Diversity and Exploration

If organic listings are ranked only on click predictions, some listings may be filtered out despite posing indirect long-term benefits to the platform. Other common ranking methods have similar pitfalls.

Promoted addresses issues that may occur with common allocation systems, and can support additional concerns unique to each marketplace. Here are some examples of common problems and Promoted's solutions:

Problem	Underlying Issue	Promoted's Solution
New listings are unable to gain traction	Newer listings have fewer reviews, conversions, and other data, so they tend to rank lower than existing listings. Lower rankings mean less engagement, exacerbating the issue.	Promoted's Blender allocation system allows marketplaces to use customizable ranking rules, including `challenger` rules, that differ based on the slot in the results. For example, a marketplace can use a `challenger` rule to set the N ^th search result to prioritize lisings with fewer than X purchases. This guarantees some visibility for new listings without compromising quality.
Not enough content or price diversity in results	Current solution lacks a robust method for increasing diversity while maintaining quality, due to various possible issues.	Promoted's Bender allocation system uses more than just click predictions to rank results; it is fully customized to your marketplace and provides special rules that directly increase diversity.
Expensive listings are pushed down in the search results or get less engagement than desired	Click and conversion predictions may be lower for expensive listings due to less engagement. Customers may also filter out more expensive listings. (These filters are usually handled by the retrieval system.)	The `maximizer` in Promoted's Bender allocation system allows marketplaces to prioritize specific business goals, like revenue maximization. For example, a current Promoted customer sets the 5 ^th search result to prioritize listings above a minimum price threshold, where applicable.