Competition Law and Machine-Driven Market Segmentation and Antitrust

1. Introduction

Machine-driven market segmentation refers to the use of algorithms, artificial intelligence (AI), machine learning, big data, and automated decision-making systems to divide consumers, suppliers, products, or geographic areas into different groups for commercial purposes.

Traditional segmentation may divide consumers according to:

age;

income;

location;

purchasing history;

product preferences.

Machine-driven segmentation can go much further. Algorithms can analyse:

browsing behaviour;

purchase history;

search activity;

device information;

location;

transaction frequency;

willingness to pay;

responsiveness to discounts;

customer loyalty;

real-time demand.

This can create significant efficiencies. However, where a firm has substantial market power, automated segmentation may also raise antitrust concerns, particularly involving price discrimination, exclusion, personalized pricing, foreclosure, exploitation, information exchange, and collusion.

The fundamental question is:

When does technologically sophisticated market segmentation become a legitimate commercial practice, and when can it be used to restrict or distort competition?

2. Meaning of Machine-Driven Market Segmentation

Machine-driven segmentation occurs when algorithms automatically identify economically relevant groups and adjust commercial strategies accordingly.

For example:

Consumer data → Machine-learning model → Consumer segment → Price/offer/product/ranking

A platform may classify consumers as:

highly price-sensitive;

premium customers;

occasional purchasers;

loyal customers;

likely repeat purchasers;

customers likely to switch;

customers with high willingness to pay.

Similarly, sellers may be classified according to:

sales volume;

reliability;

price;

delivery performance;

advertising expenditure;

customer ratings.

3. Traditional vs Machine-Driven Segmentation

The competition-law challenge is that machine segmentation can make differentiation faster, more precise, and much harder to observe.

4. Why Market Segmentation Matters to Antitrust

Segmentation can affect competition through:

Personalized pricing

Price discrimination

Customer exclusion

Geographic foreclosure

Targeted discounts

Loyalty strategies

Market allocation

Algorithmic coordination

Data-based market power

Predatory targeting

Differential access

Discriminatory platform ranking

Segmentation is therefore not inherently unlawful. The legal concern depends on how segmentation is used and its effects on competition.

5. Market Definition and Machine Segmentation

Market definition is a foundational part of competition analysis.

Machine-learning systems can reveal that consumers have very different substitution patterns.

For example:

Premium electric vehicles
Standard electric vehicles
Luxury electric vehicles

An algorithm may identify separate consumer groups with different willingness to substitute between products.

Competition authorities may therefore examine:

demand substitutability;

supply substitutability;

geographic boundaries;

consumer switching;

product characteristics;

pricing conditions.

Machine-generated consumer categories should not automatically be treated as separate legal markets.

6. Market Power Through Data

Machine segmentation requires data.

Large digital firms may possess:

millions of transactions;

detailed consumer histories;

behavioural information;

search data;

product preferences;

real-time purchasing information.

This can create a competitive advantage because the algorithm can learn more about consumer behaviour than smaller rivals.

A potential feedback loop is:

More customers → more data → better segmentation → better targeting → more customers.

However:

Data ownership alone does not establish dominance or an antitrust violation.

The competition analysis must examine whether the data advantage actually creates significant competitive barriers or contributes to exclusionary conduct.

7. Personalized Pricing

One of the most important applications of machine segmentation is personalized pricing.

Instead of:

One product → one price

the marketplace may use:

Consumer A → ₹100
Consumer B → ₹130
Consumer C → ₹160

based on predicted willingness to pay.

This may increase economic efficiency in some circumstances.

However, concerns can arise if a dominant firm uses personalized pricing to:

exploit customers;

eliminate rivals;

target vulnerable customers;

selectively undercut competitors;

prevent switching.

The legality depends heavily on jurisdiction and the specific conduct.

8. Price Discrimination

Price discrimination occurs when different customers receive different prices or conditions for reasons not fully explained by cost differences.

Machine learning makes discrimination easier because algorithms can estimate:

willingness to pay;

price sensitivity;

likelihood of switching;

urgency of purchase.

Competition concern

A dominant undertaking could potentially use targeted pricing strategically against competitors.

