Competition Law and Governance of Prediction-Based Ecosystems

1. Introduction

A prediction-based ecosystem is an economic environment in which firms use large-scale data, artificial intelligence, machine learning, algorithms, behavioural information and real-time analytics to predict future conduct of consumers, competitors, suppliers or other market participants.

Traditional competition law generally examines observable conduct such as price fixing, exclusive dealing, tying, refusal to deal, discriminatory pricing and mergers. Prediction-based ecosystems create a more complex environment because competitive decisions may be made before the relevant market event occurs.

For example, a platform may predict:

• which product a consumer is likely to purchase;
• what price a consumer will accept;
• which seller is likely to discount;
• which competitor is likely to enter a market;
• when demand will increase;
• which users are likely to switch platforms;
• which suppliers are financially vulnerable;
• what competitors' prices are likely to be in the near future.

The competition-law problem therefore shifts from merely asking "What did the undertaking do?" to also asking "What information did it possess, what did its predictive system infer, and how did it use those predictions to influence competitive conditions?"

2. Meaning of Prediction-Based Ecosystems

A prediction-based ecosystem can be represented as:

Data → Observation → Prediction → Automated Decision → Market Effect → New Data → Improved Prediction

The system becomes increasingly powerful because each transaction generates additional information.

Core characteristics

A. Continuous data collection

Platforms may collect:

• search histories;
• purchases;
• browsing behaviour;
• location information;
• transaction histories;
• supplier information;
• competitor prices;
• customer complaints;
• inventory information;
• advertising responses.

B. Predictive analytics

Machine-learning systems can transform this information into predictions concerning:

• demand;
• prices;
• consumer switching;
• competitor behaviour;
• fraud;
• creditworthiness;
• product popularity;
• supplier reliability.

C. Automated or semi-automated decisions

Predictions can determine:

• prices;
• rankings;
• recommendations;
• advertising placement;
• access to customers;
• credit limits;
• search visibility;
• commissions;
• discounts.

D. Feedback loops

The most important feature is the feedback loop.

A platform predicts what users will buy → changes recommendations → users purchase the recommended product → the platform obtains more data → its prediction becomes more accurate.

This may produce a significant competitive advantage.

3. Why Prediction Creates Competition-Law Concerns

Prediction itself is not unlawful.

Competition law becomes relevant where predictive capabilities are used to produce or reinforce:

1. market power;
2. exclusionary conduct;
3. collusion or coordination;
4. discriminatory treatment;
5. self-preferencing;
6. foreclosure of competitors;
7. exploitative conduct;
8. anticompetitive mergers; or
9. control over strategically important data.

4. Prediction as a Source of Market Power

Traditional market power often depends upon assets such as:

• factories;
• patents;
• distribution networks;
• capital;
• physical infrastructure.

Prediction-based ecosystems introduce another source of power:

predictive informational advantage.

A firm possessing superior predictive data may anticipate market developments before competitors can.

For example:

Platform A

• observes millions of transactions;
• predicts demand;
• predicts competitor reactions;
• adjusts prices;
• recommends its own products;
• collects the resulting data.

The system can create a self-reinforcing competitive advantage.

5. Data as an Input to Predictive Competition

Data can function as an essential competitive input even when it is not itself sold.

Relevant questions include:

• Who controls the data?
• Can competitors obtain comparable data?
• Is the data unique?
• Is it generated by users or suppliers?
• Is access technically restricted?
• Can the data be transferred?
• Does the platform combine datasets?
• Does the platform use competitors' information to compete against them?

The competition-law importance of data depends upon its availability, uniqueness, substitutability, scale and strategic usefulness.

6. Predictive Personalisation and Consumer Discrimination

Prediction can enable highly differentiated treatment.

A platform might predict that:

• Consumer A has a high willingness to pay;
• Consumer B is highly price-sensitive;
• Consumer C is unlikely to switch;
• Consumer D is likely to abandon the transaction.

The platform can then potentially offer different prices, promotions or conditions.

Competition law may become relevant where personalised pricing is used by a dominant undertaking to:

• exploit customers;
• discriminate between equivalent customers;
• exclude competitors;
• prevent switching; or
• reinforce dominance.

However, personalised pricing by itself is not automatically an antitrust violation.

7. Algorithmic Pricing and Tacit Coordination

One of the most significant issues is whether algorithms can facilitate coordination without an explicit human agreement.

Consider:

Competitor A → algorithm predicts B's price → A raises price

Competitor B → algorithm predicts A's price → B raises price

Repeated interaction may produce stable prices even without direct communication.

