Competition Law and Competition Implications of Behavioural Prediction Systems

1. Introduction

Behavioural Prediction Systems (BPS) are technological systems that collect, analyse and predict the behaviour of consumers, competitors, suppliers or other market participants. They may use artificial intelligence, machine learning, profiling, transaction histories, browsing patterns, location data, purchasing behaviour, social interactions, pricing responses and other datasets to predict what a person or firm is likely to do next.

In competition law, behavioural prediction is important because the same technology can produce pro-competitive efficiencies—better recommendations, lower search costs, improved matching and more accurate demand forecasting—or create competitive risks, including personalised exclusion, discriminatory pricing, self-preferencing, algorithmic coordination, foreclosure and data-driven market power.

Modern competition authorities increasingly treat data and predictive systems as potentially important sources of market power. For example, the German Facebook/Meta proceeding connected extensive data combination with Facebook's ability to build highly detailed user profiles, while EU and UK authorities have examined how control over data and ranking systems can affect competition.

2. Meaning of Behavioural Prediction Systems

A behavioural prediction system generally operates through five stages:

Data collection → Profiling → Prediction → Automated decision → Market feedback

Examples

A platform may predict:

• which product a consumer will purchase;
• what price a consumer is willing to pay;
• when a consumer is likely to switch platforms;
• which advertisement will generate a purchase;
• whether a consumer is price-sensitive;
• which competitor a customer is likely to choose;
• whether a rival is likely to reduce its price;
• which sellers are likely to leave a marketplace;
• which users are likely to respond to a promotion.

The system can therefore convert behavioural data into commercially valuable predictions.

3. Why Behavioural Prediction Matters to Competition Law

Traditional competition law generally examines:

1. market definition;
2. market power;
3. agreements or concerted practices;
4. exclusionary conduct;
5. exploitative conduct;
6. mergers and acquisitions;
7. effects on competition and consumers.

Behavioural prediction systems introduce an additional question:

Who controls the data, predictive model, computing infrastructure and resulting behavioural intelligence?

A company may not merely possess data. It may possess a continuous predictive feedback loop:

Consumer behaviour → data → prediction → personalised intervention → consumer response → new data

This can create competitive advantages that become stronger as the system receives more data.

4. Principal Competition Concerns

A. Data-driven market power

A company possessing enormous quantities of behavioural data may be able to predict consumer preferences more accurately than competitors.

This may create:

• economies of scale in data;
• learning effects;
• network effects;
• switching costs;
• entry barriers;
• superior advertising targeting;
• superior product recommendations;
• greater ability to optimise prices.

The German Facebook proceeding is particularly relevant. The Bundeskartellamt found that combining data from Facebook, Instagram, WhatsApp and third-party websites could create exceptionally detailed individual user profiles and regarded the accumulation and combination of data as relevant to Facebook's market power.

5. Personalised Pricing and Price Discrimination

Behavioural prediction can allow a firm to estimate an individual's willingness to pay.

For example:

• Consumer A is predicted to pay ₹1,000.
• Consumer B is predicted to pay ₹1,500.
• Consumer C is predicted to pay ₹700.

The system can automatically adjust offers.

Personalised pricing is not automatically unlawful. Competition law becomes particularly relevant where prediction is combined with:

• dominance;
• exclusionary conduct;
• discriminatory treatment;
• exploitation;
• coordination among competitors;
• restrictions on consumer switching.

The concern becomes greater where consumers cannot determine why different prices are being offered to different users.

6. Algorithmic Collusion

One of the most important competition risks is that predictive systems can make competitors' behaviour highly observable.

Suppose competing firms use systems that continuously:

1. observe competitors' prices;
2. predict their reactions;
3. adjust prices;
4. detect deviations;
5. retaliate against deviations.

Even without traditional face-to-face cartel meetings, technology can potentially make coordination easier.

The U.S. authorities have expressly recognised that algorithms do not provide an exemption from antitrust law. In the hotel-pricing litigation involving Caesars and other hotels, the DOJ and FTC stated that competitors cannot lawfully coordinate prices merely because an algorithm is used as the mechanism.

7. Algorithmic Information Exchange

Behavioural prediction systems may aggregate competitors' information.

For example:

Landlord A + Landlord B + Landlord C → common algorithm → pricing recommendation

If the system uses non-public competitively sensitive information supplied by competing firms, the algorithm may become a mechanism for reducing independent decision-making.

The RealPage litigation is an important contemporary example. The U.S. DOJ alleged that competing landlords supplied sensitive rental information to RealPage, which used that information in algorithmic pricing recommendations.

The subsequent RealPage settlement required restrictions concerning sharing of competitively sensitive information and algorithmic pricing mechanisms.

