1 Introduction
Over the past decade, an increasing number of firms across sectors such as travel, entertainment, retail, and platform-based services, have transitioned to algorithmic pricing mechanisms. These mechanisms involve the deployment of software that automates price setting and are mainly used by firms aiming to optimise pricing decisions in response to evolving market conditions. Pricing algorithms may combine data and assumptions on factors such as prices, volumes, inventory, and customer responsiveness to forecast market demand and elasticity; they may assess the potential impact of different pricing options on business goals like revenue, profit, and market share, factoring in expected competitor responses, and then apply the optimal price in real or near real time. Thus, their use may require vast amounts of granular data.
While the introduction of pricing algorithms can bring efficiency-enhancing and procompetitive effects (see (OECD, 2023[1])), this paper focuses on their potential risks. Their widespread use has changed market dynamics, and may also raise issues in areas such as competition, consumer protection and data privacy policies, particularly with respect to its potential implications for market competition and the legality of price discrimination and collusion practices.
Canada, in its role as 2025 G7 President, asked the OECD to prepare this note to provide general background on ‘algorithmic pricing,’ which is the topic of the G7 Joint Competition Policy Makers and Enforcers Summit. The note, prepared by the OECD Secretariat, does not necessarily represent the views or positions of individual jurisdictions. With this note, the OECD aims to provide a broad picture of the current use of algorithmic pricing and the potential risks commonly associated with it, as well as an overview of the different possible strategies that could be used to tackle them. Other G7 Presidents had also asked the OECD to prepare papers on the previous summit topics. These include work started in 2022 under Germany’s presidency and continued under Japan's presidency in 2023 and Italy's presidency in 2024, when the OECD's contributions focused first, on identifying convergences and divergences between ex-ante regulations in digital markets (OECD, 2023[2]), and secondly on the key competition concerns at the heart of multi-jurisdictional efforts in digital markets, highlighting patterns that can be identified in terms of both platforms’ conduct and enforcement activities in G7 countries (OECD, 2024[3]).
The note is structured as follows. After this introduction, the second chapter synthesises the main findings from surveys, market studies, and enforcement actions on algorithmic pricing carried out in G7 jurisdictions, drawing attention to recurring concerns identified across these jurisdictions. While some of these issues lie at the intersection of consumer protection, data protection, and competition laws, this note focuses exclusively on competition-related concerns. The third chapter presents the range of measures proposed or implemented by G7 jurisdictions in response to the identified risks. It is based on publicly available information as of July 2025 and covers recommendations following market studies, public consultations on discussion papers, reform proposals and enactment of new regulations, as well as other advocacy initiatives.
The note concludes that:
• Despite variations in legal frameworks across G7 jurisdictions, there are some commonalities in how competition authorities have sought to understand pricing algorithms, and the extent to which they raise potential competition concerns.
• Many authorities have adopted similar strategies, prioritising market studies, consultations, awareness raising initiatives, and reports to better understand the competitive dynamics, risks and benefits associated with algorithmic pricing. These efforts have been followed by the publication of guidance or statements that clarify their position and reaffirm that existing competition rules apply to algorithm-driven conduct.
• It remains too early to draw definitive conclusions about the future likelihood of significant enforcement in this fast-evolving area. While authorities have identified similar potential concerns, these have not yet translated into significant enforcement activity in every jurisdiction. Only a limited number of cases — primarily under traditional theories of harm such as price-fixing, information exchange and hub-and-spoke arrangements — have emerged.
2 Competitive risks of algorithmic pricing
In recent years, competition authorities from G7 jurisdictions have engaged in various initiatives to better understand the functioning of pricing algorithms, how companies use them to compete, and their broader impact on markets and consumers. This section summarises the key findings from the most relevant surveys, market studies, as well as enforcement cases, highlighting common concerns identified across the jurisdictions.
As highlighted below, while G7 authorities recognise many efficiency-enhancing effects of using algorithms to set prices, they have identified potential risks pertaining to both unilateral and coordinated anticompetitive behaviour arising from algorithmic pricing.1 These findings go in line with the main findings documented in recent OECD work on the topic, including the 2017 paper on Algorithms and Collusion (OECD, 2017[4]) and the 2023 note on Algorithmic Competition (OECD, 2023[1]).
