Competition Law and Algorithmic Detection of Exclusionary Conduct

1. Introduction

Algorithmic detection of exclusionary conduct refers to the use of algorithms, artificial intelligence, machine learning, economic modelling and large-scale data analysis to identify business practices that may unlawfully exclude competitors from a market.

In modern digital markets, potentially exclusionary conduct can be difficult to identify through traditional evidence alone. Pricing, rankings, recommendations, advertising allocation, access conditions and platform visibility may be determined automatically by complex software systems. A competition authority may therefore need to examine not merely contracts and emails, but also datasets, source-code logic, algorithmic outputs, ranking patterns and changes in market behaviour.

The central competition-law question remains essentially traditional: has a dominant undertaking used conduct capable of restricting effective competition through means other than legitimate competition on the merits?

In the European Union, this issue primarily arises under Article 102 TFEU, which prohibits abuse of a dominant position. Similar analytical questions arise under monopolisation and exclusionary-conduct rules in other competition-law systems.

Algorithmic detection does not create a new category of competition-law infringement. Rather, algorithms can provide new evidence and analytical methods for detecting established forms of exclusion, such as:

• predatory or selectively exclusionary pricing;
• loyalty and exclusivity arrangements;
• tying and bundling;
• discriminatory access conditions;
• margin squeeze;
• self-preferencing;
• refusal or restriction of access;
• manipulation of rankings or recommendations;
• interoperability restrictions; and
• practices designed to increase rivals' costs.

2. Why Algorithmic Detection Has Become Important

Digital markets generate enormous quantities of information.

An online platform may make millions of automated decisions concerning search results, advertisements, prices, product recommendations, seller rankings and access to customers. Manual investigation of every individual decision would be practically impossible.

Algorithmic tools can instead examine large datasets and identify recurring patterns.

For example, investigators could compare how a platform's ranking system treats its own products with the treatment received by competing products.

An algorithmic investigation might identify that competing services systematically lose visibility whenever they reach a particular competitive threshold.

This does not automatically prove an infringement. It provides evidence requiring legal and economic interpretation.

The distinction between detection and legal determination is therefore fundamental.

An algorithm can identify unusual behaviour; competition law determines whether that behaviour constitutes unlawful exclusion.

3. Basic Elements of Exclusionary Conduct

An investigation will normally consider several connected questions.

Dominance

The undertaking generally must possess substantial market power before unilateral conduct falls within Article 102 TFEU.

Relevant indicators can include:

• market shares;
• barriers to entry;
• network effects;
• economies of scale;
• access to important datasets;
• switching costs;
• control over infrastructure;
• ecosystem advantages;
• interoperability restrictions; and
• dependence of business users.

Algorithms can help analyse these factors across extremely large datasets.

Conduct

Investigators then identify the behaviour allegedly producing exclusion.

Examples include preferential ranking, exclusivity conditions, discriminatory API access, below-cost pricing, tying products together or restricting interoperability.

Capability of Exclusion

Competition authorities normally need to consider whether the practice is capable of restricting effective competition rather than simply disadvantaging an individual competitor.

This distinction matters because competition law protects the competitive process, not every competitor from ordinary competitive pressure.

Causation

Where market foreclosure is alleged, investigators may also examine whether the challenged conduct caused or was capable of causing the identified competitive effects.

Algorithms can assist by comparing actual market outcomes with appropriate counterfactual scenarios.

4. Algorithmic Detection Through Pricing Analysis

One major application concerns exclusionary pricing.

Machine-learning or statistical systems can examine millions of transactions and identify:

• prices below relevant cost measures;
• selective discounts;
• unusual customer-specific rebates;
• sudden price reductions following rival entry;
• geographic targeting of discounts;
• loyalty-inducing rebate structures; and
• subsequent price increases after competitive pressure declines.

Suppose a dominant platform operates in 500 geographic areas.

Prices remain relatively high in areas without competitors but fall sharply whenever a competing platform enters.

An automated detection system could identify the correlation.

However, correlation alone would not establish predatory pricing. Investigators would still need to consider costs, market circumstances, objective explanations and the applicable legal test.

5. Detecting Algorithmic Self-Preferencing

Self-preferencing has particular importance for digital platforms.

A vertically integrated platform might operate both:

1. an intermediary platform; and
2. products or services competing through that platform.

An algorithm may determine which products receive prominent placement.

Competition investigators can analyse ranking data to determine whether changes systematically benefit the platform's own services while reducing rivals' visibility.

Relevant variables could include:

• ranking position;
• click-through rates;
• conversion rates;
• impressions;
• demotions;
• placement frequency;
• recommendation frequency; and
• treatment following algorithm updates.