8. Self-Preferencing Based on Behavioural Predictions

A vertically integrated platform may predict:

"Users are more likely to purchase Product X."

It can then place its own Product X prominently before competing products.

This creates a potential combination of:

data advantage + prediction + ranking control + self-preferencing.

The competition concern is particularly strong where the platform controls the gateway through which competitors reach consumers.

The Google Shopping litigation is an important illustration of the broader principle. The European Commission found that Google favoured its own comparison-shopping service through the positioning and display of search results, and the General Court substantially upheld that finding.

9. Behavioural Prediction and Advertising Markets

Advertising platforms use behavioural prediction extensively.

They predict:

• likelihood of clicking;
• likelihood of purchasing;
• interests;
• demographic characteristics;
• conversion probability;
• advertising responsiveness.

This can produce substantial competitive advantages.

A dominant platform might therefore control:

user data → behavioural prediction → advertising targeting → advertiser demand → publisher access.

Competition authorities may examine whether the platform:

• restricts competitors' access to data;
• gives preferential treatment to its own advertising services;
• uses publisher data to strengthen its own advertising products;
• prevents interoperability;
• exploits its position across multiple levels of the advertising chain.

The UK's CMA, for example, provisionally alleged in its Google ad-tech investigation that Google used dominance in open-display advertising to favour its own ad-tech services. The investigation remained subject to further proceedings rather than representing a final infringement finding at that stage.

10. Behavioural Prediction and Foreclosure

A dominant platform may predict when consumers are likely to switch to competitors.

It can then strategically:

• increase switching costs;
• change ranking;
• restrict interoperability;
• provide discounts to retain users;
• impose contractual restrictions;
• limit access to data;
• degrade rival services;
• alter recommendations.

This creates a potential predictive foreclosure strategy.

Instead of merely reacting to competition, the platform can anticipate competitive threats before they materialise.

11. Behavioural Prediction and Network Effects

Behavioural prediction systems can reinforce network effects.

The cycle may be:

More users → more behavioural data → better predictions → better service → more users → more data.

This is sometimes described as a data-network-effect feedback loop.

Consequently, a small initial advantage may become substantial over time.

Competition law should therefore consider not only current market shares but also:

• data accumulation;
• switching costs;
• multi-homing;
• interoperability;
• access to alternative datasets;
• replicability of the prediction model;
• ability of new entrants to obtain training data.

12. Six Important Case Laws / Enforcement Precedents

1. Bundeskartellamt – Facebook/Meta, B6-22/16 (2019)

Facts

The Bundeskartellamt examined Facebook's practice of combining information from Facebook with data from Instagram, WhatsApp and third-party websites.

Competition issue

The authority considered Facebook dominant in the German social-network market and examined whether requiring users to accept extensive data combination constituted an abuse.

Significance

The case established an important competition-law connection between:

dominance + data collection + data combination + user profiling.

The authority specifically observed that extensive data combination could help Facebook create highly detailed profiles and strengthen its competitive position.

Relevance to behavioural prediction

A behavioural prediction system becomes more powerful as it receives more comprehensive behavioural information. The case therefore illustrates why data accumulation can become a competition concern rather than merely a privacy issue.

2. Meta Platforms Inc. v Bundeskartellamt, C-252/21 (CJEU, 2023)

Significance

The CJEU considered the relationship between competition law and data-protection rules in the Meta/Facebook context.

The case is important because it demonstrated that competition authorities may need to consider the legal circumstances surrounding personal-data processing when assessing potentially abusive conduct.

Relevance

Behavioural prediction frequently depends on personal data.

Therefore:

Competition law + data governance + behavioural profiling

may operate together rather than as completely separate legal fields.

3. Google Shopping – Google Search (AT.39740)

Facts

The European Commission found that Google had abused its dominant position by favouring its own comparison-shopping service in general search results.

The General Court substantially upheld the Commission's decision and the €2.42 billion fine.

Competition issue

Google's ranking and display mechanisms affected how consumers encountered competing services.

Relevance to behavioural prediction

Modern ranking systems increasingly use predicted user behaviour. A platform controlling the ranking mechanism can potentially use behavioural insights to decide:

• which products receive visibility;
• which sellers receive traffic;
• which competitors become less visible.

The case therefore provides an important foundation for analysing predictive ranking and algorithmic self-preferencing.

4. Amazon Marketplace – AT.40462

Facts

The European Commission investigated Amazon's use of non-public marketplace seller data.

Amazon subsequently offered commitments under which it agreed not to use certain non-public data supplied by third-party sellers for its own retail operations in competition with those sellers.

Competition issue

Amazon potentially possessed information about:

• sales;
• prices;
• demand;
• inventory;
• seller performance;
• consumer behaviour.

