Cold Start to Improve Market Thickness on Online Advertising Platforms: Data-Driven Algorithms and Field Experiments

Abstract

Cold start describes a commonly recognized challenge for online advertising platforms: with limited data, the machine learning system cannot accurately estimate the click-through rates (CTR) of new ads and, in turn, cannot efficiently price these new ads or match them with platform users. Traditional cold start algorithms often focus on improving the learning rates of CTR for new ads to improve short-term revenue, but unsuccessful cold start can prompt advertisers to leave the platform, decreasing the thickness of the ad marketplace. To address these issues, we build a data-driven optimization model that captures the essential trade-off between short-term revenue and long-term market thickness on the platform. Based on duality theory and bandit algorithms, we develop the shadow bidding with learning (SBL) algorithms with a provable regret upper bound of [Formula: see text], where K is the number of ads and d captures the error magnitude of the underlying machine learning oracle for predicting CTR. Our proposed algorithms can be implemented in a real online advertising system with minimal adjustments. To demonstrate this practicality, we have collaborated with a large-scale video-sharing platform, conducting a novel, two-sided randomized field experiment to examine the effectiveness of our SBL algorithm. Our results show that the algorithm increased the cold start success rate by 61.62% while compromising short-term revenue by only 0.717%. Our algorithm has also boosted the platform’s overall market thickness by 3.13% and its long-term advertising revenue by (at least) 5.35%. Our study bridges the gap between the theory of bandit algorithms and the practice of cold start in online advertising, highlighting the value of well-designed cold start algorithms for online advertising platforms. This paper was accepted by Gabriel Weintraub, revenue management and market analytics.

Estimating Effects of Long-Term Treatments

“…Online controlled experiments, often referred to as A/B tests, have become the gold standard for evaluating the impact of product updates on digital platforms. These updates can include the introduction of new product functions, user interface designs, and recommendation algorithms (Bakshy et al 2014, Bojinov and Gupta 2022, Larsen et al 2022, Xu et al 2015, Ye et al 2023a. By randomly assigning experimental subjects (e.g., users) to different groups and exposing them to different product versions, A/B tests can rigorously measure the effects of the product update on specific metrics of interest, facilitating data-driven business decisions.…”