This type of analysis became particularly significant in litigation concerning Google's comparison-shopping service.

6. Detecting Exclusion Through Access Restrictions

Another category involves access to infrastructure or digital ecosystems.

A dominant undertaking might control:

• operating systems;
• application stores;
• APIs;
• advertising infrastructure;
• telecommunications networks;
• essential datasets; or
• technical interoperability information.

Algorithmic monitoring can identify differences in the conditions under which competing firms receive access.

For example, an authority could compare response times, API functionality, technical restrictions or data availability provided to the dominant undertaking's own downstream service with those provided to competitors.

Persistent differences may indicate discriminatory treatment, although investigators must determine whether legitimate technical or commercial explanations exist.

7. Algorithmic Detection of Margin Squeeze

Margin squeeze can arise where a vertically integrated dominant undertaking supplies an upstream input while also competing downstream.

Algorithms can continuously compare:

Upstream access price → downstream retail price → downstream operating costs.

Where the remaining margin would not permit an equally efficient downstream competitor to compete sustainably, the pattern may warrant investigation.

Large-scale automated analysis is particularly valuable where prices vary constantly between customers, geographic locations or time periods.

8. Detecting Exclusivity and Loyalty Effects

Some exclusionary arrangements do not contain an obvious clause stating that customers cannot deal with competitors.

Instead, sophisticated rebate structures may create economic incentives producing similar effects.

An algorithmic system could analyse:

• percentage of customer demand covered;
• rebate thresholds;
• duration;
• retroactive rebate effects;
• switching behaviour;
• contestable share of demand; and
• effective price for the relevant portion of demand.

This can assist investigators in determining whether a rebate arrangement may foreclose an equally efficient competitor.

9. Counterfactual Analysis

Algorithmic detection can also assist with counterfactual analysis.

The basic question is:

What would market conditions probably have looked like without the challenged conduct?

Economic models can compare the actual market with simulated scenarios involving:

• neutral rankings;
• absence of exclusivity;
• equal interoperability;
• alternative access conditions;
• different pricing structures; or
• absence of tying.

Counterfactual models must nevertheless be treated carefully.

Their conclusions depend heavily on assumptions, input data and model design. A sophisticated model does not eliminate the need for legal reasoning and evidentiary scrutiny.

10. Important Case Laws

Case 1: Google LLC and Alphabet Inc. v European Commission — Google Shopping

Case C-48/22 P, Court of Justice, 10 September 2024

This is particularly important for algorithmic exclusion and self-preferencing.

Google operated a general search engine while also providing its own specialised comparison-shopping service. The dispute concerned Google's more favourable presentation of its own comparison-shopping results and the treatment of competing comparison-shopping services.

The Court of Justice upheld the finding concerning abuse. The case dealt directly with leveraging, self-preferencing, potential exclusionary effects, causation, counterfactual analysis and the role of an as-efficient-competitor assessment.

Its importance for algorithmic detection is substantial because search rankings are produced through complex automated systems.

An investigation can examine large quantities of ranking data to identify whether changes in algorithms systematically improve the position of an undertaking's own service while competitors are demoted.

Principle: Algorithmically implemented differences in treatment can fall within traditional abuse-of-dominance principles where the relevant legal conditions and exclusionary capability are established.

Case 2: Google LLC and Alphabet Inc. v European Commission — Google Android

Cases T-604/18 and C-738/22 P

The Android proceedings concerned restrictions connected with Google's mobile ecosystem, including product bundling, exclusive pre-installation arrangements and restrictions affecting Android forks.

The General Court largely upheld the Commission's infringement findings in 2022. On appeal, the Court of Justice delivered its judgment on 2 July 2026, addressing contractual restrictions, tying, exclusionary effects, counterfactual analysis, exclusive pre-installation payments and obstruction of Android forks.

This litigation illustrates why algorithmic detection should not focus exclusively on pricing.

Digital exclusion can emerge from combinations of contractual, technological and ecosystem restrictions.

Investigators can therefore analyse device-level and platform-level data to determine how pre-installation requirements, defaults, technical restrictions and contractual conditions affect rival distribution.

Principle: In digital ecosystems, exclusionary analysis may require examination of several connected mechanisms rather than one isolated algorithm or contractual provision.

Case 3: Intel Corp. v European Commission

Case C-413/14 P, Court of Justice, 6 September 2017

Intel concerned rebates granted by a dominant microprocessor producer.

The case became particularly important for economic assessment of exclusionary rebates. Where a dominant undertaking disputes the capability of a rebate scheme to restrict competition, factors relevant to the analysis can include market position, market coverage, conditions and duration of rebates, their size and mechanisms, and their capacity to exclude an equally efficient competitor.

