1. Introduction

Predictive antitrust frameworks are competition-law methods that use economic modelling, statistical analysis, artificial intelligence (AI), machine learning, and market intelligence to identify, assess, and prevent potential anticompetitive conduct before it causes substantial harm to competition.

Traditional antitrust enforcement often investigates conduct after it has occurred. Predictive antitrust, by contrast, seeks to anticipate developments such as emerging monopolies, exclusionary pricing, coordinated algorithmic behaviour, anticompetitive mergers, discriminatory platform access, and the foreclosure of innovative competitors.

The objective is not to punish firms merely because an algorithm predicts that they may become dominant. Rather, predictive analysis helps competition authorities identify credible risks, investigate them using legally admissible evidence, and take proportionate action under established legal standards.

Objectives of predictive antitrust

Early detection: Identify emerging market-power concerns before competitive conditions deteriorate.

Merger screening: Predict whether a proposed acquisition may eliminate a future competitor.

Algorithmic monitoring: Detect suspicious pricing patterns, parallel conduct, or potential algorithmic coordination.

Market foreclosure analysis: Anticipate whether dominant firms may restrict access to platforms, data, infrastructure, or distribution channels.

Dynamic competition assessment: Evaluate effects on innovation, entry, consumer choice, and future market development.

Evidence-based enforcement: Direct investigative resources towards markets and conduct presenting credible competition risks.

2. Concept and theoretical foundations

Predictive antitrust draws on several interconnected disciplines.

A. Industrial organisation economics

Analyses market concentration, entry barriers, pricing incentives, economies of scale, network effects, and the relationship between market structure and competitive behaviour.

B. Artificial intelligence and machine learning

Identifies unusual pricing patterns, predicts changes in market concentration, and prioritises transactions or conduct requiring further investigation.

C. Causal inference and counterfactual analysis

Distinguishes genuine competitive harm from correlations caused by demand shocks, input-cost changes, regulation, or other market conditions.

D. Network and ecosystem analysis

Examines how control over data, operating systems, payment infrastructure, app stores, cloud services, and digital distribution may reinforce market power.

E. Legal and institutional safeguards

Ensures that predictions remain investigative indicators rather than substitutes for proof, procedural fairness, and the applicable legal test.

A central distinction is between predicting market outcomes and proving an infringement. A model may indicate an elevated risk of exclusionary conduct, but a competition authority must still establish the relevant facts, legal elements, and appropriate causal connection.

3. Legal framework

Predictive antitrust is not generally a standalone offence. It operates through existing competition-law provisions and, in some jurisdictions, specialised digital-market legislation.

The precise legal threshold differs across jurisdictions. Merger control is inherently forward-looking: authorities assess likely future competitive effects. Conduct enforcement, however, ordinarily requires proof of conduct satisfying the applicable statutory test, rather than a prediction of possible future wrongdoing alone.

4. Predictive antitrust frameworks: operational model

A well-designed framework should connect predictive analytics to legal investigation without allowing automated systems to make final infringement decisions.

Stage 1 — Market data collection

Prices, output, market shares, merger documents, switching patterns, entry barriers and platform-access conditions

Stage 2 — Predictive risk screening

Statistical models, anomaly detection, merger simulations and network analysis

Stage 3 — Economic and legal assessment

Market definition, theories of harm, counterfactuals, efficiencies and alternative explanations

Stage 4 — Human-led investigation

Document requests, witness evidence, source-code review where appropriate, and verification of economic findings

Stage 5 — Legally justified outcome

No action, continued monitoring, commitments, merger remedies, or enforcement proceedings according to applicable law

Principal predictive tools

The U.S. Department of Justice's 2023 Merger Guidelines, for example, expressly discuss how pricing algorithms, information exchange and rapid competitive responses can affect the assessment of coordination risks. Predictive techniques can strengthen that analysis, but must remain connected to the legal and economic evidence.

5. Detailed case laws — at least six

The following decisions illustrate the legal principles underlying predictive antitrust, including forward-looking merger assessment, emerging competition, coordinated conduct, electronic systems, and digital-market foreclosure. Not every case involved AI or predictive analytics directly; their relevance lies in the doctrines that predictive frameworks can help authorities investigate.

Case 1: United States v. Microsoft Corp. (2001)

Jurisdiction: United States Court: U.S. Court of Appeals for the District of Columbia Circuit

Facts: Microsoft was accused of using its dominant position in operating systems to restrict competition from Netscape's browser and Java technologies.

