Target return-on-ad-spend (tROAS) autobidding has become a dominant paradigm in large-scale advertising systems because it replaces manual bid tuning with a single, business-aligned objective: to deliver conversion value efficiently while meeting a return-on-investment target.

In the paper, we present an end-to-end tROAS system for Wayfair Sponsored Products in a multi-slot cost-per-click (CPC) auction, where an auction-time value-based bid is scaled by a campaign-level control parameter that is updated from delayed and noisy realized ROAS. We specify the interfaces among bid generation, score-based ranking, generalized second-price (GSP)-like charging, and campaign-level feedback control; introduce deployment guardrails for sparse-conversion campaigns, including cold-start handling, adaptive learning rates, and bounded multiplier updates; and quantify the intrinsic achievability of ROAS targets through an offline analysis. This analysis establishes a marketplace-specific noise floor for ROAS attainment and yields campaign-eligibility criteria based on data volume. Finally, we report findings from a large-scale online deployment and subsequent post-launch analysis, showing that 74% of enrolled campaigns with sufficient order volume achieved ROAS within ±20% of their target after the initial learning period, with attainment approaching the intrinsic achievability estimated offline. An observational study comparing tROAS to Wayfair’s manual bidding baseline finds tROAS delivers a 11-15% efficiency gain. In general, the paper contributes to a deployable, interpretable, and scalable tROAS system, together with the operational lessons required to make value-based bidding work in a sparse and high-variance marketplace.