For example:

Customers likely to switch → aggressive discounts
Customers unlikely to switch → higher prices

This can make exclusionary strategies more difficult for competitors to detect.

9. Case Law 1: United Brands v Commission

United Brands Company v Commission, Case 27/76 (1978)

This is a foundational EU competition case involving dominance and discriminatory pricing.

The Court examined United Brands' conduct concerning different customers and markets.

Principle

Differential commercial treatment can raise competition concerns when imposed by a dominant undertaking under circumstances that amount to abuse.

Machine-segmentation relevance

An algorithm could potentially identify customer groups and automatically apply different commercial conditions.

The legal question would not be simply:

“Were different prices charged?”

It would be:

“Did the dominant undertaking use its market power in a manner that constitutes prohibited discrimination or exclusion?”

10. Case Law 2: British Airways v Commission

British Airways plc v Commission, Case C-95/04 P (2007)

The case concerned loyalty-related rebate arrangements.

The Court considered whether a dominant undertaking's incentive structure could foreclose competitors.

Machine-segmentation relevance

AI systems can create highly sophisticated loyalty segmentation.

For example:

frequent customers receive special discounts;

customers considering competitors receive targeted rebates;

highly valuable customers receive exclusive benefits.

If such a system substantially forecloses competitors, it can raise concerns analogous to traditional loyalty-rebate cases.

11. Case Law 3: Intel v Commission

Intel Corp. v Commission, Case C-413/14 P (2017)

Intel concerned conditional rebates offered by a dominant undertaking.

The judgment is particularly relevant because it emphasizes the importance of considering the economic circumstances and potential exclusionary effects of rebate practices where appropriate.

Machine-driven segmentation

An AI system could automatically identify:

“customers whose switching would most threaten the dominant firm's position.”

It could then provide those customers with highly targeted discounts.

The competition analysis may therefore need to consider:

coverage;

duration;

market position;

foreclosure;

competitors' ability to compete;

efficiencies.

12. Case Law 4: Hoffmann-La Roche v Commission

Hoffmann-La Roche & Co. AG v Commission, Case 85/76 (1979)

This is one of the leading European cases on loyalty-inducing practices by dominant firms.

The case established important principles concerning conduct capable of tying customers to a dominant undertaking.

Machine segmentation relevance

Machine learning can make loyalty strategies much more sophisticated.

An algorithm could identify:

customers with high switching probability;

customers with high lifetime value;

customers likely to purchase competing products.

The company could then target those customers with individualized incentives.

Such conduct is not automatically unlawful, but a dominant firm's targeted loyalty strategy may require competition-law scrutiny where it has exclusionary effects.

13. Case Law 5: Michelin v Commission

NV Nederlandsche Banden Industrie Michelin v Commission, Case 322/81 (1983)

Michelin involved a dominant undertaking's rebate system.

The Court examined whether the rebate arrangements could tie customers to the dominant undertaking and restrict competition.

Relevance

Machine-driven segmentation can make rebate systems substantially more complex.

Instead of a uniform rebate:

“All customers receive 5%.”

an algorithm might create:

Customer A → 3%
Customer B → 8%
Customer C → 12%

based on purchasing patterns and competitive threats.

The important competition-law question is whether the resulting system has an exclusionary effect.

14. Case Law 6: Tomra Systems v Commission

Tomra Systems ASA v Commission, Case C-549/10 P (2012)

Tomra concerned exclusionary rebate arrangements and customer segmentation in the context of a dominant undertaking.

The case is particularly relevant to strategies directed toward specific customer groups.

Machine-segmentation relevance

Modern AI can identify precisely which customers are strategically important.

A dominant firm could theoretically prioritize:

customers most valuable to rivals;

customers located in contested geographic areas;

customers likely to switch;

customers necessary for a competitor's expansion.

This can make traditional foreclosure strategies more targeted.

15. Case Law 7: AKZO Chemie v Commission

AKZO Chemie BV v Commission, Case C-62/86 (1991)

AKZO is a leading predatory-pricing case.

The Court examined pricing conduct in circumstances involving a dominant undertaking and a smaller competitor.

Machine-segmentation relevance

Machine learning could make predatory strategies more precise.

