Decision Density Drives Marketplace Personalization
Andrew Yates, CEO of Promoted.ai, on driving marketplace ARPU with personalization
The real constraint on marketplace machine learning is not catalog size, it is decision density. A marketplace becomes optimizable when the app must continuously choose among ranking, price, timing, availability, subsidy, and fulfillment tradeoffs. That is why a dark store with a short list of SKUs can still need heavy optimization once delivery time, stock levels, margin, and supplier priorities start changing order by order, and why Uber needed major ML despite selling a simple ride request.
-
In the Promoted interview, the key idea is that ML only matters when the product has multiple live decisions embedded in it. In Uber, the decision is not just which ride, it is which driver, at what ETA, at what price, with what downstream effect on marketplace balance and unit economics.
-
Owning supply does not remove the need for optimization, it shifts where the decisions sit. In a JOKR style model, the app can suppress demand for constrained inventory by raising price or reducing placement, while inventory ownership adds new constraints around what is in stock and what suppliers want pushed.
-
The biggest marketplaces turn these decisions into profit centers. Amazon built a $360B third party marketplace and $48B advertising stream, while Mirakl sells the software layer that helps retailers create similar seller, ranking, and ad decisions. Once there are enough sellers and enough user attention to allocate, optimization compounds into margin.
The next wave of marketplaces will look less differentiated by whether they own inventory and more by how many decisions they can instrument in real time. As logistics, merchandising, and retail media converge inside one app, the winners will be the companies that turn every impression, search result, and delivery promise into a measurable economic choice.