Collusion
The predominant focus of competition authorities’ activities related to pricing algorithms has been the relationship between algorithmic competition and collusion. While most of the concerns presented below can also be raised by the use of other types of algorithms, this section will focus on how algorithms that set or recommend prices may increase risks of collusion. As discussed below, one broader concern relates to accountability of firms for autonomous algorithmic behaviour, an issue that remains largely unresolved in some jurisdictions due to the absence of precedent.
In its 2021 study on the use of algorithms, the Japan Fair Trade Commission (JFTC) recognises that the different types of price-setting algorithms, which contain automatic updating tools, machine learning strategies to predict demand and reinforcement learning to maximise profits, may change the competitive environment. The study highlighted that the use of the algorithms increases market transparency between competitors, as well as the frequency of their interactions. Both characteristics are well-known for increasing the possibilities of reaching and sustaining a collusive outcome. Canada’s latest discussion paper on algorithmic pricing and competition also notes that algorithmic coordination is also more likely “in markets with high transparency due to better availability of data, and in markets with frequent interactions, as algorithms may enable firms to punish deviations more effectively. Coordination is also more easily sustained in markets with fewer firms and higher barriers to entry” (Competition Bureau Canada, 2025[5]).
Reports, market studies, and inquiries conducted by authorities in Canada, France, Germany, Italy and Japan, as well as Guidelines and notes issued by the European Commission and statements, and enforcement cases – addressed in detail in the following chapter – reveal some commonalities amongst G7 countries on the issues identified related to algorithmic competition and collusion.
While there are some differences in how competition authorities have classified the types of algorithm driven collusion, concerns generally fall into four main categories: (i) algorithms as facilitators of traditional price-fixing agreements, (ii) hub-and-spoke schemes, (iii) vertical agreements, and (iv) tacit collusion through autonomous learning.
Algorithms as facilitators of traditional anticompetitive price-fixing agreements
Most G7 Competition authorities have identified concerns related to the use of pricing algorithms as a tool to implement, facilitate, better conceal and/or monitor traditional price-fixing agreements, usually including prior contact between the cartelists.
• Implement/Facilitate: once companies have explicitly colluded, pricing algorithms can serve as a tool to implement the agreements by embedding explicit assumptions or rules within the algorithm, such as not to undercut competitors’ prices or block poaching each other’s customers, among others (Autorité de la Concurrence and Bundeskartellamt, 2019[6]). Similarly, the increased availability of pricing data from competitors and the use of automated pricing systems may facilitate explicit coordination. They reduce the chance of errors or accidental deviation, making explicit collusion between firms more stable. Moreover, the algorithm could automatically adjust prices according to market changes, eliminating the need for cartel members to renegotiate the terms of their agreement (JFTC, 2021[7]). Another feature of algorithms that facilitates price-fixing agreements is that they can increase the stability of such agreements by reducing “agency slack”. With their use, there can be less scope for individuals within a company to deviate from the price fixing agreement e.g. by undercutting the collusive price (CMA, 2018[8]).
• Conceal: algorithms could serve as a tool to hide the anticompetitive behaviour. In their joint study, the French and German competition authorities found that algorithms could be used to artificially generate occasional price heterogeneity or instability (e.g. when there is low or no demand), while maintaining a collusive behaviour in general. This makes it more difficult for authorities to detect the cartel. They could also be used to conceal communication between competitors, by allowing encrypted messaging among other strategies (Autorité de la Concurrence and Bundeskartellamt, 2019[6]).
• Monitor: pricing algorithms may also monitor competitors’ price, even by accessing real-time data. This makes it easier and faster to detect and respond to deviations, reducing incentives for firms to deviate in the first place (CMA, 2021[9]).
