Competitive Homogeneity Induced by Shared Algorithms

1. Introduction

Competitive homogeneity induced by shared algorithms refers to a situation in which competing firms use the same, substantially similar, or commonly supplied algorithmic system to make strategic decisions—particularly prices, discounts, output, inventory, advertising, wages, or other competitive variables—with the result that their behaviour becomes increasingly uniform.

The central competition-law concern is not simply that several firms use algorithms. Algorithms ordinarily improve efficiency and can intensify competition. The concern arises where a common algorithm, common data source, common intermediary, or common optimization objective reduces independent decision-making and makes competitors behave as though they were coordinating.

For example:

Competitor A and Competitor B independently purchase the same pricing software. If the software merely helps each firm process its own data, competition may remain vigorous. If the software uses competitors' confidential information and recommends substantially aligned prices, the algorithm can become a mechanism for coordination.

The UK Competition and Markets Authority has specifically identified the use of the same algorithm or third-party pricing service by competing businesses as a potential mechanism through which competitively sensitive information can be shared and competitive behaviour can become coordinated.

2. Meaning of Competitive Homogeneity

Competitive homogeneity occurs when firms that should ordinarily make independent strategic choices increasingly converge on:

• identical or near-identical prices;
• similar discounts;
• similar output levels;
• similar inventory decisions;
• similar advertising strategies;
• similar contract terms;
• similar wages or commission structures;
• similar capacity decisions;
• similar responses to competitors;
• similar timing of price changes; or
• similar market-entry and exit decisions.

The important question is therefore not:

“Are the firms using algorithms?”

but:

“Does the algorithm reduce or eliminate the independent competitive uncertainty that normally exists between rivals?”

3. How Shared Algorithms Can Produce Competitive Homogeneity

A. Common pricing software

Several competitors may purchase the same algorithm from a software provider.

The algorithm may observe:

• competitors' prices;
• historical transactions;
• demand;
• inventory;
• capacity;
• customer behaviour; and
• market trends.

If it then produces similar recommendations for every competitor, independent pricing can progressively disappear.

B. Common data pool

A particularly significant risk arises where competing firms contribute their confidential information to a common database.

For example:

Firm A's prices → common database → algorithm → Firm B's pricing recommendation

and simultaneously:

Firm B's prices → common database → algorithm → Firm A's pricing recommendation

The software can thereby become an indirect information-exchange mechanism.

C. Common optimization objective

Algorithms may be programmed to maximize:

• revenue;
• price;
• margin;
• occupancy;
• utilization;
• market share;
• yield; or
• long-term profitability.

If competing firms rely on the same optimization logic and sufficiently similar data, their strategies may converge.

D. Algorithmic reaction to competitors

Traditional human pricing may involve delays and imperfect observation.

An algorithm can react within seconds:

Competitor raises price → Algorithm detects change → Firm raises price → Competitor's algorithm detects change → Competitor raises price.

Repeated interactions can therefore create highly stable price alignment.

E. Hub-and-spoke structure

A third-party software provider can become the hub connecting competing firms.

The structure can be represented as:

Firm A ↘

Firm B → Common Algorithm/Provider → Pricing Recommendations

Firm C ↗

The intermediary may therefore occupy a position through which competitively sensitive information is aggregated, processed and redistributed.

4. Competitive Homogeneity Is Not Automatically Illegal

This distinction is fundamental.

The mere fact that firms:

• use identical software;
• use the same AI model;
• adopt similar prices;
• react similarly to market conditions; or
• independently reach similar commercial strategies

does not automatically establish an infringement.

Competition law generally requires additional evidence showing, depending on the jurisdiction:

1. an agreement;
2. concerted practice;
3. exchange of competitively sensitive information;
4. participation in a facilitating mechanism;
5. exclusionary conduct by a dominant undertaking; or
6. another legally recognized anticompetitive mechanism.

The UK CMA has expressly recognized three broad algorithmic-collusion concerns: algorithms facilitating explicit coordination, common algorithmic systems creating information-exchange or hub-and-spoke structures, and potentially autonomous tacit coordination.

5. Relevant Legal Framework

A. Article 101 TFEU

Under EU competition law, Article 101 may apply where the use of a shared algorithm forms part of:

• an agreement;
• a decision by an association;
• a concerted practice; or
• an information-exchange arrangement

that has the object or effect of restricting competition.

The crucial concept is independent market conduct.