For example, a dominant firm could identify:

geographic area + vulnerable competitor + highly contested customers

and reduce prices only in those segments.

This could potentially make predatory pricing more difficult to detect because the losses are concentrated in specific segments rather than across the entire market.

However, algorithmic targeting does not itself establish predatory pricing; the relevant legal tests must still be satisfied.

16. Case Law 8: United States v Microsoft

United States v Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001)

Microsoft involved exclusionary conduct associated with control over an important technological platform.

Relevance to machine segmentation

Modern platforms may similarly use customer data to identify:

users vulnerable to switching;

developers likely to migrate;

competitors' customers;

strategic market segments.

The broader lesson is that control over a technological platform can be used to affect competition in adjacent markets.

17. Case Law 9: Google Shopping

Google and Alphabet v Commission, Case T-612/17 (2021)

Google Shopping concerned the treatment of competing comparison-shopping services within Google's search ecosystem.

Machine-segmentation relevance

Modern AI systems can combine:

segmentation + recommendation + ranking + personalization.

For example:

User profile → predicted preference → selected products → ranking.

If a dominant platform uses this process to systematically favor its own services while disadvantaging competitors, competition concerns may arise.

18. Case Law 10: Aspen Skiing

Aspen Skiing Co. v Aspen Highlands Skiing Corp., 472 U.S. 585 (1985)

Aspen Skiing is an important U.S. refusal-to-deal case.

Its relevance to machine segmentation lies in the possibility that a dominant company might use detailed information to selectively terminate or restrict dealings with particular customer or competitor groups.

Again, the case represents an exceptional category of refusal-to-deal conduct rather than a general duty to cooperate.

19. Geographic Machine Segmentation

Algorithms can divide markets geographically.

For example:

Region A → low prices
Region B → high prices
Region C → no service

Potential competition concerns can arise where geographic segmentation is used to:

exclude competitors;

allocate markets;

prevent arbitrage;

reinforce territorial restrictions.

Geographic differentiation itself is not unlawful.

The relevant question is whether the conduct has an anticompetitive purpose or effect under the applicable law.

20. Machine Segmentation and Territorial Allocation

A particularly serious concern arises where competing businesses use algorithms to divide customers geographically.

For example:

Firm A → Northern region
Firm B → Southern region

If competitors reach an agreement to allocate territories, this may constitute a form of cartel or market allocation.

The use of AI does not make traditional cartel rules disappear.

21. Algorithmic Customer Allocation

Machine systems can allocate customers automatically.

For example:

Customer → algorithm → preferred seller.

This can improve efficiency.

But if competing sellers coordinate their algorithms to divide customers, the conduct may raise serious antitrust concerns.

The crucial distinction is between:

Independent optimization

Each company independently chooses its algorithm.

and

Coordinated allocation

Competitors coordinate their systems to avoid competing for particular customers.

22. Personalized Discounts and Predatory Conduct

Personalized discounts can create a particularly difficult problem.

A dominant firm might identify customers served by a new entrant and provide them with aggressive discounts.

The strategy may be:

Competitor customer identified → personalized discount → customer retained → competitor loses scale.

The competition authority would need to determine whether the conduct satisfies the applicable test for predatory or exclusionary pricing.

Important evidence could include:

costs;

prices;

duration;

targeting;

competitor vulnerability;

internal documents;

market effects.

23. Machine Segmentation and Loyalty Schemes

Algorithms can create individualized loyalty programs.

For example:

High-value customer → premium membership
Switching customer → special discount
Loyal customer → exclusive offer

Such systems can improve customer retention.

However, when implemented by a dominant undertaking, authorities may examine whether the system:

forecloses rivals;

covers a substantial portion of demand;

creates switching costs;

conditions access to important customers;

prevents effective competition.

24. Machine Segmentation and Tying

Segmentation can also facilitate tying.

A platform might identify customers who purchase Product A and automatically offer:

Product A + Product B + mandatory platform service.

Potential competition concerns arise if the firm has market power in Product A and uses that power to foreclose competition in Product B.

The legality depends upon the applicable tying doctrine and evidence.

25. Machine Segmentation and Bundling

AI can create highly individualized bundles.

