TLDR: A new research paper finds that when e-commerce sellers use AI algorithms for both pricing and advertising, and consumers have high search costs, these algorithms can surprisingly learn to collude on lower advertising bids. This reduces seller costs, leading to lower product prices for consumers, increased demand, and higher profits for sellers and the platform. Empirical evidence from Amazon.com supports these findings, suggesting that algorithmic coordination isn’t always harmful.
In the bustling world of e-commerce, where online sellers increasingly rely on artificial intelligence (AI) algorithms to set prices and manage advertising, a new study sheds light on a surprising phenomenon: algorithmic collusion that can actually benefit consumers, sellers, and even the platforms themselves. Traditionally, the concern with AI-driven pricing has been that algorithms might learn to secretly coordinate, leading to higher prices that hurt consumers. However, this new research explores a more complex scenario where algorithms manage both pricing and advertising decisions simultaneously.
The paper, titled “Algorithmic Collusion of Pricing and Advertising on E-commerce Platforms,” by Hangcheng Zhao and Ron Berman from The Wharton School of the University of Pennsylvania, delves into how multi-agent reinforcement learning algorithms behave in a competitive online marketplace. Their empirical strategy involved calibrating these learning algorithms to a vast dataset collected from Amazon.com, covering over 2 million products.
A “Win-Win-Win” for Everyone?
One of the study’s most significant findings is the identification of conditions under which these learning algorithms can create a “win-win-win” outcome. This happens particularly when consumers face high “search costs” – meaning they are less willing to browse through many product listings before making a purchase. In such cases, the algorithms learn to coordinate on lower advertising bids. This reduction in advertising costs for sellers then translates into lower prices for consumers. The lower prices, in turn, enlarge the overall demand on the platform, benefiting the platform through increased sales commissions.
The intuition behind this is fascinating. While algorithms might naturally tend to increase prices for higher profits, they also learn that advertising is a cost. When search costs are high, the benefit of colluding on lower advertising bids (reducing costs) outweighs the benefit of colluding on higher prices. This leads to a scenario where lower prices actually result in higher seller profits due to increased demand and reduced ad spending.
Empirical Evidence from Amazon
To validate their theoretical findings, the researchers analyzed a large-scale, high-frequency dataset from Amazon.com. They estimated consumer search costs for over 2,000 product search keywords, finding a wide range of costs across different products. For many keywords, consumers indeed do not search beyond a small portion of the results page, confirming the presence of significant search costs.
Furthermore, the study developed an “algorithm usage index” based on how correlated product prices were over time. They found a negative interaction between estimated consumer search costs and this algorithm usage index. This means that in markets where consumers have higher search costs, higher algorithm usage is associated with lower prices, providing real-world evidence for their “beneficial collusion” theory.
Also Read:
- Navigating the Legal Landscape of AI Agents: Loyalty and Disclosure in the Age of Autonomous AI
- Enhancing AI Collaboration: How Language Models Learn to Navigate Complex Social Games
Platform’s Strategic Response
The paper also examines how e-commerce platforms might respond to sellers using these algorithms. Platforms typically earn revenue from sales commissions and advertising fees. The study found that while adjusting the “reserve price” in ad auctions (the minimum bid required) might not be an effective tool for the platform, adjusting the “sales commission rate” can be. Increasing commission rates can help platforms recoup some revenue lost from lower advertising bids, while still maintaining the beneficial outcomes for sellers and consumers.
This research offers a nuanced perspective on algorithmic competition, suggesting that not all forms of algorithmic coordination are detrimental. In specific market conditions, particularly those with high consumer search costs, these AI-driven decisions can lead to outcomes that are favorable for all parties involved: consumers enjoy lower prices, sellers benefit from reduced advertising costs and increased demand, and platforms see higher overall sales volume. For more details, you can read the full research paper here.