Competitors must ordinarily determine their market behaviour independently.

B. Competition Act 1998 — United Kingdom

In the UK, Chapter I of the Competition Act 1998 prohibits agreements, decisions and concerted practices that have the object or effect of preventing, restricting or distorting competition.

A shared algorithm can become relevant where it facilitates:

• price fixing;
• market sharing;
• coordination;
• exchange of commercially sensitive information; or
• implementation of an existing cartel.

The CMA has stated that businesses should be particularly careful when the same algorithmic system is used by competitors.

C. Sherman Act — United States

In the United States, algorithmic coordination can potentially implicate:

• Section 1 Sherman Act — agreements restraining trade;
• Section 2 Sherman Act — monopolization or attempted monopolization; and
• related doctrines concerning information exchange and facilitating mechanisms.

The essential problem under Section 1 remains whether the algorithm is merely independently used technology or an instrument through which competitors coordinate.

6. Case Laws

Case 1 — Eturas UAB v Lietuvos Respublikos konkurencijos taryba

C-74/14, Court of Justice of the European Union, 2016

This is one of the most important cases concerning technology-enabled coordination.

Facts

Travel agencies used the common E-TURAS computerized booking system.

A system message announced a restriction on online discounts. The software subsequently technically limited discounts available through the system.

The Lithuanian competition authority treated the conduct as a concerted practice.

Legal issue

Could competitors' use of a common computerized system, together with knowledge of an algorithmically implemented restriction, constitute evidence of concerted conduct?

Decision

The CJEU held that the existence of a concerted practice could be established where users knew of the anticompetitive system message and continued participating, subject to the applicable evidentiary requirements.

The Court also emphasized that firms must have the opportunity to rebut the inference—for example, by demonstrating that they did not know of the message or that they took appropriate steps to distance themselves.

Relevance to competitive homogeneity

Eturas demonstrates that a common technological infrastructure can become the mechanism through which competitors' commercial conduct is standardized.

The algorithm does not need to negotiate directly with every firm. The technological system itself can implement the common restriction.

Principle

Common digital infrastructure can be legally relevant when it facilitates coordinated competitive behaviour.

7. Case 2 — Trod Ltd / GB eye Ltd

UK CMA, 2016

This is a leading UK example of algorithmically implemented price coordination.

Facts

Trod and GB eye competed in selling posters and frames through Amazon Marketplace.

They agreed that they would not undercut one another.

The businesses then used automated repricing software to implement the arrangement.

The software monitored prices and adjusted them so that neither business undercut the other.

Legal significance

The CMA treated the underlying agreement as illegal price fixing.

The fact that software was used to execute the arrangement did not make the conduct technologically neutral.

Relevance

The case establishes an important distinction:

Human agreement + algorithmic implementation = ordinary cartel principles still apply.

The algorithm merely increased the effectiveness and speed of the cartel.

The CMA imposed a fine of more than £160,000 on Trod, while GB eye received immunity after reporting and cooperating with the investigation.

Principle

Technology does not immunize an otherwise unlawful agreement.

8. Case 3 — United States v David Topkins

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

This was an important early U.S. prosecution involving algorithmic pricing.

Facts

David Topkins and co-conspirators agreed to fix prices of posters sold through Amazon Marketplace.

The participants adopted pricing algorithms and wrote computer code designed to implement coordinated pricing.

The Department of Justice charged the conduct as horizontal price fixing.

Significance

The case demonstrated that:

• online commerce is not outside antitrust law;
• computer code can implement a cartel;
• algorithmic sophistication does not change the underlying economic character of price fixing.

Competitive-homogeneity dimension

The competitors' use of compatible algorithmic pricing strategies helped ensure that their prices moved according to the coordinated arrangement rather than independent competitive incentives.

Principle

The use of computer code to execute an agreement does not change the underlying antitrust analysis.

9. Case 4 — United States and State Plaintiffs v RealPage, Inc.

U.S. District Court, Middle District of North Carolina, filed 2024; developments continued through 2025

This is particularly important for the modern concept of shared algorithmic decision-making.

Allegations

The U.S. Department of Justice and participating states alleged that RealPage's revenue-management software facilitated coordination among competing landlords.

According to the complaint, participating landlords supplied RealPage with non-public, competitively sensitive information concerning rental prices and lease terms.

The algorithm then generated pricing recommendations using information concerning competing properties.

Why it matters

This differs from Topkins and Trod/GB eye.