Legal principle: A dominant firm's conduct may unlawfully maintain monopoly power when it impairs competitive threats and protects its position from emerging rivals.

Relevance to predictive antitrust: The case demonstrates why competition analysis must consider the future competitive significance of nascent technologies. A predictive framework could identify strategies that make entry more difficult, such as restrictive defaults, interoperability barriers, or exclusionary distribution arrangements.

Significance: Competition authorities should not assess digital markets solely by looking at current market shares. They should also investigate whether present conduct prevents a new technology from becoming a meaningful competitive constraint.

Case 2: United States v. Apple Inc. (2013)

Jurisdiction: United States Court: U.S. District Court for the Southern District of New York

Facts: The litigation concerned arrangements involving Apple and publishers in the electronic-book market. The court found that Apple had orchestrated a conspiracy to raise e-book prices.

Legal principle: An intermediary may incur antitrust liability where its conduct facilitates or participates in an unlawful agreement among competitors.

Relevance to predictive antitrust: Digital intermediaries and common service providers can create information channels through which competitors coordinate. Predictive monitoring can flag common pricing changes, shared information flows, or suspicious patterns for further investigation.

Significance: A pricing pattern alone does not establish conspiracy. Investigators must distinguish independent responses to market conditions from coordination supported by evidence of an agreement or other legally sufficient concerted conduct.

Case 3: Airtours plc v Commission (2002)

Jurisdiction: European Union Court: Court of First Instance, now the General Court

Facts: The European Commission prohibited Airtours' proposed acquisition of First Choice, expressing concern that the transaction would facilitate collective dominance in the package-holiday market.

Legal principle: An assessment of coordinated effects requires a sufficiently substantiated explanation of how coordination could emerge and remain sustainable in the relevant market.

Relevance to predictive antitrust: The case illustrates the value of scenario-based analysis. Authorities may need to assess whether firms can monitor one another's conduct, whether deviations can be detected, and whether competitors have incentives to retaliate.

Significance: Predictions of coordination must be supported by the market's actual characteristics. A model cannot replace a rigorous explanation of how coordination would operate and why it would be sustainable.

Case 4: Wood Pulp — Ahlström Osakeyhtiö and Others v Commission (1993)

Jurisdiction: European Union Court: Court of Justice of the European Communities

Facts: The Commission investigated parallel pricing behaviour in the pulp market and concluded that producers had engaged in concerted practices.

Legal principle: Parallel conduct does not, by itself, establish concertation where the observed behaviour can be explained by the structure and conditions of the market.

Relevance to predictive antitrust: Machine-learning systems may detect simultaneous price increases across competing firms. Such findings can identify useful investigative leads, but authorities must examine alternative explanations such as common input costs, demand changes, and transparent market conditions.

Significance: Predictive enforcement must guard against false positives. Similar conduct is a signal to investigate, not automatic proof of collusion.

Case 5: T-Mobile Netherlands BV and Others (2009)

Jurisdiction: European Union Court: Court of Justice of the European Communities

Facts: Mobile telecommunications operators participated in a meeting at which competitively sensitive information concerning dealer remuneration was discussed.

Legal principle: An exchange of strategic information may restrict competition where it reduces uncertainty about competitors' future market conduct and satisfies the applicable requirements of Article 101 TFEU.

Relevance to predictive antitrust: Information-exchange monitoring can help identify conduct that reduces strategic uncertainty. In digital markets, relevant data may be transmitted through shared analytics services, industry platforms, or third-party pricing tools.

Significance: Authorities must assess the information exchanged, the market context, and the legal requirements for establishing an infringement. Automated detection of communication or parallel pricing is not sufficient on its own.

Case 6: Eturas UAB and Others (2016)

Jurisdiction: European Union Court: Court of Justice of the European Union

Facts: A common online travel-booking system was used by travel agencies. A system message informed participating agencies of a restriction on discounts, and technical measures were implemented within the system.

Legal principle: Conduct facilitated through an electronic platform may constitute a concerted practice, but the existence and attribution of participation must be assessed under the applicable evidentiary rules.

Relevance to predictive antitrust: The case is particularly relevant to automated commercial systems. A platform's configuration, shared system messages, or algorithmic restrictions may affect the competitive decisions of multiple businesses.

Significance: Predictive systems can help identify suspicious platform-wide changes, but they must distinguish technical design choices from evidence that firms knowingly participated in unlawful coordination.