This framework is well suited to computational analysis.

Algorithms can process enormous transaction datasets and calculate effective prices, rebate thresholds and the proportion of customer demand affected.

They can also identify which portions of demand are genuinely contestable.

Principle: Data-driven economic evidence can play a major role in determining whether rebate arrangements are capable of producing anticompetitive foreclosure.

Case 4: Post Danmark A/S v Konkurrencerådet

Case C-209/10, Court of Justice, 27 March 2012

Post Danmark concerned selectively low prices offered by a dominant postal undertaking to customers previously served by a competitor.

The Court stressed that not every pricing difference imposed by a dominant undertaking automatically constitutes exclusionary abuse.

Competition on price can be legitimate and beneficial.

The analysis must therefore examine the actual economic characteristics of the conduct and whether it is capable of producing exclusionary effects.

This principle provides an important safeguard for algorithmic enforcement.

A detection system may identify:

competitor enters → dominant firm's price decreases.

That pattern alone cannot establish illegality.

Investigators must consider costs, competitive circumstances and possible objective explanations.

Principle: Algorithmic detection should operate as an evidentiary screening mechanism rather than automatically converting unusual pricing patterns into competition-law violations.

Case 5: Slovak Telekom a.s. v European Commission

Case C-165/19 P, Court of Justice, 25 March 2021

Slovak Telekom concerned broadband infrastructure, conditions governing access to the local loop and margin squeeze.

The case is especially useful for algorithmic detection because telecommunications markets contain enormous amounts of measurable information.

An automated monitoring system could examine:

• wholesale prices;
• retail prices;
• downstream costs;
• access delays;
• rejection rates;
• customer switching;
• network availability; and
• competitor margins.

The Court's treatment of access conditions, margin squeeze and the position of an equally efficient competitor illustrates how economic data can support exclusionary-conduct analysis.

Principle: Exclusion may occur not only through complete denial of access but also through access conditions and pricing structures capable of weakening effective downstream competition.

Case 6: Microsoft Corp. v Commission

Case T-201/04, Court of First Instance, 17 September 2007

Microsoft remains an important authority concerning technology markets.

The proceedings concerned, among other matters, Microsoft's refusal to supply and authorise the use of interoperability information and the tying of Windows Media Player with its PC operating system.

The case demonstrates how exclusion can arise through technical architecture and interoperability restrictions rather than simply through prices.

Modern algorithmic monitoring could examine whether competitors systematically receive:

• reduced interoperability;
• restricted API functionality;
• slower technical access;
• incomplete documentation; or
• technologically disadvantageous treatment.

Principle: Control over interoperability and technological architecture can be relevant to exclusionary-abuse analysis where the applicable legal requirements are satisfied.

Case 7: Servizio Elettrico Nazionale SpA and Others v AGCM

Case C-377/20, Court of Justice, 12 May 2022

This case provides an important general statement about exclusionary abuse under Article 102 TFEU.

It arose from the liberalisation of the Italian electricity market and involved the use of commercially valuable customer information connected with a former statutory monopoly.

The Court considered whether conduct capable of producing exclusionary effects involved methods different from normal competition on the merits.

The judgment is particularly relevant to algorithmic systems because access to unique customer datasets can itself provide significant competitive advantages.

An algorithmic investigation might therefore examine whether a dominant undertaking uses privileged historical datasets to preserve or extend market power in markets opened to competition.

Principle: Competition authorities can examine whether a dominant firm's conduct relies on advantages or resources that competitors cannot realistically reproduce and whether the conduct departs from competition on the merits.

11. Practical Algorithmic Detection Model

A competition authority could construct an exclusion-detection system using several stages.

Stage 1 — Market Data Collection

The authority collects:

• prices;
• costs;
• search rankings;
• customer transactions;
• advertisements;
• platform commissions;
• access requests;
• API performance;
• contractual conditions;
• rebate information;
• product visibility; and
• competitor entry and exit data.

Stage 2 — Establishing Baselines

The system determines ordinary competitive behaviour.

For example:

Normal ranking → normal pricing → normal access conditions.

Material deviations from the baseline are then flagged.

Stage 3 — Rival-Specific Pattern Detection

The system determines whether behaviour changes specifically when competitive threats emerge.

For example:

Competitor entry → targeted discount → competitor exit → price restoration.

Such a sequence can justify further investigation, although it does not by itself establish abuse.

Stage 4 — Foreclosure Assessment

Authorities examine whether the identified conduct is capable of reducing competitors' ability to compete effectively.