Competition authorities therefore distinguish between:

Explicit algorithmic collusion

Human actors communicate or agree to use algorithms to implement an anticompetitive arrangement.

Algorithmically facilitated coordination

Algorithms make coordination easier or more stable.

Independent algorithmic adaptation

Each firm independently uses an algorithm to respond to market conditions.

The third category is not automatically unlawful merely because prices become similar.

8. Prediction and Abuse of Dominance

A dominant digital platform may use predictions generated from its ecosystem to disadvantage rivals.

Potential theories include:

A. Self-preferencing

The platform predicts consumer preferences and places its own products ahead of competitors.

B. Leveraging

Predictive information obtained in one market is used to expand dominance into another.

C. Refusal of access

The dominant firm prevents competitors from obtaining strategically important information or data.

D. Discriminatory access

The platform gives predictive tools or information to some firms but not others.

E. Margin or commission manipulation

The platform uses predictions concerning sellers' dependence to impose differentiated commercial conditions.

9. Prediction-Based Ranking

Search engines and marketplaces increasingly predict:

"Which result is most likely to satisfy this particular user?"

Ranking can therefore become an important competitive instrument.

The system may consider:

• user history;
• product characteristics;
• seller performance;
• conversion probability;
• advertising revenue;
• platform commissions;
• predicted engagement.

Competition concerns arise where a dominant intermediary systematically manipulates predictive ranking to disadvantage competing suppliers or favour its own services.

10. Prediction and Network Effects

Prediction-based ecosystems can create powerful data-network effects.

The mechanism is:

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

This can produce a data-driven competitive feedback loop.

A smaller competitor may therefore face a substantial disadvantage even if it possesses technically comparable software.

The relevant competition-law question is whether the feedback loop merely represents legitimate competition on the merits or whether it is reinforced through exclusionary conduct.

11. Prediction and Switching Costs

Prediction systems become more valuable when users remain inside an ecosystem.

Examples include:

• personalised recommendations;
• predictive assistants;
• stored preferences;
• purchasing profiles;
• financial histories;
• health or fitness profiles;
• enterprise behavioural analytics.

A user may therefore face a cost when switching because the new platform lacks the historical information required to provide comparable predictions.

This creates potential competition concerns concerning:

• interoperability;
• data portability;
• tying;
• exclusivity;
• ecosystem foreclosure.

12. Prediction and Mergers

Prediction-based ecosystems create new merger questions.

A merger may involve:

Company A's consumer data + Company B's behavioural data + Company C's predictive technology

The resulting entity may obtain a significant informational advantage.

Authorities may therefore consider:

• data concentration;
• loss of potential competition;
• access to datasets;
• interoperability;
• vertical foreclosure;
• portfolio effects;
• innovation competition;
• ability to combine datasets;
• increased accuracy of prediction.

Traditional turnover-based thresholds may sometimes fail to capture strategically important acquisitions by large digital firms.

13. Six Important Case Laws

1. Eturas v Lietuvos Respublikos Konkurencijos Taryba

Court of Justice of the European Union, Case C-74/14

Facts

Eturas operated an online travel-booking platform used by travel agencies. A system message effectively introduced a restriction on the level of discounts that participating agencies could provide.

Legal significance

The case is important for technology-mediated coordination.

The Court examined whether businesses could be responsible for participating in an anticompetitive arrangement implemented through an electronic system.

Relevance to prediction-based ecosystems

The case demonstrates that competition law can apply where commercial conduct is implemented through a centralised technological platform, even where traditional face-to-face communication is absent.

It is particularly relevant to:

• automated restrictions;
• platform-mediated coordination;
• algorithmic implementation;
• electronic communication between competitors.

14. United States v Topkins

U.S. District Court, Northern District of California, 2015

Facts

Topkins and other online sellers were involved in an agreement concerning the pricing of posters and other goods sold through online marketplaces. Algorithms were used to implement agreed prices.

Legal significance

The case is a prominent example of algorithm-supported price fixing.

The important principle is that the use of an algorithm does not transform an otherwise unlawful agreement into independent competitive conduct.

Relevance

It establishes an important distinction:

Technology may implement collusion, but technology does not legalise collusion.

Prediction-based ecosystems therefore require investigation of the relationship between:

• human agreement;
• algorithmic instructions;
• data;
• automated pricing;
• resulting market conduct.

15. FTC v Amazon.com, Inc.

U.S. Federal Trade Commission and State Attorneys General litigation

Competition significance

The U.S. litigation concerning Amazon addresses a range of alleged practices involving Amazon's marketplace, sellers, pricing and competitive conditions.

