Pricing algorithms can be fantastic tools for market intelligence, but they are now firmly on the radar of the Commerce Commission. Business leaders are aware that there are legal risks. Less obvious is how competition law will ultimately respond once this issue reaches the courts. The most active enforcement right now is occurring in the United States. The cases so far do more to demonstrate the many shades of grey involved in this area than to provide clear rules or advice for businesses.
In this Insight, we look deeper into the US decisions and draw out some threads to provide guidance on how the use of pricing algorithms might be approached under the Commerce Act 1986 and what risks businesses need to be aware of.
We have four key messages for New Zealand businesses:
independently choosing to use the same pricing software as your competitors is unlikely, by itself, to violate competition law; but
be cautious about delegating a significant portion of pricing decisions to software commonly used in the industry;
be cautious about using software that: (i) includes commercially sensitive non-public competitor data as an input; or (ii) involves an AI model trained on such data; and
know your pricing software and how it operates, including what information is collected and how that information is used, especially whether it is shared or used to train an AI model.
The degree of legal risk will depend on the nature of the algorithm (which can range from rudimentary spreadsheets to sophisticated AI models), pricing decisions, business and industry involved. Seeking advice is recommended to understand how those risks apply to your business.
What precisely might regulators be concerned about?
Using better data to make sharper commercial decisions is not anticompetitive. Rather, it may promote competition by enabling businesses to price more competitively. The legal risk arises where pricing algorithms become a mechanism for coordination. That can occur in two ways when competitors use the same pricing algorithm:
Hub-and-spoke cartel – where multiple competitors effectively delegate a significant portion of their pricing decisions to a commonly used algorithm, there is a risk that the algorithm will reduce independent pricing decisions with the effect of coordinating prices, including through tacit collusion.
Sensitive information-exchange – where a pricing algorithm makes recommendations based on commercially sensitive non-public competitor data, there is a risk that the algorithm may recommend less competitive prices.
These two concerns have been the driving features of early enforcement action in the US. While the US courts have yet to distinguish clearly between these two factors, the US Department of Justice has been pursuing each as a standalone ground for liability and has been aggressively advocating for that position through amicus briefs in private class actions.
What have the US courts said on these issues?
While the high-profile claim involving RealPage Inc, a revenue management software firm operating in the residential rental market, receives the most attention, it is far from the only pricing algorithm case.
The RealPage case is an extreme example, with several hallmarks of conventional price fixing. One judge described RealPage as having created a "robust multi-layered monitoring system" involving the assignment of "pricing advisors or revenue managers, employed by RealPage, who closely monitor clients' conformance with RealPage's pricing recommendations"[1]. RealPage also allegedly organised regular meetings of its users. Other US cases paint a more balanced and complex picture.
Initial decisions have been made in several proceedings, but no case has yet reached trial. The outcomes have been mixed, with a variety of views expressed. We provide a brief snapshot of those decisions at the end of this Insight. We are also following active US litigation over pricing algorithms in the construction equipment rental and mortgage markets.
How do these concerns map onto the Commerce Act?
When might it be cartel conduct?
For New Zealand businesses, the most obvious risk is price fixing under section 30 of the Commerce Act. The Commission has already flagged (in its guidance published in July 2025) the possibility of viewing common use of a pricing algorithm as a hub-and-spoke cartel; that is, a horizontal agreement between competitors that is facilitated through a series of vertical agreements with the pricing software provider.
The danger of a hub-and-spoke cartel is that businesses may be investigated even where they have never spoken directly with their competitors about use of the pricing algorithm. That said, the mere independent use of common software should not be enough. There will need to be evidence from which the Commission, or a court, can infer a contract, arrangement or understanding that contains a provision with the purpose, effect or likely effect of price fixing (or restricting output or market allocating). The threshold for establishing an understanding is lower than for a contract and can be established where there is a meeting of the minds involving a mutual expectation as to future conduct.
The following conduct might be considered evidence of a price fixing agreement:
sharing of commercially sensitive non-public data (which would otherwise be counter to business self-interest);
correspondence with the software provider regarding competitors;
correspondence with competitors regarding which software is used;
unrestrained use of "autopilot" pricing;
high acceptance rates of pricing recommendations without evidence of independent commercial judgment being applied;
structures designed to minimise departures from pricing recommendations, such as targets, incentives or monitoring of staff;
internal documents suggesting that the pricing algorithm was adopted to avoid undercutting or to raise prices; or
marketing by the software provider promising to achieve market discipline or higher prices.
By contrast, genuine independent pricing discretion and routine departures from pricing recommendations would mitigate risk. There are two dimensions to consider:
First, how robust is the commercial judgment? Simply interposing a human to review a pricing recommendation may not suffice if they habitually adopt the recommendations. In some circumstances, evidence of genuine commercial consideration may be expected.
