Ghost Cartels: How Pricing Algorithms Quietly Kill Competition
AI-driven pricing tools can mimic cartel behavior without human collusion, raising antitrust concerns.
1The Nessie Blueprint
Amazon's Project Nessie pricing algorithm identified products where competitors would follow price hikes, generating $1B+ in excess profits according to FTC allegations. The tool was allegedly toggled off during regulatory scrutiny periods. While Amazon denies these claims and states Nessie was discontinued, it exemplifies designed algorithmic anticipation of competitor behavior—a deliberate form of pricing coordination.
2Gas Station Revelation
A 2017 German case study showed when two competing gas stations adopted automated pricing software, margins rose 38% without any communication. This 'ghost cartel' effect only occurred when both stations used algorithms—single adopters saw no margin changes. Published in 2024 in the Journal of Political Economy, this real-world evidence confirmed what simulations had predicted about algorithmic tacit collusion.
3Three Collusion Pathways
Researchers identify three algorithmic collusion models: 'Ghost' (independent systems learning truces), 'Mirror' (Amazon-style prediction of rivals), and 'Hub' (shared vendor systems like RealPage). The ghost model is most insidious—requiring no data exchange or design intent. Only hub models have been successfully challenged by regulators, leaving major enforcement gaps for other types.
4AI's Cartel Classroom
A 2020 American Economic Review study placed reinforcement-learning algorithms in pricing competition scenarios. Without communication, they consistently learned to charge above competitive levels and enforce this through retaliatory pricing—mirroring human cartel behavior. Even with varying costs or competitor counts, this pattern persisted, demonstrating AI's capacity for emergent collusion.
5Regulatory Blind Spot
Current antitrust laws require evidence of human agreement—meetings or understandings—to prove collusion. Machine-learned coordination leaves no such trail. The 2024 RealPage case highlights these challenges: despite using competitor data to recommend rents, the settlement involved no penalties or admissions, merely restricting certain data types and adding monitoring.
6Executive Illusion
Managers often misinterpret stable prices and comfortable margins as competitive victory. Algorithmic pricing breaks this intuition—the same calm market could indicate algorithms have learned non-aggression pacts. This creates invisible risks: systems might be working exactly as designed (optimizing margins) while achieving outcomes companies couldn't publicly justify.
7The Wharton Warning
In a 2020 Science paper, economists including Joseph Harrington cautioned that algorithmic pricing creates collusion backdoors. AI can learn anti-competitive rules without human intent or oversight. Harrington argues competition law must adapt to address coordination without agreement—a paradigm shift from traditional antitrust frameworks.
8Implementation Paradox
While some studies show algorithmic collusion risks, others debate how readily lab results translate to real markets. This uncertainty itself warrants proactive board oversight rather than waiting for regulatory clarity. The German gasoline study provides concrete evidence that ghost cartels aren't just theoretical—they're already occurring in some sectors.
9Compliance Time Bomb
Companies using pricing algorithms face hidden compliance risks. Unlike obvious pricing errors, sophisticated systems might produce economically optimal but legally questionable outcomes. With regulators struggling to address even clear-cut cases like RealPage, businesses must self-police algorithmic behavior to avoid future liability as laws evolve to cover emergent collusion.
Original article: Fortune — Ghost Cartels: How Pricing Algorithms Quietly Kill Competition