The case is relevant to prediction-based ecosystems because large marketplaces possess extensive information concerning:

• seller behaviour;
• consumer demand;
• prices;
• product performance;
• transaction histories.

Relevance

The broader competition issue is whether a vertically integrated platform can use its informational and technological advantages to influence marketplace competition.

It illustrates why competition authorities increasingly examine platform architecture, seller dependence, data advantages and algorithmically mediated marketplace conditions together, rather than considering pricing practices in isolation.

16. Google Search (Shopping)

European Commission, Google Shopping Decision, 2017

Facts

The European Commission found that Google had abused its dominant position by favouring its comparison-shopping service in general search results and displaying it prominently while competing comparison-shopping services were subject to less favourable treatment.

Legal significance

The case is fundamental to the relationship between:

• search algorithms;
• ranking;
• dominance;
• self-preferencing;
• platform intermediation.

Relevance to prediction-based ecosystems

Search rankings increasingly depend on predictions concerning relevance and user behaviour.

Where a dominant platform controls the ranking mechanism, it can potentially determine which competitors receive visibility.

The competition-law concern therefore extends beyond price to control over discoverability.

17. Google Android

European Commission, 2018

Facts

The European Commission examined Google's conduct concerning Android, including restrictions relating to mobile applications, search and browser distribution.

Legal significance

The decision addressed:

• tying;
• default status;
• distribution;
• mobile ecosystems;
• barriers to competing services.

Relevance

Prediction-based ecosystems frequently operate through interconnected products rather than isolated markets.

A dominant firm may use one ecosystem component to influence another.

The case therefore helps explain the competition-law concept of ecosystem leveraging.

18. Microsoft v Commission

General Court of the European Union, T-201/04

Facts

The case concerned Microsoft's conduct involving interoperability information and the incorporation of Windows Media Player into Windows.

Legal significance

The case addressed important questions concerning:

• interoperability;
• leveraging;
• tying;
• exclusion of competitors;
• technological ecosystems.

Relevance

Prediction-based ecosystems increasingly depend upon interoperability between:

• platforms;
• applications;
• databases;
• APIs;
• AI systems;
• predictive services.

Microsoft therefore provides an important conceptual foundation for examining whether technological control can be used to restrict competition in adjacent markets.

19. Google Android Auto / Enel X

European Commission / CJEU digital-platform interoperability jurisprudence

The broader European digital-platform jurisprudence concerning interoperability is significant for prediction-based ecosystems.

Where a platform controls an operating environment, refusing interoperability may prevent competing applications from accessing users.

Relevance

The principle becomes particularly important when predictive services require:

• APIs;
• operating-system access;
• sensor information;
• user permissions;
• platform interfaces.

A dominant ecosystem can potentially make its own predictive service more effective by restricting rivals' access to equivalent technological inputs.

20. Meta Platforms / Bundeskartellamt

CJEU, Case C-252/21

Facts

The case concerned Meta's combination of personal data collected from different services and the relationship between data processing and competition law.

Legal significance

The CJEU examined the relationship between:

• competition law;
• personal data;
• dominance;
• data combination.

Relevance to prediction-based ecosystems

This is especially important because predictive systems often become more powerful when datasets from multiple services are combined.

For example:

Social-network data + browsing data + marketplace data + advertising data

may create significantly greater predictive capability than any dataset alone.

The case demonstrates that data practices can become relevant to competition-law analysis where they are connected with the conduct of a dominant undertaking.

21. United Brands v Commission

CJEU, Case 27/76

Although predating digital ecosystems, United Brands remains important to prediction-based markets.

Principle

The case is foundational for the analysis of:

• dominance;
• discriminatory conditions;
• exploitation;
• market power.

Relevance

In a prediction economy, discrimination may occur through automated systems rather than human decision-making.

The underlying competition-law question remains:

Does a dominant undertaking use its market position to impose materially different conditions on comparable trading partners without an objectively justified basis?

22. Google Shopping, Eturas and Topkins: Combined Importance

These cases illustrate three different dimensions of technology-mediated competition:

23. Competition Law Framework

A. Agreement and concerted-practice rules

Competition authorities should examine whether algorithms merely implement an independent commercial decision or implement:

• an agreement;
• communication between competitors;
• coordinated pricing;
• information exchange;
• market allocation.

Central question

Was there human coordination behind the algorithm?

24. Abuse of Dominance

For a dominant prediction-based platform, authorities may investigate:

1. Self-preferencing

Giving the platform's own predictive service preferential treatment.

2. Data foreclosure

Preventing competitors from obtaining comparable information.

3. API discrimination

Providing superior technological access to affiliated businesses.