Case 7: Google Shopping — Google and Alphabet v Commission (2024)

Jurisdiction: European Union Court: Court of Justice of the European Union

Facts: Google was found to have abused its dominant position by favouring its own comparison-shopping service in its general search results while positioning and displaying competing services less favourably.

Legal principle: Self-preferencing by a dominant undertaking may constitute an abuse under Article 102 TFEU when the conduct satisfies the applicable legal test.

Relevance to predictive antitrust: Ranking algorithms influence which businesses receive visibility, traffic and commercial opportunities. Predictive monitoring can test whether ranking changes systematically disadvantage competing services, particularly where a platform controls access to users.

Significance: Ranking outcomes should be assessed alongside the platform's market position, the design and effects of its conduct, and possible objective justifications. A ranking difference alone does not establish abuse.

Case 8: Intel Corp. v Commission (2024)

Jurisdiction: European Union Court: Court of Justice of the European Union

Facts: The litigation concerned Intel's conditional rebates to computer manufacturers and the Commission's assessment of their exclusionary effects.

Legal principle: Where the relevant legal framework requires an assessment of the exclusionary capability of rebate practices, the authority must properly evaluate the relevant evidence and arguments concerning that capability.

Relevance to predictive antitrust: Predictive models can examine customer dependence, switching costs, rival costs, rebate conditions and the likelihood that competitors can reach sufficient scale. Such analysis may help identify whether a pricing strategy could foreclose equally efficient rivals.

Significance: The use of economic models must be accompanied by a legally sound assessment of the conduct and its competitive effects. The Court's 2024 judgment is especially important in ensuring that economic evidence is evaluated properly.

Case 9: United States v. Topkins (2015)

Jurisdiction: United States Court: U.S. District Court for the Northern District of California

Facts: An online seller participated in a conspiracy to fix prices for posters sold through Amazon Marketplace, using pricing algorithms to implement the arrangement.

Legal principle: The use of an algorithm does not shield an explicit price-fixing agreement from antitrust liability.

Relevance to predictive antitrust: This case provides a direct example of algorithm-assisted collusion. Monitoring may identify suspicious patterns in online prices and prompt examination of communications, software instructions, data-sharing arrangements, and pricing rules.

Significance: The key issue was the underlying agreement, not the mere fact that prices were set by software. This distinction remains central when authorities investigate modern AI-driven pricing systems.

6. Application of predictive antitrust to emerging markets

A. AI-driven pricing and algorithmic coordination

Pricing algorithms can monitor competitors, react to demand, and change prices repeatedly within short periods. Predictive antitrust can examine whether those systems increase the risk of coordinated outcomes or implement an existing unlawful agreement.

Three situations should be distinguished:

Explicit algorithm-assisted collusion: Competitors agree to fix prices or allocate markets and use software to implement their agreement.

Coordination facilitated by a common intermediary: Competing firms share competitively sensitive information through a pricing provider or platform.

Autonomous algorithmic convergence: Independently operating algorithms produce similar prices without a proven agreement.

The first two situations can raise conventional antitrust concerns where the necessary legal elements are established. The third presents more difficult questions: similar prices or even supracompetitive outcomes do not automatically establish an unlawful agreement under every jurisdiction's law.

B. Predictive merger control

Traditional merger control is already forward-looking. Predictive methods can strengthen assessments of:

Whether an acquisition removes a likely future competitor.

Whether a transaction increases incentives for unilateral price increases.

Whether a vertically integrated firm could restrict rivals' access to essential inputs.

Whether a transaction consolidates control over datasets, cloud infrastructure, AI models or distribution channels.

Whether a merger makes coordination more likely or sustainable.

For example, if a dominant AI platform proposes to acquire a small developer of a promising competing model, current revenue may understate the target's competitive significance. Authorities could examine technical capability, user adoption, innovation trajectories, entry barriers and internal documents concerning the acquisition rationale.

A prediction that the target might become a significant rival would be relevant evidence, but the authority would still need to establish the applicable merger-control test.

C. Digital-platform foreclosure

Predictive tools can examine whether a platform's decisions concerning ranking, interoperability, access fees, default settings or API restrictions may exclude competitors.

Useful indicators include:

Persistent declines in competitor visibility following platform-policy changes.

Disproportionate access restrictions imposed on competing services.

Increased switching costs or reduced multi-homing.

Changes in conversion rates associated with self-preferencing.