In the UK, there have been two relevant cases illustrating these concerns. The first one is the 2016 decision on the Trod/GBE cartel on the online sales of posters and frames.2 The Competition and Markets Authority (CMA) discovered that the companies had agreed that they would not undercut each other on prices for certain posters and frames sold on the Amazon UK Marketplace, and that they used a pricing software to implement this agreement. The second case was an agreement to allocate customers investigated by the energy regulator Ofgem. The regulator found that two energy suppliers had an agreement preventing them from actively targeting each other’s customers and used a common algorithm to block such behaviour.3
In a similar case to the CMA’s decision on Trod/GBE, in the United States, sellers of posters on the Amazon Marketplace were found to have entered into a price-fixing agreement, that they implemented through pricing algorithms. Specifically, the cartelists “wrote computer code that instructed algorithm-based software to set prices in conformity with this agreement” (US DoJ, 2015[10]).
In these scenarios, algorithms act simply as the tool to implement a traditional anti-competitive agreement. For competition authorities, the use of pricing algorithms (shared algorithms or individual ones) to agree on prices, should be treated as any other price-fixing agreement, which usually is considered a by object / per se restriction of competition. While this type of collusion would normally not require adjustments in the analytical framework compared to any other price-fixing agreement, the use of the algorithm may require authorities to consider their evidence-gathering tools and practices (for example, if as a result of the use of the pricing algorithm there are fewer direct exchanges between the involved competitors that can be detected through traditional evidence-gathering practices). On the other hand, the ability to analyse the functioning and role of an algorithm (e.g. by auditing) can provide new type of evidence. The European Commission has highlighted this relevance stating their view that, “although in this scenario it is also possible to rely directly on more traditional concepts in the legal assessment, the use of algorithms could certainly have a relevant impact on the authorities’ ability to prove the infringement through the usual investigative tools” (OECD, 2023[11]).4
Hub-and-spoke schemes
Third-party involvement in collusive practices have generally generated two further concerns to most G7 competition authorities. One of them results from pricing recommendations or price setting by common intermediaries and the second from the use of the intermediary to exchange confidential information relevant to reach the anticompetitive agreement.
If multiple competitors use the same pricing algorithm, certain aspects of its use may generate competitors to react in a similar way to external events, better predict their competitors’ responses to price changes and even introduce a common pricing strategy. In cases in which the third party optimise prices and recommendations for each user independently, the use of the algorithm is not necessarily restrictive of competition.
In their joint study from 2019, France and Germany identified two types of alignment of algorithmic decision-making when companies rely on the same algorithm to make their strategic decisions (Autorité de la Concurrence and Bundeskartellamt, 2019[6]). For the authorities, both types of alignment may coincide.
In their joint study from 2019, France and Germany identified two types of alignment of algorithmic decision-making when companies rely on the same algorithm to make their strategic decisions (Autorité de la Concurrence and Bundeskartellamt, 2019[6]). For the authorities, both types of alignment may coincide.
1. Alignment at code level: when the third party provides algorithms with a shared purpose and a similar or related methodology. This can go from completely identical algorithms to including commonalities while keeping some degree of individualisation for the respective customer. This includes the software merely suggesting prices but not providing automated price setting. Alignment at code level may also involve full delegation of the strategic decision to the common third party.
2. Alignment at data level: this refers to the extent to which the algorithm facilitates information exchange among competitors. This can go from the algorithm facilitating active information sharing to the algorithm using a shared pool of data to pursue a common goal (i.e. maximising joint profits). Even when prices are calculated individually, the use of competitors’ data to train the algorithm may be of concern.
In addition to the degree of algorithmic alignment, other relevant factors identified in the report include the market shares of the competitors sharing the algorithm and the way in which the data is used, stored and shared, which would determine whether their conduct can be understood as a hub-and-spoke scheme.
Furthermore, the report underscores a distinction between two different scenarios. One that involves competitors knowingly delegating certain strategic decisions—such as pricing—to a third party, which then makes those decisions using a single algorithm. Another, when the third party provides a coordinated algorithmic solution without explicitly informing the users of the existence of the common strategy. The key difference according to the joint study is that awareness of the conduct may differentiate between said conduct been lawful or unlawful.