For example:

Customer profile → smartphone + cloud storage + insurance + payment service.

Bundling can produce efficiencies.

But a dominant undertaking could potentially use bundling to make it difficult for competing suppliers to access customers.

This may raise foreclosure concerns where the legal requirements for an abuse are satisfied.

26. Machine Segmentation and Self-Preferencing

Machine learning can identify which consumers are most likely to purchase a platform's own product.

The algorithm can then place that product prominently in the consumer's personalized interface.

This raises a possible self-preferencing issue.

Relevant factors include:

dominance;

market structure;

degree of preferential treatment;

competitive foreclosure;

consumer effects;

legitimate ranking criteria.

Self-preferencing should not automatically be classified as unlawful.

27. Machine Segmentation and Seller Discrimination

Marketplaces may also segment sellers rather than consumers.

For example:

Segment A

Large established sellers.

Segment B

Small sellers.

Segment C

Platform-affiliated sellers.

Segment D

New entrants.

The algorithm could assign different:

commissions;

ranking;

advertising opportunities;

data access;

delivery options.

Differential treatment may be commercially justified, but discriminatory treatment by a dominant marketplace can raise competition concerns depending on its effects.

28. Machine Segmentation and Information Exchange

Machine systems can process huge amounts of competitor information.

Suppose algorithms receive:

competitors' prices;

inventories;

customer demand;

discounts;

capacity.

This information can improve market efficiency.

But systematic sharing of competitively sensitive information between competitors may facilitate coordination.

Competition authorities may therefore need to distinguish:

legitimate market intelligence

from

information exchange that facilitates collusion.

29. Algorithmic Collusion

Machine segmentation can contribute to algorithmic coordination.

For example:

Algorithm identifies competitor's price.

Algorithm identifies customers likely to switch.

Algorithm changes its price.

Competitor's algorithm responds.

Prices converge.

This sequence is not automatically proof of a cartel.

However, where firms deliberately design systems to coordinate or implement an agreement, competition-law risks become much more serious.

30. Privacy and Competition

Machine segmentation often depends on personal data.

There can therefore be overlap between:

competition law;

privacy law;

consumer protection;

data governance.

A dominant platform's data advantage may become competitively significant where consumers cannot realistically switch without losing their data or personalized services.

Nevertheless, privacy regulation and competition law serve different purposes and should not automatically be treated as interchangeable.

31. Consumer Exploitation

Machine segmentation may also create concerns about exploitation.

An algorithm could theoretically identify consumers who are:

less price-sensitive;

dependent on a product;

unlikely to switch;

willing to pay more.

A dominant undertaking could then charge higher prices to those consumers.

Competition law in some jurisdictions can address exploitative conduct by dominant firms, including excessive or discriminatory pricing in appropriate circumstances.

However, personalized pricing is not automatically unlawful.

32. Machine Segmentation and Barriers to Entry

A new entrant may have difficulty competing because it lacks sufficient data to train its algorithms.

This can create:

data-driven entry barriers.

Existing platform:

millions of users → enormous data → sophisticated algorithm

New entrant:

few users → limited data → less accurate algorithm → fewer users.

This can become a self-reinforcing competitive cycle.

But the presence of such a cycle does not by itself establish an antitrust violation.

33. Dynamic Competition

Machine-driven markets are rapidly changing.

A firm with a large market share today may face strong competition tomorrow.

Therefore authorities may need to consider:

innovation;

potential competition;

technological change;

consumer switching;

entry possibilities;

future market developments.

This is particularly important when evaluating mergers involving AI and data-rich companies.

34. Efficiency Benefits

Machine segmentation has legitimate economic benefits.

Lower search costs

Consumers receive relevant products quickly.

Better matching

Consumers are matched with products they actually need.

Inventory efficiency

Businesses can forecast demand more accurately.

Lower marketing costs

Advertising can be directed toward interested consumers.

Innovation

Companies can develop products for specific market segments.

Better customer service

AI can identify customer needs more quickly.

Competition law should therefore avoid treating segmentation itself as suspicious.

35. Risks of Over-Regulation

Excessive restrictions on machine segmentation could:

reduce personalization;

increase search costs;

reduce innovation;

make advertising less efficient;

discourage investment in AI;

prevent legitimate price differentiation.

