

Normalise before you rank. A product that is cheaper on sticker and more expensive delivered is not cheaper, and a comparison that ranks it first is worse than showing nothing.

For discount depth, use percentile against the product's own history at that seller rather than percent-off a reference price the seller chose. Comparison sites get into trouble over inflated reference prices; they do not get into trouble over an auditable statement about what a product has actually cost.