4. Algorithmic discrimination

Using ranking or prediction systems to disadvantage particular competitors.

5. Tying

Making access to one predictive service conditional upon use of another product.

6. Exclusivity

Preventing users or suppliers from using competing predictive systems.

25. Essential Facility and Prediction Infrastructure

A particularly difficult question is whether certain predictive infrastructure can constitute an economically indispensable facility.

Possible examples include:

• critical transaction data;
• interoperability interfaces;
• payment data;
• market-wide datasets;
• digital identity infrastructure;
• technical APIs.

However, not every valuable dataset is an essential facility.

Authorities generally need to examine:

1. indispensability;
2. lack of realistic alternatives;
3. feasibility of access;
4. competitive harm;
5. justification for refusal.

26. Algorithmic Collusion

A useful analytical model is:

Competitor A's data

↓

Predictive algorithm

↓

Prediction of Competitor B

↓

Pricing response

↓

Competitor B observes A's behaviour

↓

B's algorithm predicts A

↓

Stable high-price equilibrium

The difficulty is distinguishing:

Legitimate adaptation

from

Coordinated conduct

and from

Explicit collusion implemented through algorithms.

Competition authorities should therefore investigate the design, inputs, instructions, communication and implementation of algorithms.

27. Algorithmic Transparency

Transparency can assist competition enforcement.

Important questions include:

• What data does the algorithm use?
• Who controls the model?
• What objectives are programmed?
• Are competitors' prices incorporated?
• Are competitors' commercially sensitive data used?
• Can employees modify the algorithm?
• Are there safeguards against coordinated pricing?
• Are algorithmic outputs reviewed?
• Is the algorithm trained using competitors' information?

However, competition law must balance transparency against:

• trade secrets;
• cybersecurity;
• intellectual property;
• privacy.

28. Predictive Data Sharing

Data sharing can have opposite competitive effects.

Pro-competitive

Sharing can:

• improve interoperability;
• reduce transaction costs;
• increase innovation;
• facilitate entry;
• improve consumer choice.

Anti-competitive

It can also:

• facilitate collusion;
• reveal future pricing;
• expose commercially sensitive information;
• strengthen dominant firms;
• exclude rivals.

Therefore, data sharing should not be classified automatically as either pro-competitive or anti-competitive.

The relevant question is its effect in the particular market structure.

29. Prediction-Based Mergers

Competition authorities should consider whether a merger combines:

Large dataset + large user base + predictive technology + distribution infrastructure.

Potential consequences include:

• greater predictive accuracy;
• higher barriers to entry;
• increased switching costs;
• foreclosure of rivals;
• increased advertising power;
• improved personalised pricing;
• reduced innovation.

Merger analysis should therefore consider not only current market shares, but also the parties' control over future competitive capabilities.

30. Remedies

Competition authorities can employ several remedies.

Structural remedies

• divestiture;
• separation of business units;
• asset separation.

Behavioural remedies

• non-discrimination;
• interoperability;
• data portability;
• API access;
• restrictions on data combination;
• ranking neutrality;
• transparency obligations.

Algorithmic remedies

• independent audits;
• monitoring;
• documentation;
• human oversight;
• prohibition of certain algorithmic inputs;
• compliance controls.

31. Governance Model for Prediction-Based Ecosystems

An effective governance framework can be represented as:

DATA GOVERNANCE

↓

ALGORITHM GOVERNANCE

↓

MARKET-POWER ASSESSMENT

↓

COMPETITIVE-EFFECT ANALYSIS

↓

MONITORING

↓

REMEDIES

Stage 1 — Data

Determine:

• ownership;
• access;
• portability;
• exclusivity;
• sensitivity;
• substitutability.

Stage 2 — Algorithm

Determine:

• objective;
• inputs;
• decision rules;
• degree of automation;
• human intervention.

Stage 3 — Market

Assess:

• dominance;
• network effects;
• switching costs;
• entry barriers;
• multi-homing.

Stage 4 — Conduct

Examine:

• tying;
• exclusivity;
• discrimination;
• self-preferencing;
• information exchange;
• collusion.

Stage 5 — Effects

Assess:

• foreclosure;
• higher prices;
• reduced choice;
• reduced innovation;
• reduced quality;
• reduced entry.

32. Role of Competition Authorities

Competition authorities increasingly need technological expertise.

A modern investigation may require:

• algorithmic auditing;
• data analysis;
• source-code examination;
• econometric analysis;
• digital forensics;
• network analysis;
• machine-learning expertise.

Traditional evidence such as contracts and emails remains important, but investigators may also need to examine:

• model logs;
• training datasets;
• API records;
• algorithmic changes;
• A/B testing;
• ranking changes;
• automated pricing records.