There, firms had an identifiable agreement to coordinate prices.

The RealPage litigation raises a broader question:

Can an intermediary's algorithm create competitive alignment by aggregating competitors' sensitive information and producing recommendations based upon it?

The DOJ alleged violations of Sections 1 and 2 of the Sherman Act.

Later development

In November 2025, the DOJ announced a proposed settlement involving RealPage addressing information sharing and pricing alignment among competitors.

Principle

A common algorithmic intermediary may create competition concerns where competitors surrender independent pricing decisions to a system influenced by rivals' confidential information.

10. Case 5 — Wood Pulp / Ahlström Osakeyhtiö and Others v Commission

Joined Cases 89/85 etc., CJEU, 1993

This case is not an algorithm case, but it provides an important foundational principle for analysing algorithmic homogeneity.

Principle

The EU competition framework distinguishes between:

• genuinely independent parallel conduct; and
• conduct resulting from coordination.

Parallel behaviour alone is not necessarily sufficient to establish concerted practice.

Importance for algorithms

Suppose ten competitors independently use the same commercially available AI pricing system and prices converge.

That convergence does not automatically prove collusion.

An enforcement authority would need to establish additional evidence connecting the similarity to a prohibited coordination mechanism.

Principle

Parallel conduct must not automatically be equated with collusion.

This principle is extremely important in AI markets because machine-learning systems can independently produce similar outputs from similar economic conditions.

11. Case 6 — Suiker Unie v Commission

Joined Cases 40–48, 50, 54–56, 111, 113 and 114/73, CJEU, 1975

This is another foundational case concerning concerted practices.

Principle

EU competition law recognizes that competitors can unlawfully reduce the uncertainty surrounding their future conduct without necessarily entering into a traditional formal contract.

The concept of concerted practice therefore reaches beyond conventional written agreements.

Relevance to shared algorithms

An algorithmic system can potentially reduce strategic uncertainty by providing competitors with:

• current prices;
• future pricing signals;
• production information;
• capacity information;
• demand information; or
• recommendations reflecting rivals' behaviour.

Thus, an algorithm may perform a function that previously required direct human communication.

Principle

Competition law can address coordination that operates through conduct and information exchange even without a conventional written cartel agreement.

12. Case 7 — Anic Partecipazioni SpA v Commission

Case C-49/92 P, CJEU, 1999

This case provides an important explanation of the relationship between:

• agreement;
• concerted practice; and
• parallel conduct.

Relevance

The Court recognized that the concepts of agreement and concerted practice can overlap in the broader Article 101 framework.

For algorithmic markets, this matters because coordination may be difficult to characterize using traditional categories.

A firm may not explicitly tell its competitor:

“Raise your price to £100.”

Instead, the information may pass through a common algorithmic mechanism.

Principle

Competition law focuses on the substance of coordinated market conduct rather than merely on formal contractual language.

13. Case 8 — A. G. v Uber / Meyer v Kalanick — Algorithmic Pricing Context

U.S. litigation concerning Uber's pricing system has also generated important discussion concerning the competitive implications of algorithmic pricing.

The underlying issue was whether Uber's algorithmic surge-pricing structure could facilitate coordination among independent drivers.

Although the procedural and substantive questions differ from cartel cases such as Topkins, the litigation illustrates the difficulty of applying traditional antitrust concepts to a platform that uses a centralized algorithm to influence the pricing decisions of economically independent participants.

Relevance

The case highlights a critical question:

When does a platform merely provide technological infrastructure, and when does the platform's algorithm become a mechanism that coordinates competitive behaviour?

That distinction is increasingly important for AI-mediated marketplaces.

14. Comparative Significance of the Cases

15. The Three Main Models of Algorithmic Homogeneity

Model 1 — Explicit coordination

This is the easiest category to prosecute.

Human agreement → algorithm → uniform prices

Example:

“We will not undercut each other.”

The algorithm simply implements the cartel.

Trod/GB eye and Topkins are particularly illustrative.

Model 2 — Algorithmic information exchange

Here, firms may not communicate directly.

Instead:

Firm A → confidential data → common provider

Firm B → confidential data → common provider

Provider → algorithmic recommendation → A + B

The algorithm therefore creates an indirect communication channel.

RealPage is highly relevant to this model.

Model 3 — Autonomous algorithmic coordination

This is the most difficult category.