Increasing dependence on a dominant platform for customer acquisition.

The Google Shopping judgment illustrates why the competitive consequences of platform design and ranking can matter under established competition law.

D. Predatory pricing and limit pricing

Predictive models can identify pricing strategies that may undermine competition over time. They can assess price-cost relationships, the duration of below-cost pricing, competitors' exit risks and the possibility of recouping losses through later increases in market power.

However, low prices are generally beneficial to consumers. Authorities must distinguish aggressive but lawful competition from conduct that meets the relevant jurisdiction's legal requirements for predation or exclusion.

E. Data concentration and ecosystem power

In data-intensive markets, competitive advantages can reinforce themselves through feedback loops:

More users generate more data.

More data may improve a service or model.

Better performance may attract more users.

Greater scale may increase entry barriers and reduce rivals' access to customers.

Predictive analysis can test whether these effects are substantial and durable, whether alternatives exist, and whether competitors can obtain equivalent inputs. It should not assume that possessing more data necessarily creates dominance or constitutes an infringement.

7. Proposed institutional framework for competition authorities

A competition authority adopting predictive antitrust could organise its work around the following framework.

Risk classification

An authority could classify potential cases into three investigative categories.

Level 1

Routine monitoring

Ordinary market changes or weak predictive signals. Continue observation and improve data quality.

Level 2

Enhanced review

Multiple independent indicators of possible foreclosure, coordination or harmful concentration. Undertake targeted economic and factual investigation.

Level 3

Formal enforcement assessment

Credible evidence supporting a legally cognisable theory of harm. Consider formal proceedings or interim measures where authorised and justified.

These levels are a proposed administrative design, not an existing universal legal classification. A model's numerical risk score must not determine liability or automatically trigger sanctions.

8. Risks and limitations of predictive antitrust

First, false positives. Similar prices, market concentration and declining entry may reflect lawful competitive dynamics rather than unlawful conduct.

Second, false negatives. Sophisticated firms may change their behaviour, conceal coordination, or operate in markets where the available data do not reveal exclusionary strategies.

Third, historical bias. Models trained on past enforcement decisions may reproduce earlier investigative priorities or overlook new forms of competitive harm.

Fourth, causal uncertainty. A prediction that prices will rise after a merger does not establish that the merger caused the increase. Demand shocks, input costs and regulatory changes must be considered.

Fifth, explainability and due process. Firms should be able to understand and challenge the substantive evidence relied upon against them, subject to legitimate confidentiality protections.

Sixth, regulatory overreach. Authorities should not treat business success, innovation, high prices, or market concentration as sufficient evidence of an infringement without applying the relevant legal standards.

Seventh, strategic adaptation. Businesses may alter pricing, product design or data practices in response to known screening methods, reducing the reliability of static predictive systems.

The appropriate safeguard is a combination of transparent methodology, independent validation, human decision-making, legally authorised evidence collection and periodic reassessment.

9. Future development of predictive antitrust

Future frameworks are likely to focus on five areas:

Continuous merger-risk assessment: Monitoring acquisitions involving nascent competitors, AI capabilities and critical digital infrastructure.

Algorithmic accountability: Examining how pricing systems use competitor data and whether their operation facilitates unlawful coordination.

Dynamic market definition: Assessing innovation pipelines, switching, multi-homing and potential competition rather than relying exclusively on historical market shares.

Cross-border cooperation: Coordinating investigations where data, platforms and commercial decisions span multiple jurisdictions.

Outcome-based evaluation: Measuring whether enforcement preserves entry, innovation, consumer choice and effective rivalry rather than relying only on case counts or fines.

These developments should complement, not displace, established competition-law doctrines.

10. Conclusion

Predictive antitrust frameworks represent an important evolution in competition-law enforcement. They enable authorities to identify emerging risks, analyse complex digital markets, evaluate future merger effects and investigate potentially exclusionary or coordinated conduct more efficiently.

The cases discussed—particularly Microsoft, Airtours, Wood Pulp, T-Mobile Netherlands, Eturas, Google Shopping, Intel and Topkins—demonstrate the importance of forward-looking analysis, economic evidence, digital-system scrutiny and legally sufficient proof.

The governing principle is that prediction should guide investigation, not replace adjudication. A robust framework combines data science with sound economics, statutory authority, procedural fairness and independent human judgment. This approach allows competition authorities to respond to rapidly evolving markets while protecting both competitive freedom and the rule of law.