How Promoted Works

Introduction optimizes marketplace search, feed, and native ads for an extra 5-10% revenue per user. Each marketplace has unique needs, and Promoted offers a variety of options to accommodate your own platform. Promoted’s advanced model builds on top of your existing search and discovery stack — not instead of it — so results are always incremental and leverage any existing investments. Promoted's surfacing optimizations increase revenue, relevancy, and engagement with extremely low latency data logging, so buyers and sellers will have an improved experience without sensing any changes in the platform. Promoted encrypts your data, and does not own or sell any of your data or intellectual property.

Promoted listings. In search and discovery, promote the best listings at the top in real time, by leveraging Promoted's search ranking ML while protecting the user experience with ad quality controls.

Search ranking. Time-tested strategies like A/B testing reveal increases in gross sales, conversions, profits, new seller activation, and other customizable, long-term business objectives.

Deep introspection. An unparalleled ML data-centric view into how search ranking works to understand what is relevant, what is most profitable, what is promoted, and why, avoiding bugs before they happen.


  1. Promoted’s team will help you integrate with Promoted’s performant SDKs or APIs.
  2. You'll log existing rankings and engagement and content data to Promoted.
  3. Promoted’s system optimizes ranking using our advanced model, including Blender, our internal ML tool that combines metrics and outputs a ranking order. There is no change to the end user experience except that search, recommendations, and ads are now dramatically better.

Responding to Search Queries

The main steps are summarized below. For more detail, see the next section for a comprehensive example using a current Promoted client. This is the general process after a user submits a search query:

  1. You will fetch and send hundreds of candidate response items to Promoted for selection and ranking. Along with the items, send all metadata, including any ranking, quality, relevance, or personalization scores. Promoted incorporates all of your existing signals into its own model using “model stacking.”
  2. Promoted’s heavyweight machine learning scores and chooses among these items. Our efficient C++ and Flink systems combine millions of signals extracted from your data, user engagement metrics, in-house or vendor scores, item metadata, and contextual information, like location, time, and season. For ads, this includes computing "insertion bids" for different ad optimizations (e.g., CPC) to be used in pricing.
  3. Using Blender, Promoted accounts for specific filtering rules, item diversity, business objectives (e.g., optimizing profitability), and ad load and quality controls (if applicable), to choose an optimal presentation of organic listings and ads.
  4. Promoted logs and returns the results to your in-house platform to display as if they were returned by your existing retrieval system. The only change to the end-user's display or experience is that results will be better and more relevant.
Process of ranking and displaying results for a user's search query, with Promoted

Process of ranking and displaying results for a user's search query, with Promoted