Relevance

A behavioural prediction system can transform marketplace data into predictions concerning:

which products will sell, at what price, and under what conditions.

This creates a potential platform-as-competitor information advantage.

5. United States v Topkins and related Amazon Marketplace algorithmic pricing conduct

Facts

U.S. enforcement involved sellers using pricing algorithms in connection with an agreement to maintain prices on Amazon Marketplace.

The DOJ described algorithms that were programmed to monitor competing prices and implement the participants' pricing arrangement.

Competition issue

The important principle was that automation does not remove the underlying antitrust problem.

Relevance

A behavioural prediction system may:

• observe competitors;
• predict their responses;
• automatically alter prices;
• make collusion more sustainable.

Therefore, competition authorities can examine the human agreement underlying the algorithm, as well as the algorithm's operation.

6. RealPage Algorithmic Pricing Litigation – United States

Facts

The DOJ alleged that RealPage obtained competitively sensitive rental information from competing landlords and used that information in algorithmic pricing systems.

The DOJ's complaint alleged violations involving Sections 1 and 2 of the Sherman Act.

Subsequent settlements with major landlords prohibited specified forms of algorithmic coordination and sharing of competitively sensitive information.

Competition issue

The case raises the question whether an algorithm can become a hub through which competitors exchange information and align competitive behaviour.

Relevance

It is one of the clearest contemporary examples of:

competitor data → algorithm → prediction → pricing recommendation → reduced competitive independence.

13. Additional Important Precedent: Eturas v Lietuvos Respublikos Konkurencijos Taryba

The CJEU's decision in Eturas (C-74/14) is highly relevant to automated systems.

The case involved an electronic booking system through which a software mechanism could facilitate coordinated discount restrictions among travel agencies.

The significance lies in the proposition that competition law can apply when an electronic system facilitates coordinated behaviour.

For behavioural prediction systems, the broader lesson is that the technical form of communication or coordination does not determine whether competition law applies.

14. Behavioural Prediction and Abuse of Dominance

In a dominance case, authorities may examine whether the system facilitates:

1. Exclusion

The dominant firm uses predictions to identify and disadvantage emerging competitors.

2. Discrimination

Different users or business partners receive different treatment based on predicted behaviour.

3. Self-preferencing

The platform predicts demand and directs consumers toward its own products.

4. Tying

Predictions are used to encourage consumers to remain within an ecosystem.

5. Refusal of access

Competitors are denied access to datasets or interfaces necessary to compete.

6. Exploitative conduct

The system uses behavioural information to impose excessively disadvantageous conditions.

15. Behavioural Prediction and Essential Data

A particularly difficult issue is whether behavioural data can become an essential input.

Relevant questions include:

1. Is the dataset genuinely unique?
2. Can competitors obtain equivalent data?
3. Can data be replicated?
4. Is historical data sufficient?
5. Does the dataset improve prediction accuracy?
6. Are there alternative datasets?
7. Is access technically feasible?
8. Is access legally permissible?
9. Would compulsory access reduce incentives to innovate?

The answer should generally be determined through the facts of the particular market rather than by assuming that every large dataset constitutes an essential facility.

16. Behavioural Prediction in Merger Control

Behavioural prediction systems are also relevant to mergers.

Consider:

Company A: large consumer dataset
Company B: advanced prediction technology

A merger may combine:

Data + AI model + distribution platform

and create a significant competitive advantage.

Authorities may therefore examine:

• data concentration;
• interoperability;
• foreclosure;
• vertical integration;
• access to behavioural datasets;
• algorithmic advantages;
• entry barriers;
• innovation competition;
• future competitors;
• control over AI training data.

A merger may be competitively significant even when the parties have relatively small conventional market shares if the transaction substantially increases control over strategically important data or predictive infrastructure.

17. Behavioural Prediction and Consumer Choice

Behavioural prediction may reduce consumers' effective choice without formally preventing them from choosing competitors.

For example:

Prediction: Consumer is unlikely to switch.

The platform may then:

• reduce incentives to switch;
• make cancellation difficult;
• prioritise its own service;
• personalise offers;
• suppress competing recommendations.

The legal issue is therefore not merely:

"Can the consumer technically choose?"

but potentially:

"How does the dominant firm's predictive system influence the competitive conditions under which that choice is made?"

18. Behavioural Prediction and Dynamic Competition

Traditional market analysis often examines existing competitors.

Behavioural prediction requires greater attention to future competitive threats.

A dominant platform could predict:

• which start-up will grow;
• which seller is gaining market share;
• which technology is becoming attractive;
• which customers are likely to migrate;
• which product could become a substitute.

It could then respond before the competitor becomes a serious threat.

This raises difficult questions concerning innovation competition and nascent competitors.