The appropriate focus should be on competitive harm rather than technological sophistication itself.

36. Remedies

Where unlawful conduct is established, possible remedies may include:

1. Prohibition of exclusionary pricing

Prevent targeted predatory strategies.

2. Non-discrimination

Require fair treatment of competing sellers.

3. Data portability

Allow customers to transfer relevant information.

4. Algorithmic monitoring

Require independent monitoring in appropriate cases.

5. Transparency

Provide sufficient explanation of ranking or pricing practices where legally appropriate.

6. Access remedies

Require access to relevant infrastructure where the legal requirements are satisfied.

7. Behavioural commitments

Prohibit specific exclusionary strategies.

8. Structural remedies

Reserved for exceptional circumstances where behavioural remedies are inadequate.

37. Evidence in Machine-Segmentation Cases

Competition authorities may need to examine:

algorithmic source code;

model documentation;

training data;

pricing records;

customer segmentation;

internal communications;

A/B testing;

ranking changes;

competitor response;

profit margins;

consumer switching;

market shares;

internal strategy documents.

This makes algorithmic auditing and digital evidence increasingly important in antitrust enforcement.

38. UAE Perspective

Machine-driven segmentation is increasingly relevant to UAE markets involving:

e-commerce;

fintech;

ride-hailing;

food delivery;

tourism;

digital advertising;

cloud computing;

smart-city services;

logistics;

AI platforms.

Under the UAE competition framework, the principal questions would include:

What is the relevant market?

Does the undertaking possess a dominant position?

What conduct is being implemented?

Does the segmentation restrict competition?

Is there foreclosure?

Is there objective justification?

Are efficiencies generated?

Are consumers or competitors harmed?

Are competitors realistically able to enter or expand?

The mere fact that AI is being used does not create an independent competition-law violation.

39. Important Distinctions

Machine segmentation ≠ market allocation

Independent segmentation is different from competitors agreeing to divide markets.

Personalization ≠ discrimination automatically

Different prices or offers do not automatically constitute unlawful discrimination.

Data advantage ≠ dominance automatically

Large datasets do not necessarily establish market power.

Algorithmic pricing ≠ cartel automatically

Parallel algorithmic responses do not by themselves prove an agreement.

Targeted discounts ≠ predatory pricing automatically

Predatory pricing requires application of the relevant legal and economic tests.

AI curation ≠ abuse automatically

The underlying conduct and competitive effects remain decisive.

40. Major Case-Law Revision Table

41. Quick Revision Notes

Machine-driven market segmentation = AI/algorithms automatically divide consumers, sellers or markets into commercially relevant groups.

Major antitrust concerns

personalized pricing;

discriminatory pricing;

loyalty rebates;

predatory targeting;

market allocation;

geographic foreclosure;

customer exclusion;

tying;

bundling;

self-preferencing;

algorithmic collusion;

data-driven entry barriers;

seller discrimination.

Important cases

United Brands — dominance and discriminatory treatment.

Hoffmann-La Roche — loyalty and exclusion.

Michelin — rebate systems.

British Airways — loyalty rebates.

Intel — exclusionary rebates and effects.

Tomra — customer foreclosure.

AKZO — predatory pricing.

Microsoft — platform exclusion.

Google Shopping — preferential treatment.

Aspen Skiing — exceptional refusal to deal.

42. Conclusion

Machine-driven market segmentation is not inherently anticompetitive. It can substantially improve consumer choice, product matching, inventory management, pricing efficiency and innovation.

The competition-law problem arises when a powerful undertaking uses automated segmentation as a strategic instrument of exclusion or exploitation.

The most important risks include:

targeted exclusion + personalized pricing + loyalty strategies + data advantages + foreclosure + algorithmic coordination.

Modern antitrust analysis must therefore examine not merely the existence of an algorithm, but what the algorithm does, who controls it, what market power exists, how competitors are affected, whether consumers can switch, and whether legitimate efficiencies or objective justifications exist.

The long-term challenge is to preserve the efficiency of AI-based segmentation while preventing dominant firms from turning granular consumer and market information into a mechanism for permanent competitive foreclosure.