A routine apartment hunt in Seattle has turned into the starting point for three University of Washington researchers exploring algorithmic collusion and the unseen coordination of AI-driven markets.
In brief:
- The research shows how recommendations shoppers see on platforms can intensify or curtail algorithmic collusion, a phenomenon where interacting pricing algorithms drive prices artificially high
- When platforms prioritize revenue, they help sellers maintain higher prices by keeping expensive products visible to shoppers
- Counterintuitively, showing consumers more product choices can sometimes hurt them when pricing algorithms are involved
- Findings suggest regulators should look beyond pricing algorithms alone and consider the role of recommendation systems on platforms like Amazon in enabling elevated prices
- The study contributes to a growing field examining how AI affects the economy, but it does something most research doesn’t: It looks at how different AI systems interact, rather than studying each one separately
A Seattle Apartment Search Raises Bigger Questions
When Xingchen (Cedric) Xu was searching for apartments in Seattle in 2023, he noticed that rental prices were consistently high. He searched online and came across news about algorithmic collusion in the rental market. He had a thought: If algorithms were coordinating prices in housing, what about online shopping sites like Amazon? Unlike rental platforms, Amazon has not only pricing algorithms but also recommendation systems that decide what shoppers see. Xu, a doctoral student at the Foster School of Business, had a call with Stephanie Lee, a Foster faculty member, on moving day, somewhere between his old place and his new one. That conversation would eventually launch award-winning research examining how AI systems interact to shape markets in ways that aren’t always visible to consumers or regulators.
The resulting study, titled Algorithmic Collusion or Competition: the Role of Platforms’ Recommender Systems, was guided by Xu’s advisor, Yong Tan. The paper recently won the 2025 INFORMS ISS Cluster Best Paper Award.
Critically, the study reveals that when pricing algorithms and recommendation systems work together, they can fundamentally reshape market competition, sometimes in ways that harm consumers.
“This work is one of the examples that examine how AI to AI interactions determine the market structure and competitive outcome,” says Tan, Michael G. Foster Endowed Professor of Information Systems. “It can be expanded to many settings of competitive algorithmic decision making, human-AI interactions, etc., to produce insights into market design and platform regulation.”
The hidden power of recommender systems
Regulators and researchers have increasingly focused on algorithmic collusion, which is when pricing algorithms independently learn to maintain artificially high prices without any direct communication between sellers. In 2024, RealPage faced lawsuits for using pricing algorithms that allegedly harmed millions of renters. Congress proposed the “Preventing Algorithmic Collusion Act” to address these concerns.
But there’s been a critical blind spot: the role of platform recommendation systems.
“What was lacking in the previous analysis was the presence of recommendation systems,” says Lee, assistant professor of information systems. “We take a look at how, depending on these recommendation systems’ configurations, it intensifies the collusion or mitigates the collusion.”
The team developed a simulation that mirrors the functioning of real online markets. They created an economic model to capture how consumers make decisions when faced with different prices and product recommendations. They then developed virtual pricing algorithms for sellers and various types of recommendation systems for the platform. During the revision process, to ensure their model reflected real consumer behavior, they analyzed an open-source hotel booking dataset from Expedia to calibrate the model parameters. They then watched what happened as these systems interacted over millions of simulated transactions.
Platform goals shape what shoppers pay
The research reveals a striking finding: The platform’s intended objective fundamentally alters the outcome of prices.
“If you want to regulate these shopping platforms, you should not just focus on the pricing algorithm teams,” Xu says. “Maybe you should also focus on the recommender system teams there.”
When a recommendation system aims to maximize revenue, which is the typical goal for platforms that take a cut of each sale, the platform tends to show higher-priced products to shoppers because these products generate a higher commission. Over time, the sellers’ pricing algorithms learn that they can charge more and still get visibility. Prices gradually creep up and stabilize at these higher levels.
The platform’s recommendations ultimately reward higher-priced products by keeping them visible, thereby reinforcing elevated pricing. This dynamic leads to what researchers call supracompetitive pricing—prices higher than what normal market competition would produce.
