Hyper-Personalized Pricing Systems and Discrimination Risks

1. Introduction

Hyper-personalized pricing refers to pricing systems in which an algorithm determines the price, discount, surcharge, ranking, or commercial offer presented to an individual consumer by using highly granular information about that consumer or their context.

Traditional personalized pricing might use broad categories such as loyalty status or customer segment. Hyper-personalized pricing can go much further by combining:

• browsing and search history;
• previous purchases;
• device and operating-system information;
• location and time;
• income or inferred purchasing power;
• creditworthiness;
• willingness-to-pay estimates;
• behavioral patterns;
• demographic proxies;
• social-network information;
• responses to previous prices;
• competitor prices;
• real-time demand;
• biometric or behavioral signals; and
• predictions generated by AI models.

The competition-law concern is not simply that two consumers pay different prices. Price differentiation can be economically legitimate. The more difficult question is whether a firm with substantial market power can use individualized data and algorithmic prediction to systematically extract consumer surplus, exclude rivals, exploit vulnerable consumers, or discriminate through protected or proxy characteristics.

2. What Makes Hyper-Personalized Pricing Different?

A conventional pricing model might be:

Product price = ₹1,000 for everyone.

A segmented model might be:

Loyalty members = ₹900; non-members = ₹1,000.

A hyper-personalized system could instead calculate:

Individual price = predicted willingness to pay + predicted urgency + predicted switching cost + contextual demand adjustment.

For example, an AI system could infer that:

• Consumer A urgently needs a flight and has few alternatives;
• Consumer B is highly price-sensitive;
• Consumer C regularly purchases premium products;
• Consumer D is unlikely to compare competing offers.

The platform could then present different prices or discounts to each consumer.

The legal difficulty is that the consumer may never know that personalization occurred.

3. Main Competition-Law Risks

A. Exploitative Individualized Pricing

A dominant firm may use detailed consumer data to determine the maximum price each individual is likely to accept.

This transforms traditional price discrimination into potentially extreme forms of surplus extraction.

The concern is particularly serious where:

1. the firm has substantial market power;
2. consumers cannot realistically switch;
3. prices are opaque;
4. consumers cannot compare individualized offers;
5. the algorithm systematically exploits dependency; and
6. the firm possesses unique data unavailable to competitors.

4. First-Degree Price Discrimination

Hyper-personalization can approximate first-degree price discrimination, where a seller attempts to charge each consumer a price approaching that consumer's reservation price.

Perfect first-degree discrimination is rarely achievable in practice, but machine learning can bring firms closer to it.

Example

Suppose a dominant platform predicts:

The platform may obtain substantially more consumer surplus than under a uniform ₹1,500 price.

The competition concern therefore shifts from:

“Is the price high?”

to:

“How did the firm obtain the informational capacity to extract different amounts from individual consumers?”

5. Discrimination Through Proxies

One of the most significant risks is that the algorithm does not explicitly use a protected characteristic but uses variables that correlate strongly with it.

For example:

• postcode may correlate with ethnicity or socioeconomic status;
• device type may correlate with income;
• browsing patterns may correlate with age;
• language may correlate with nationality;
• employment history may correlate with socioeconomic status;
• purchasing patterns may correlate with disability or health conditions.

Thus:

No explicit discriminatory variable is required for discriminatory pricing to emerge.

An AI model can reproduce discriminatory outcomes through apparently neutral variables.

6. Algorithmic Opacity

Traditional price discrimination could sometimes be detected because consumers observed different prices.

Hyper-personalized systems can make discrimination difficult to identify because:

• prices change continuously;
• offers are individually generated;
• experiments are conducted in real time;
• algorithms use hundreds or thousands of variables;
• consumers receive different interfaces;
• prices may change according to predicted behavior; and
• the underlying model may be a proprietary black box.

This creates an information asymmetry between the platform and consumers.

7. Competition-Law Framework

Hyper-personalized pricing can potentially implicate several competition-law doctrines.

Article 102 TFEU

A dominant undertaking may face concerns involving:

• unfair prices or trading conditions;
• discriminatory conditions;
• exclusionary conduct;
• leveraging of data advantages;
• exploitative use of dependency.