Suppose:

• A uses Algorithm X;
• B uses Algorithm X;
• neither communicates with the other;
• both algorithms observe market prices;
• both algorithms learn that aggressive undercutting reduces profits;
• both gradually converge toward higher prices.

There may be no traditional human agreement.

This creates the difficult legal question:

Can competition law address coordinated outcomes produced by autonomous systems when no human expressly instructed the algorithms to coordinate?

Current competition authorities regard this as an important emerging issue, but autonomous tacit coordination should not automatically be treated as equivalent to an established cartel. The CMA has specifically distinguished autonomous tacit coordination from explicit coordination facilitated by algorithms.

16. Competitive Homogeneity and Tacit Collusion

Algorithms can make tacit coordination more stable because they can:

1. observe rivals continuously;
2. detect deviations immediately;
3. react rapidly;
4. remember historical behaviour;
5. calculate optimal responses;
6. punish deviations automatically; and
7. repeatedly interact without human delay.

A simplified model is:

Price A ↑

↓

Algorithm B observes

↓

Price B ↑

↓

Algorithm A observes

↓

Price A ↑

↓

Stable high-price equilibrium

The legal problem is determining whether this is:

• lawful independent adaptation;
• conscious parallelism;
• algorithmic tacit coordination; or
• evidence of an underlying agreement or concerted practice.

17. Competitive Harm

Shared algorithms can create several forms of harm.

1. Price homogenization

Competitors stop aggressively undercutting each other.

2. Reduced price dispersion

Independent firms normally have different costs and strategies. Algorithmic standardization may narrow price differences.

3. Reduced innovation

Competition may shift from:

“Who can develop the better product?”

to:

“Who can optimize the same algorithm?”

4. Reduced consumer choice

Uniform algorithmic recommendations can produce similar:

• prices;
• discounts;
• contract terms; and
• service offerings.

5. Higher barriers to entry

New entrants may need access to:

• the same data;
• the same software;
• the same infrastructure; and
• the same computational resources

to compete effectively.

6. Strategic transparency

Excessive algorithmic transparency can reduce uncertainty among competitors.

18. Important Distinction: Homogeneity vs Efficiency

Competitive similarity is not necessarily harmful.

A common algorithm can generate legitimate efficiencies by:

• reducing transaction costs;
• improving inventory management;
• reducing waste;
• improving demand forecasting;
• lowering search costs;
• enabling small businesses to compete;
• improving matching;
• reducing logistics costs.

Indeed, the CMA notes that pricing algorithms can generate benefits such as faster price adjustment, lower business costs and potentially more intense competition.

Therefore, the competition-law inquiry should distinguish:

Efficiency-driven convergence

from

coordination-driven convergence.

19. Evidence Relevant to Enforcement

Authorities investigating competitive homogeneity may examine:

A. Algorithm design

• What objective does the algorithm optimize?
• Does it maximize individual profit or industry-wide profitability?
• Does it incorporate rivals' information?

B. Data architecture

• Who supplies the data?
• Is the data confidential?
• Can one competitor's information influence another competitor's recommendation?

C. Governance

• Who controls the algorithm?
• Can competitors modify its parameters?
• Can competitors see how the system works?

D. Contractual arrangements

Authorities may examine agreements between:

• competing firms;
• software vendors;
• data providers;
• platforms; and
• consultants.

E. Communication records

Important evidence may include:

• emails;
• internal messages;
• software specifications;
• implementation documents;
• pricing policies;
• algorithm-development instructions.

F. Output patterns

Authorities may compare:

• prices;
• timing of changes;
• discounts;
• responses to shocks;
• margins; and
• deviations.

However, similar outputs alone should be treated cautiously, because machine-learning systems may independently produce similar results when exposed to similar market conditions.

20. Liability of the Algorithm Provider

A particularly difficult issue concerns the third-party software provider.

Possible structures include:

Neutral technology provider

The provider simply supplies software.

Facilitating intermediary

The provider collects competitors' confidential information and incorporates it into recommendations.

Coordinating hub

The provider deliberately structures the system so competitors can align their commercial behaviour.

The legal characterization depends heavily upon:

• knowledge;
• contractual arrangements;
• data flows;
• system architecture;
• communications;
• purpose;
• participation; and
• effect.

The RealPage proceedings illustrate why the intermediary itself can become central to the competition-law analysis.

21. Compliance Measures

Businesses using common algorithms should consider:

1. Independent data architecture

Competitors' confidential information should not unnecessarily influence another firm's pricing decisions.