Following European case law and practice, the report establishes that hub-and-spoke conspiracies where information exchange takes place through a common third party may be anti-competitive. if competitors are aware – or could reasonably have foreseen – that they are relying on the same provider and using the same or coordinated algorithm. Moreover, collusion via a third party and the role of a facilitator in a cartel have been explicitly recognised by case law of the European Court of Justice and hub-and-spoke schemes have been investigated and sanctioned, although in cases not involving the use of an algorithm (OECD, 2023[11]).
The EC’s Guidelines on the applicability of Article 101 TFEU to horizontal cooperation agreements state that when competitors subscribe to the same third-party pricing tool, and the tool uses commercially sensitive information from competitors, this may result in an unlawful information exchange (European Commission, 2023[12]). Indirect contact amongst competitors was key in the E-Turas case. The Supreme Administrative Court of Lithuania requested a preliminary ruling of the EU General Court concerning an interpretation of Article 101 TFEU to conclude whether using a common information system that recommends certain pricing practices can be considered to infringe Article 101 TFEU.5 The Court concluded that if competitors were aware of the practice, they could be presumed to have participated in the concerted practice, “unless they publicly distanced themselves from that practice, reported it to the administrative authorities or adduce other evidence to rebut that presumption, such as evidence of the systematic application of a discount exceeding the cap in question”.6
The JFTC identified similar distinctions and enforcement implications under the Japanese Antimonopoly Act. If firms using the same algorithm share a common understanding that it will coordinate prices—despite no direct communication between the competitors—this may constitute a violation of Japan’s Antimonopoly Act. In contrast, if an algorithm provider sets their prices, without the firms’ knowledge, it may be considered private monopolisation by the provider, if it provides a great majority of the price-setting algorithms used in a specific market. Similar concerns have been identified in the United Kingdom. The CMA noted in its 2018 paper that hub-and-spoke scenarios are likely to present the most immediate risk for competition when algorithms are involved, noting however that the mere use of a shared algorithm does not, by itself, constitute a violation of competition law. For the conduct to be deemed anticompetitive, there must still be some intention among the companies to suppress competition (CMA, 2018[8]).
The CMA’s Guidance on the application of the Chapter I prohibition in the Competition Act 1998 to horizontal agreements highlight hub-and-spoke agreements as a type of indirect information exchanges between companies. The Guidelines describe how shared algorithms can be used to agree on price levels or margins and specifies that while “using publicly available data to feed algorithmic software is legal, the aggregation of competitively sensitive information into a pricing tool offered by a single IT company to which various competitors have access could amount to horizontal collusion.” (CMA, 2023[13])
In the United States, courts have started to review cases related to price recommendations by algorithms and the use of common algorithms to make strategic decisions. The US Department of Justice (DoJ) Antitrust Division has filed statements of interest in some of these cases to address relevant legal issues.
Canada’s consultation on Artificial Intelligence and Competition revealed hub-and-spoke agreements as a recurrent concern, where a single algorithm acts as a hub and multiple competitors use it to set prices (Competition Bureau Canada, 2025[14]). One of the main concerns on the common use of algorithms to set prices also appears to be linked to the sharing of competitively sensitive information from different competitors, to train the software and set or recommend prices. In 2024, the Bureau opened an investigation to determine whether the pricing guidance offered by a data service provider to retail energy companies has an adverse effect on competition between gas stations. As part of its investigation, the Bureau obtained a court order to gather information on the company’s pricing services) and the way it operates to provide pricing guidance to gas station operations. In 2025 the Bureau also confirmed an ongoing investigation on possible uses of algorithms in real estate markets to set rents.
In summary, most G7 competition authorities appear to be broadly aligned in raising concerns about the potential for companies to engage in hub-and-spoke agreements, where the algorithm provider acts as the hub and competing firms as the spokes, collectively coordinating pricing. Key elements of this theory of harm include: the algorithm’s use of commercially sensitive data from multiple competitors to set prices and the competitors’ awareness or intent to reduce competition, even in the absence of direct communication or full alignment with the pricing strategy.
Click to read more