UK Competition Law

Under the Competition Act 1998, particularly the Chapter II prohibition, individualized pricing can become relevant where a dominant undertaking uses personalization to:

• impose unfair conditions;
• discriminate between equivalent trading partners;
• exploit customers;
• foreclose competitors; or
• reinforce market power.

Article 101 / Chapter I

If competing firms use coordinated pricing algorithms or exchange individualized pricing information, the issue may become one of algorithmic coordination or collusion, rather than unilateral discrimination.

8. Important Case Laws

1. United Brands v Commission — 1978

United Brands Company v Commission, Case 27/76, is one of the foundational Article 102 cases concerning unfair pricing and discriminatory conditions.

The European Court of Justice established the framework for assessing whether a dominant undertaking imposes an unfair price.

Relevance to hyper-personalized pricing

The case provides an important foundation for asking whether an individualized price generated by a dominant platform becomes legally problematic because the pricing system enables the firm to impose unfair trading conditions.

Hyper-personalization could make the assessment considerably more complicated because the price may differ from consumer to consumer.

The relevant inquiry could therefore include:

• the economic value supplied;
• the individualized price;
• the degree of market power;
• the consumer's dependency;
• the methodology used to determine the price; and
• whether the personalization mechanism produces systematically exploitative outcomes.

2. MEO v Autoridade da Concorrência — 2017

MEO – Serviços de Comunicações e Multimédia SA v Autoridade da Concorrência, Case C-525/16, is particularly important for discriminatory pricing under Article 102(c).

The Court emphasized that not every difference in treatment between trading partners automatically constitutes an abuse. The discrimination must be capable of placing certain trading partners at a competitive disadvantage.

Importance

This principle is highly relevant to personalized pricing.

Suppose a dominant platform gives:

• Consumer group A a 20% discount;
• Consumer group B a 5% discount.

The existence of different prices alone does not necessarily establish an Article 102 infringement.

The legal question becomes whether the differentiated treatment produces a relevant competitive disadvantage.

For hyper-personalization, this means authorities may need to examine:

• magnitude of price differences;
• duration;
• affected consumers;
• competitive significance;
• downstream effects; and
• whether similarly situated transactions receive materially different treatment.

3. British Airways v Commission — 2007

British Airways plc v Commission, Case C-95/04 P, concerned a dominant airline's incentive and commission arrangements.

The Court examined whether differentiated remuneration arrangements could distort competition by placing competitors at a disadvantage.

Relevance

The case illustrates how apparently commercial discounts can become problematic when employed by a dominant undertaking in a manner capable of producing exclusionary effects.

Hyper-personalized discounts could potentially operate similarly.

A dominant digital platform might provide:

extraordinarily attractive personalized prices to consumers showing signs of switching,

while charging less favorable prices to consumers already locked into the platform.

The individualized pricing system could therefore become an anti-switching mechanism.

4. Intel v Commission — 2017

Intel Corp v Commission, Case C-413/14 P, is principally an exclusionary-abuse case involving rebates.

The Court emphasized the importance of assessing the capacity of rebates to foreclose equally efficient competitors where the undertaking contests the Commission's assessment.

Relevance

Hyper-personalized discounts can function like sophisticated individualized rebates.

An AI system might determine:

“This customer is likely to switch to Rival X.”

It could then automatically provide that customer with an unusually large discount.

Another consumer who is unlikely to switch might receive no discount.

The system therefore transforms pricing into a dynamic retention mechanism.

This creates a possible competition-law concern where individualized discounts selectively make entry or expansion by competitors more difficult.

5. Hoffmann-La Roche v Commission — 1979

Hoffmann-La Roche & Co AG v Commission, Case 85/76, is a foundational Article 102 case concerning exclusionary loyalty arrangements.

The case established the importance of assessing conduct by a dominant undertaking that can tie customers to the dominant supplier.

Relevance

Hyper-personalized pricing may create a modern form of loyalty mechanism.

Instead of explicitly saying:

“You will receive this discount only if you purchase exclusively from us,”

an algorithm could infer loyalty and continuously optimize individualized offers to reduce the likelihood of switching.

