States and municipalities looking to govern algorithmic pricing should learn from the legal saga of the real estate software company RealPage, which is often seen as the exemplar of algorithmic foul play. RealPage has been on a roll lately. In October, a court approved the software company’s $360 million class action settlement. In September, the Southern District of New York granted a preliminary injunction to the company, which is suing the state for outlawing its business model.
RealPage’s proprietary algorithm suggests profit-maximizing rents and occupancy rates to property managers, a practice that can be both collusive and downright miserly. With several settlements under its belt, the company has now positioned itself for a likely win in New York. But in its legal maneuvering, it has managed to sidestep the million-dollar question: whether algorithms run primarily on public data can ever cause sufficient chaos to merit government limits.
The law RealPage claims is unconstitutional has been billed as a 21st-century update to New York’s 127-year-old antitrust law. Among other changes, the 2025 act bans software like RealPage that gathers data about other rental properties. While dozens of states have introduced bills to regulate algorithmic pricing, New York is especially prolific. Shortly before banning the practice for landlords, the state passed a law requiring companies that set prices using personal data to explicitly acknowledge doing so to consumers. The law survived a First Amendment challenge, but is poised to become obsolete anyway: in June, the legislature passed a bill to ban personalized pricing altogether. Last week, New York City’s council passed a similar prohibition.
New York’s concern with RealPage is that it enables collusion by coordinating pricing strategies among competitors. This coordination could be intentional, where RealPage coordinates prices without the paper trail of an explicit agreement, or unintentional, where RealPage’s suggestions unthinkingly lead competitors to stop competing. Neither is necessarily true. More market information could easily lead property managers to more effectively compete, including by lowering prices to win over rivals’ tenants. The trouble is that proving undesirable coordination is hard enough if parties use an intermediary; it is harder still when algorithms unintentionally lead firms to split the market instead of compete.
RealPage has played its cards well: as part of a DOJ settlement last year, it stopped basing recommendations on competitors’ proprietary information or local market fluctuations, and stopped training models on real-time data. In effect, the settlement steered RealPage into the realm of price consultant and further away from collusion coordinator. It sued New York’s attorney general days after the settlement.
The settlement has impoverished what could have been a bellwether case. Much of the nascent case law hinges on whether price recommendations rely on nonpublic data, but tacit collusion depends on many factors: Does the model consider hyper-local geographic effects? Are the price updates instant? Do public listings reflect the final price? How often do landlords override the recommendations?
Whether the New York law is a “reasonable fit“ for its goal to prevent runaway rents depends on its ability to stop algorithms from instantaneously and automatically price-matching local competitors. Such behavior could lead to price-fixing, as algorithms learn that undercutting price lowers revenue without stealing customers. Importantly, in many markets this could happen regardless of whether competitors use the same software or pool proprietary data. As AI becomes more entwined in commerce, the courts will play a crucial role in establishing the extent to which states and municipalities can regulate algorithmic price setting. Because New York’s law does not distinguish between nonpublic and public data, it could have been a testing ground if the facts of the case were different.
Unfortunately for legislators, it does not look good for New York. RealPage’s settlement forbids auto-updates based on its recommendations and adds enough noise and lag to the data that it can no longer effectively facilitate instantaneous local price-matching. Meanwhile, the New York law bears not on the conduct of landlords setting prices (less protected), but on the speech of RealPage offering advice (more protected). This framing has proven to be a major weakness of the law. New York must now convince the court it should be able to curb protected speech that is wholly unremarkable except for its being mediated by software. If the state has any argument at all, it is that the settlement still permits the model to train on historical proprietary information that over time could lead it to anticipate real-time fluctuations.
The phrasing of the law, which in effect bars communication, does not help the state’s case, but the judicial decision seems to imply that whether the law bans speech or conduct reduces down to whether the source of the data is nonpublic. The crucial question, whether circumstances can align such that algorithmic pricing leads to tacit collusion without pooling proprietary information, is thus brushed aside without much commentary. That is a shame, since it is almost certainly too soon to tell how agent coordination can distort markets. No state should have to tolerate independent algorithms converging on artificially high rents or vacancy rates that worsen housing shortages and drain tenants’ pocketbooks.
The implications run far beyond housing, too. A major private equity firm recently warned that personal agents could cause a bank run by optimizing personal finances away from low-interest deposits. Public information, when systematically harnessed to make economic decisions, can harm consumers in the aggregate. States deserve latitude in addressing that.
But RealPage, despised as it is, has agreed to bring its operations within the bounds of acceptable business practice and protected speech. It will win this case. The question is who will win the next one.