2. Data firewalls

Competitively sensitive information should be segregated.

3. Algorithmic audits

Businesses should periodically examine:

• input data;
• output behaviour;
• optimization objectives;
• competitor-related variables.

4. Human oversight

Important strategic decisions should not automatically be delegated to systems whose recommendations depend on competitors' confidential information.

5. Competition-law clauses

Contracts with software providers should address:

• data confidentiality;
• permitted data use;
• competitor information;
• algorithmic recommendations; and
• compliance responsibilities.

6. Documentation

Companies should document legitimate business reasons for:

• adopting an algorithm;
• choosing particular variables;
• rejecting competitor information; and
• modifying recommendations.

22. A Competition-Law Test for Shared Algorithms

A useful analytical framework is:

Step 1 — Identify the algorithm

What does it actually do?

Step 2 — Identify the users

Are the users competitors?

Step 3 — Identify the data

Does the algorithm receive competitors' confidential information?

Step 4 — Identify the information flow

Can Firm A's information influence Firm B's recommendation?

Step 5 — Identify the decision-maker

Is the firm independently making the decision, or effectively outsourcing the decision to a common mechanism?

Step 6 — Identify the competitive variable

Does the system control:

• price;
• output;
• wages;
• capacity;
• discounts;
• inventory;
• advertising; or
• contract terms?

Step 7 — Examine communications

Was there any explicit or implicit communication between competitors?

Step 8 — Examine purpose

Was the system designed to:

• compete more effectively; or
• stabilize competitive conditions?

Step 9 — Examine effects

Did competitive independence materially decline?

Step 10 — Apply the appropriate legal doctrine

Possible doctrines include:

• cartel;
• concerted practice;
• information exchange;
• hub-and-spoke coordination;
• abuse of dominance;
• exclusionary conduct; or
• other applicable competition-law theories.

23. Algorithmic Homogeneity and Market Structure

The danger becomes greater where the market has:

• few competitors;
• high entry barriers;
• frequent interaction;
• transparent prices;
• standardized products;
• high switching costs;
• common data infrastructure;
• common software suppliers; or
• strong network effects.

In such markets, algorithmic systems can potentially make competitive behaviour substantially more predictable.

Conversely, where markets have:

• numerous competitors;
• differentiated products;
• volatile demand;
• low entry barriers; and
• limited information sharing,

similar algorithmic outputs may be more plausibly explained by ordinary competitive conditions.

24. Relationship with AI and Foundation Models

The issue becomes more complex with modern AI.

Imagine that competing retailers independently use the same foundation model:

Foundation model

↓

Pricing agent

↓

Retailer A

Retailer B

Retailer C

The model may learn from large-scale market information and generate recommendations that systematically push firms toward similar strategies.

This raises questions concerning:

• common model providers;
• model fine-tuning;
• shared market data;
• confidential information;
• agent-to-agent communication;
• autonomous pricing;
• model updates;
• common optimization objectives; and
• responsibility for emergent coordination.

The CMA's 2026 work specifically recognizes the growing significance of AI and agentic systems for competition and the possibility of new forms of collusion.

25. Key Doctrinal Proposition

The emerging principle can be summarized as follows:

Competition law is concerned not with algorithmic technology itself, but with whether the technology undermines the independent competitive decision-making that competition law requires.

Thus:

Algorithm + independent decisions = ordinarily legitimate

Algorithm + explicit cartel = unlawful

Algorithm + competitors' confidential information = potentially unlawful

Common algorithm + coordinated hub-and-spoke structure = potentially unlawful

Common algorithm + autonomous convergence without agreement = legally more uncertain

26. Conclusion

Competitive homogeneity induced by shared algorithms represents a significant evolution in competition-law analysis.

The traditional cartel model assumes that human beings communicate and agree:

Firm A ↔ Firm B → Agreement → Price increase

Algorithmic markets can create much more sophisticated structures:

Firm A → Algorithm ← Firm B

or:

Firm A → Common Data Pool → AI System → Firm A

Firm B → Common Data Pool → AI System → Firm B

The cases of Eturas, Trod/GB eye, Topkins and RealPage demonstrate progressively different ways in which technology can facilitate or potentially generate competitive coordination. The foundational cases such as Wood Pulp, Suiker Unie and Anic remain important because they prevent the analysis from equating mere parallel behaviour with unlawful coordination.