Wayfair Sponsored Products (WSP) is a cost-per-click (CPC) advertising program aimed at improving product discovery and driving sales. At the heart of WSP is a ranking function that determines the placement of ads on a page. In this paper, we introduce a new ranking function derived from an optimization framework that maximizes overall profitability by jointly considering advertising revenue from clicks and profit from resulting sales. To maintain a positive customer experience, the formulation includes a relevance constraint. We evaluate the approach through offline simulations using counterfactual estimates of clicks and orders, analyzing the impact of different parameter settings on key metrics such as ad revenue, total profit, and estimated conversion rate (CVR). Finally, we share results from a successful online test, that validated the effectiveness of the proposed method, leading to its full deployment in production, and highlight key learnings from the study.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Profit Aware Ad Ranking with Relevance Constraint

  • Manavender Malgireddy,
  • Dongyue Xie,
  • Sergey Kolbin,
  • Kurt Zimmer,
  • Masoum Mosmer,
  • Patrick Phelps

摘要

Wayfair Sponsored Products (WSP) is a cost-per-click (CPC) advertising program aimed at improving product discovery and driving sales. At the heart of WSP is a ranking function that determines the placement of ads on a page. In this paper, we introduce a new ranking function derived from an optimization framework that maximizes overall profitability by jointly considering advertising revenue from clicks and profit from resulting sales. To maintain a positive customer experience, the formulation includes a relevance constraint. We evaluate the approach through offline simulations using counterfactual estimates of clicks and orders, analyzing the impact of different parameter settings on key metrics such as ad revenue, total profit, and estimated conversion rate (CVR). Finally, we share results from a successful online test, that validated the effectiveness of the proposed method, leading to its full deployment in production, and highlight key learnings from the study.