Indicators can include declining traffic, reduced customer acquisition, worsening margins, reduced platform visibility or increasing entry barriers.

Stage 5 — Counterfactual Testing

Investigators model what market conditions could have been without the challenged practice.

Stage 6 — Legal Assessment

Finally, the evidence must be assessed under the relevant legal standard.

The authority considers dominance, nature of the conduct, competitive context, foreclosure capability, causal relationship where required, possible objective justification and efficiencies.

12. False Positives and Algorithmic Enforcement

One of the greatest dangers is the false positive.

An algorithm may classify aggressive but legitimate competition as exclusionary.

For example, a dominant company reducing prices after entry by a rival may simply be competing more vigorously.

Likewise, a ranking algorithm placing the platform's own product higher does not automatically establish an infringement without the necessary legal and factual context.

Human review therefore remains essential.

Competition authorities should not use:

Algorithmic flag = legal infringement.

Instead, the appropriate approach is:

Algorithmic flag → economic investigation → evidentiary assessment → legal determination.

13. Explainability and Transparency

Competition authorities may need to understand why an algorithm generated particular outcomes.

This can become difficult where machine-learning models are highly complex.

Investigators may therefore examine:

• model objectives;
• training data;
• optimisation criteria;
• ranking variables;
• weighting systems;
• internal experiments;
• A/B testing;
• historical versions of algorithms; and
• effects following algorithm updates.

An important question is whether a particular competitive disadvantage results from legitimate optimisation or from rules capable of producing unlawful foreclosure.

14. Intent Versus Effects

Algorithmic systems also complicate the role of intent.

A company might argue that nobody manually instructed an algorithm to exclude competitors.

But the absence of a simple instruction such as "exclude competitor X" does not necessarily resolve the legal question.

The investigation may instead focus on the design, implementation and competitive effects or capability of the practice under the relevant legal test.

Evidence concerning intent may nevertheless help explain why particular parameters, objectives or restrictions were adopted.

15. Algorithmic Evidence and Due Process

Companies must have an adequate opportunity to challenge algorithmic evidence used against them.

This includes the ability, subject to legitimate confidentiality restrictions, to question:

• datasets;
• economic assumptions;
• model specifications;
• statistical significance;
• benchmark selection;
• counterfactual assumptions; and
• possible alternative explanations.

Otherwise, an apparently sophisticated AI system could conceal methodological weaknesses.

Algorithmic enforcement should therefore remain explainable, reviewable and subject to ordinary procedural safeguards.

16. Relationship Between the Major Cases

The seven cases collectively provide a useful framework.

Google Shopping demonstrates how preferential treatment within an algorithmically organised platform can become relevant to exclusionary-abuse analysis.

Google Android demonstrates how contractual restrictions, defaults, tying and ecosystem architecture can interact to affect competition.

Intel demonstrates the importance of rigorous economic analysis when assessing exclusionary rebates.

Post Danmark shows why aggressive competitive behaviour cannot automatically be classified as unlawful exclusion.

Slovak Telekom demonstrates the importance of access conditions and margin-squeeze analysis.

Microsoft demonstrates how interoperability restrictions and tying can affect competition in technology markets.

Servizio Elettrico Nazionale provides broader guidance concerning exclusionary conduct and competition on the merits.

Together, they demonstrate that algorithms do not replace existing competition-law principles. Instead, algorithmic tools can provide new methods of discovering, measuring and proving conduct already addressed by those principles.

17. Conclusion

Algorithmic detection of exclusionary conduct represents an important development in modern competition enforcement.

Markets increasingly operate through automated pricing systems, search engines, recommendation engines, digital advertising systems, app stores and platform-ranking mechanisms. Competition authorities consequently need analytical tools capable of examining millions of automated decisions.

Algorithms can help identify patterns involving selective pricing, self-preferencing, discriminatory access, margin squeeze, exclusivity, tying, interoperability restrictions and other potentially exclusionary practices.

Nevertheless, algorithmic detection must remain separate from the ultimate legal determination of abuse.

A statistical anomaly is not automatically an infringement.

The strongest enforcement model therefore combines:

large-scale algorithmic detection + economic analysis + documentary evidence + counterfactual assessment + procedural safeguards + human legal judgment.

The established jurisprudence in Google Shopping, Google Android, Intel, Post Danmark, Slovak Telekom, Microsoft and Servizio Elettrico Nazionale provides a substantial legal foundation for this approach. These authorities demonstrate that even where exclusion is implemented through sophisticated digital architecture or automated decision systems, the central legal inquiry remains whether a dominant undertaking has engaged in conduct capable of restricting effective competition in circumstances prohibited by competition law.