But when the platform instead tries to maximize shopper satisfaction, the opposite occurs. The recommendation system hides expensive products and highlights better deals. The pricing algorithms quickly learn that raising prices means losing visibility and sales. Competition intensifies and prices fall.
Why more choices can mean worse outcomes
Perhaps the study’s most surprising discovery challenges the basic assumption that giving shoppers more options is always better.
Conventional wisdom says that when consumers see more products, they can skip what they don’t like and pick what they do. Traditional economics supports this intuition. But when AI pricing algorithms are involved, this logic breaks down.
The researchers found that when platforms try to maximize shopper satisfaction, showing more products can actually make shoppers worse off. It sounds paradoxical, but two forces compete here: more options help people find what they like, but showing more products reduces the platform’s ability to pressure sellers on price. High-priced items remain visible, and some customers still purchase them; as a result, pricing algorithms learn that they can maintain elevated prices.
This effect becomes stronger based on what researchers call “product differentiation”—the degree to which products vary in features, quality, or other attributes beyond just price.
Consider hotels: if they’re mostly identical except for price, shoppers will simply pick the cheapest one. But when hotels vary widely in style, location, and amenities, some shoppers will pay more for their preferred option even when shown multiple choices. Pricing algorithms pick up on this and hold prices higher.
What this means for regulators, platforms, and businesses
The research provides practical guidance for various stakeholders in online markets.
For regulators, the findings suggest that focusing solely on pricing algorithms overlooks a critical piece of the puzzle.
“If you want to regulate these shopping platforms, you should not just focus on the pricing algorithm teams,” Xu says. “Maybe you should also focus on the recommender system teams there.”
Regulators may want to require platforms to disclose how their recommendation systems are designed, including whether they optimize for revenue maximization, customer satisfaction, or other goals.
For platforms like Amazon, research indicates that they should consider how sellers will respond when designing recommendation systems.
“In the end, they want to maximize their profits,” Xu explains. “But in certain cases, they also consider consumer welfare because they want to focus on their long-term benefits. Let’s say we want to keep the consumers within the platform, not going to other platforms.”
The work suggests platforms may need to track not just customer behavior and recommendation performance, but also how pricing algorithms are responding.
For sellers using pricing algorithms, the takeaway is that these tools don’t work in a vacuum. Product design choices—how differentiated your product is from competitors—can influence how the platform’s recommendations treat you, which in turn affects how your pricing algorithm performs. Sellers need to think about pricing and product strategy together, not separately.
A new way of studying AI’s impact
The study contributes to a growing field examining how AI affects the economy, but it does something most research doesn’t: It looks at how different AI systems interact, rather than studying each one separately.
“We are looking at how the recommender systems and the pricing algorithms can interact with each other,” Xu explains. “That’s the most interesting part about the interactions between different kinds of AI.”
When Tan presented the work at UC Davis and the University of Minnesota, the response was enthusiastic.
“They found the study novel, relevant, and interesting,” Tan reports. “They were especially impressed with the granularity in our models, which capture the underlying mechanisms.”
The award from INFORMS, the leading professional association for operations research and analytics, recognizes both the originality of the approach and its practical significance.
AlgoritHmic collusion and apartment hunting
For Xu, the research journey that began with apartment hunting in Seattle has revealed something fundamental about how modern markets operate. As AI becomes increasingly prevalent in business, understanding how different AI systems interact becomes essential.
The researchers emphasize that this work is just beginning to reveal the aggregate economic effects of multiple AI systems interacting—essential work as AI becomes more prevalent in business and society.
The message is clear: In online marketplaces, pricing algorithms and recommendation systems interact in ways that fundamentally reshape competition and what consumers pay. Understanding these interactions will be critical for ensuring fair markets in the age of AI.
Read the research here.
Xingchen Xu is a Ph.D. student at the Foster School of Business. In this research, he examines algorithmic collusion and how pricing algorithms and recommendation systems interact to influence market competition and prices.
Stephanie Lee is Assistant Professor of Information Systems at the Foster School of Business.
Yong Tan is the Michael G. Foster Endowed Professor of Information Systems and Chair of the Information Systems and Operations Management at the Foster School of Business.