33. Indian Competition-Law Perspective

In India, prediction-based ecosystems can be examined principally under the Competition Act, 2002, particularly:

• Section 3 — anti-competitive agreements;
• Section 4 — abuse of dominant position;
• Section 5 — combinations;
• Section 19 — inquiry by the Competition Commission of India;
• Section 26 — investigation procedure;
• Section 27 — orders against infringement.

The Indian digital-market cases involving platforms, online marketplaces and technology ecosystems demonstrate the growing importance of:

• data;
• network effects;
• platform dependence;
• self-preferencing;
• discriminatory access;
• tying;
• interoperability.

34. Important Indian Case-Law Connections

1. Matrimony.com Ltd. v Google LLC

The CCI examined Google's practices in relation to search and search-related services.

Relevance: search ranking, platform power and preferential treatment.

2. Umar Javeed v Google LLC

The CCI examined Google's conduct concerning the Android ecosystem.

Relevance: ecosystem effects, tying and technological leverage.

3. Federation of Hotel & Restaurant Associations of India v MakeMyTrip

The CCI considered conduct involving an online platform and hotel businesses.

Relevance: platform dependence, distribution and contractual restrictions.

4. Samir Agarwal v ANI Technologies

The case concerned allegations involving algorithmically determined pricing in the cab-hailing sector.

Relevance: algorithmic pricing and the distinction between platform-controlled pricing and independent competitive conduct.

These Indian cases are useful when applying traditional Competition Act principles to predictive and algorithmically mediated markets.

35. Major Legal Challenges

A. Causation

It can be difficult to prove that an algorithm caused an anticompetitive outcome.

B. Counterfactual analysis

Authorities must determine what the market would have looked like without the challenged algorithmic conduct.

C. Black-box models

Complex machine-learning models may be difficult to interpret.

D. Rapid technological change

An investigation may take years while the underlying technology changes within months.

E. Data confidentiality

Authorities must balance disclosure with protection of commercially sensitive information.

F. False positives

Similar algorithmic outcomes do not necessarily prove collusion.

36. Future Competition-Law Issues

Prediction-based ecosystems are likely to create disputes involving:

1. Generative AI

AI systems may predict:

• consumer demand;
• purchasing decisions;
• competitor responses;
• optimal prices.

2. Autonomous pricing

Pricing may occur continuously without direct human intervention.

3. Predictive credit markets

AI may predict default, switching and purchasing behaviour.

4. Predictive advertising

Platforms may predict which consumers are most valuable to advertisers.

5. Autonomous marketplaces

AI agents could negotiate prices and contractual conditions independently.

6. Predictive supply chains

Dominant platforms could forecast supplier vulnerabilities and demand conditions.

7. AI-agent competition

Future consumers may use autonomous agents to search, negotiate and purchase products, potentially changing the meaning of consumer choice and market power.

37. Key Principles

The governance of prediction-based ecosystems can be reduced to ten principles:

1. Prediction itself is not an antitrust offence.
2. Algorithmic conduct remains subject to competition law.
3. Algorithms cannot disguise an underlying cartel agreement.
4. Data can constitute an important source of competitive advantage.
5. Dominant platforms may face heightened scrutiny when controlling predictive infrastructure.
6. Self-preferencing can become significant where ranking determines market access.
7. Interoperability can be critical in ecosystem competition.
8. Data sharing may be either pro-competitive or anti-competitive depending on circumstances.
9. Merger analysis must consider future predictive capabilities, not merely existing market shares.
10. Competition enforcement increasingly requires technical and algorithmic expertise.

38. Conclusion

Prediction-based ecosystems represent a transition from reactive competition to anticipatory competition. Firms no longer merely respond to existing demand; sophisticated systems can forecast consumer behaviour, competitor strategies, market movements and future demand and then automatically modify commercial conduct.

Competition law must therefore examine the entire chain:

Data → Prediction → Algorithm → Decision → Market Effect → Feedback

The principal legal challenge is not to prohibit predictive technology. Prediction can generate substantial efficiencies, better products, lower transaction costs and improved matching between consumers and suppliers.

The competition-law task is instead to determine when predictive capability becomes a source or instrument of market power, particularly where it is combined with dominance, exclusive data access, self-preferencing, discriminatory algorithms, tying, foreclosure or coordinated conduct.

The jurisprudence of Google Shopping, Google Android, Microsoft, Eturas, Topkins, Meta/Bundeskartellamt and United Brands, together with emerging Indian digital-platform cases, provides a foundation for applying established competition principles to these increasingly automated ecosystems.