Second, what portion of pricing decisions are genuinely independent? If a significant majority of pricing decisions are derived from a common algorithm without independent judgment, then some human oversight will do less to eliminate legal risk.
The factor attracting the most focus in US cases has been the sharing of commercially sensitive non-public data. Data that is recent, specific, directly relevant to pricing, and ordinarily inaccessible to competitors is most at risk of attracting scrutiny. Historic or abstracted data raises fewer concerns. The riskiest scenario is where shared data is used as a direct input into the pricing algorithm.
Businesses should be aware, however, that pricing algorithms may attract scrutiny for price fixing even if the algorithm uses no commercially sensitive non-public competitor data. Our view, however, is that anticompetitive coordination would be much more difficult to prove in this scenario and that there are good reasons independent use of pricing algorithms should not be treated as price fixing (as the US cases tend to recognise).
When might it substantially lessen competition?
Separately, where a pricing algorithm does use commercially sensitive non-public competitor data, there may additionally be potential for liability under section 27 of the Commerce Act where there is evidence that the information-exchange has the purpose, effect or likely effect of substantially lessening competition. In this context, as above, it is the sharing of commercially sensitive non-public data that risks being classified as an understanding.
The starting point for the Commission and courts is likely to be the traditional theory of harm found in ordinary information exchange cases; namely, whether "exchange of information between competitors … reduces or removes the degree of uncertainty as to the operation of the market in question".[2]
The answer to that question would depend on the nature of the data shared and the uses to which it is put, as described above. The nature of the market will also be relevant. Information transparency can be pro-competitive, as better data may support sharper pricing, lower costs, improved inventory management and better customer matching. But in other cases, removing the uncertainty from a market that has historically made tacit coordination unstable may drive higher prices and be considered anticompetitive. That risk is likely to be greatest in concentrated markets, markets with relatively homogeneous products, or markets where prices can be changed quickly in response to competitor behaviour.
How much uncertainty is there about this new legal risk?
There remains a significant amount of uncertainty in this area, including overseas. No US court has yet examined evidence about pricing algorithms. That is important because these are complex technological tools, where the untutored assumptions of courts (based on the unproven allegations of litigants) will often be wrong.
Duffy v Yardi Systems Inc [3] is a prime example. A federal court allowed a class action to proceed on the assumption (alleged by plaintiffs) that the pricing algorithm used commercially sensitive non-public competitor data in pricing decisions. In a parallel state court proceeding, where Yardi had the opportunity to give evidence about how its algorithm works, it transpired that the claim was false. Non-public data collected from users was isolated from competitors and only used in tailoring pricing recommendations for each user. The state court dismissed the claim as a result.
It would be dangerous to assume that all pricing algorithms are the same and will be treated alike. Understanding how your pricing algorithm operates is critical. Make sure you are familiar with what information is being collected by the algorithm, how that information is being used and whether it is being shared with other users or used to train an AI model.
There is also an open question about how the use of commercially sensitive non-public data for the training of AI models will be treated. Notably, the RealPage settlement extended to AI model training but the court in Gibson v Cendyn Group LLC [4], a consumer class action against Las Vegas casino hotels using the same pricing software, appears to have been unconcerned about suggestions that non-public data may have been used in AI model training.
AI is a rapidly developing field, so where might future risks arise?
The key development we are following is the emerging use of large language model (LLM) agents as the architecture for pricing algorithms. Pricing algorithms using AI typically rely on a specialised process called Q-learning, which is a form of machine learning that essentially uses trial and error to optimise pricing. LLMs, by contrast, are the general-purpose kind of AI that underpin the generative AI models that consumers are familiar with. Agents are an autonomous function that uses an LLM to execute a series of tasks without ongoing user involvement and can call on the use of "tools" that allow the agent to interact with the outside world. Pricing algorithms that use LLM agents may raise new competition law risks, such as communicating with competitors or obtaining non-public data without the user's knowledge. For example, the recent Hugging Face incident demonstrates both the capability and propensity of agents to communicate and collaborate with other agents. These concerns remain on the frontier of competition law for now and have not featured in enforcement action to date.
Is there also a regulatory risk?
In September 2025, Cabinet agreed to amend the Commerce Act to ensure its prohibitions apply to conduct carried out using artificial intelligence or algorithmic tools. That clarificatory reform was not carried through to the Commerce (Promoting Competition and Other Matters) Amendment Bill (as we explained in a previous Insight). By contrast, several US states have gone further and adopted broad prohibitions on some pricing algorithms. For example:
California has banned the use of "common pricing algorithms", which includes any software that uses competitor data (non-public or public) to recommend prices.
Meanwhile, New York has banned pricing algorithms in the residential rental market that perform "a coordinating function", which includes training algorithms on publicly available historic competitor data.
These bans have wider scope than existing enforcement action in the US. We are monitoring for any renewed suggestions of similar regulatory efforts in New Zealand.