19. Behavioural Prediction and Algorithmic Transparency

Competition authorities may increasingly need to understand:

• training datasets;
• model objectives;
• optimisation functions;
• ranking variables;
• pricing rules;
• feedback loops;
• audit logs;
• model updates;
• automated interventions;
• human involvement.

However, competition law does not necessarily require businesses to disclose their entire source code.

The more relevant question is generally:

What information is necessary to determine whether the system facilitates or causes anticompetitive conduct?

20. Regulatory Remedies

Possible remedies include:

A. Data separation

Preventing combination of datasets obtained from different services.

B. Data portability

Allowing consumers or businesses to transfer relevant data.

C. Interoperability

Allowing competing systems to interact.

D. Non-discrimination

Preventing discriminatory access to data or infrastructure.

E. Algorithmic auditing

Independent examination of predictive systems.

F. Data-use restrictions

Preventing use of competitors' confidential information.

G. Structural separation

Separating platform, data and competing commercial functions in exceptional circumstances.

H. Transparency obligations

Requiring disclosure of relevant ranking, recommendation or pricing practices.

I. Restrictions on algorithmic coordination

Preventing competing firms from using shared systems containing competitively sensitive information.

The CMA's former Privacy Sandbox investigation illustrates how competition remedies can be directed at the design of technological infrastructure itself. The CMA accepted commitments concerning Google's proposed replacement of third-party cookies, and later released those commitments in October 2025 after concluding that the relevant competition concerns no longer arose under the changed circumstances.

21. Pro-Competitive Uses of Behavioural Prediction

Competition law should not treat behavioural prediction as inherently anticompetitive.

It can generate substantial efficiencies.

Consumer benefits

• personalised recommendations;
• fraud detection;
• better search;
• lower transaction costs;
• improved product discovery;
• faster matching.

Business benefits

• improved inventory management;
• demand forecasting;
• logistics optimisation;
• reduced waste;
• better capacity planning.

Competitive benefits

Prediction systems may also help smaller businesses understand demand and compete with larger firms.

Therefore, the proper competition-law question is generally:

Does the behavioural prediction system improve competition or is it being used to suppress, distort or exploit competition?

22. Competition-Law Analytical Framework

A useful framework is:

Step 1 – Identify the system

What does the behavioural prediction system actually predict?

Step 2 – Identify the data

Where does the data come from?

Step 3 – Identify control

Who controls:

• the data;
• model;
• infrastructure;
• interface;
• outputs?

Step 4 – Define the market

Identify relevant product, service and geographic markets.

Step 5 – Assess market power

Examine:

• market share;
• network effects;
• switching costs;
• data advantages;
• entry barriers;
• multi-homing.

Step 6 – Identify conduct

Determine whether the system facilitates:

• coordination;
• exclusion;
• self-preferencing;
• discrimination;
• tying;
• refusal of access;
• exploitative conduct.

Step 7 – Assess effects

Consider:

• prices;
• quality;
• innovation;
• consumer choice;
• entry;
• rivals' access;
• market structure.

Step 8 – Examine efficiencies

Consider whether the conduct creates verifiable benefits that cannot reasonably be achieved through less restrictive means.

Step 9 – Select remedy

Potential remedies include:

data separation → interoperability → access → non-discrimination → algorithmic safeguards → behavioural commitments → structural remedies

23. Key Legal Principles Emerging from the Case Law

24. Conclusion

Behavioural Prediction Systems represent a major development in competition law because they transform behavioural information into a strategic competitive resource.

Their competition implications arise principally through five mechanisms:

1. Data accumulation
Large datasets can improve predictive accuracy and reinforce market power.

2. Predictive advantage
Superior predictions can create barriers to entry and strengthen network effects.

3. Automated market conduct
Predictions can directly determine prices, rankings, recommendations and access.

4. Competitive coordination
Shared algorithms and common datasets can reduce independent decision-making among competitors.

5. Strategic foreclosure
Dominant platforms may use behavioural intelligence to identify, disadvantage or prevent competitive threats.

The Facebook/Meta, Google Shopping, Amazon Marketplace, Topkins, RealPage and Eturas precedents collectively demonstrate that competition law can address different dimensions of technologically mediated behaviour—from data accumulation and ranking to automated pricing and electronic coordination. The central legal challenge is to distinguish legitimate predictive efficiency from the use of predictive systems to obtain, preserve or exploit market power.

The modern competition-law inquiry can therefore be expressed as:

Data → Prediction → Decision → Market Effect → Competitive Assessment

That framework is particularly important for AI-driven platforms, personalised pricing, digital advertising, recommendation systems, autonomous marketplaces and other markets in which behavioural prediction becomes a core competitive capability.