The resulting economic effect may resemble a technologically sophisticated loyalty strategy.

6. AKZO Chemie v Commission — 1991

AKZO Chemie BV v Commission, Case C-62/86, is a leading authority on predatory pricing and dominant-firm pricing strategies.

The case developed important principles concerning prices below relevant cost benchmarks and the potential exclusionary consequences of aggressive pricing.

Relevance to hyper-personalization

An algorithmically personalized system could potentially identify:

• consumers whose switching would threaten the incumbent;
• geographic areas where a rival is expanding;
• customers strategically important to a competitor;
• users who are particularly responsive to discounts.

The dominant firm could then selectively offer extremely low prices to those consumers.

Unlike conventional predatory pricing, the strategy could be micro-targeted rather than market-wide.

That makes detection substantially more difficult.

9. Additional Important Case: Deutsche Telekom v Commission

Deutsche Telekom AG v Commission, Case C-280/08 P, concerns exclusionary pricing and margin squeeze.

The case is relevant because competition law may examine pricing not merely by asking whether a price is high or low, but by considering whether the pricing structure forecloses competitors.

Hyper-personalized pricing can create similar concerns when individualized offers are strategically designed to weaken competitors.

10. Individualized Pricing and Consumer Exploitation

Competition law traditionally focuses heavily on protecting the competitive process.

But hyper-personalized pricing creates a closer relationship between competition law and consumer protection.

Consider an AI system that identifies:

• consumers with low financial literacy;
• consumers under time pressure;
• consumers who repeatedly purchase essential goods;
• consumers with high switching costs.

It could charge those consumers more because the algorithm predicts they are less likely to resist.

This creates a form of behavioral exploitation.

The concern is stronger where the individualized price is generated from characteristics that consumers cannot realistically control.

11. Hyper-Personalization and Vulnerable Consumers

AI pricing can potentially identify consumer vulnerability with extraordinary precision.

For example:

Consumer A repeatedly searches for emergency accommodation.

The system may infer urgency.

It can then increase the price presented specifically to that consumer.

The important distinction is between:

Demand-based dynamic pricing

and

individualized exploitation of vulnerability.

Dynamic pricing may reflect genuine changes in supply and demand.

Hyper-personalized exploitation instead uses information about the particular consumer to determine how much that consumer can be charged.

12. Geographic and Socioeconomic Discrimination

Location is particularly powerful as a pricing variable.

A platform may learn that consumers from certain neighborhoods generally:

• have higher purchasing power;
• have fewer alternatives;
• buy premium products;
• exhibit lower price sensitivity.

The algorithm can then adjust prices accordingly.

This creates a potential chain:

Location data → socioeconomic inference → willingness-to-pay prediction → individualized price.

Even if socioeconomic status is never explicitly entered into the model, the outcome may reproduce socioeconomic discrimination.

13. Device-Based Discrimination

Device information can also become a pricing signal.

For example, an algorithm might infer that users of certain:

• smartphones;
• operating systems;
• browsers;
• connected devices;

have greater willingness to pay.

The platform could therefore provide different offers.

The competition issue becomes particularly serious if the dominant platform systematically disadvantages consumers using particular devices or channels.

14. Dynamic Pricing Versus Hyper-Personalized Pricing

These concepts should not be confused.

Dynamic pricing

Prices change because:

• demand changes;
• supply changes;
• inventory changes;
• time changes;
• market conditions change.

Hyper-personalized pricing

Prices change because:

• this particular consumer is predicted to tolerate a particular price.

The distinction is important.

A hotel charging ₹10,000 because demand is high is not necessarily engaging in discriminatory personalization.

A hotel charging Consumer A ₹10,000 and Consumer B ₹15,000 because its AI predicts that B will tolerate the higher price is a substantially different phenomenon.

15. Algorithmic A/B Testing

Platforms may also continuously experiment with prices.

For example:

• 5% of users receive ₹900;
• 5% receive ₹950;
• 5% receive ₹1,000;
• the system measures conversion;
• the algorithm updates the optimal price.

Over time, the platform may learn each consumer's approximate reservation price.

This creates a continuous price-discovery machine.

The competition concern is that a dominant platform can acquire an informational advantage that smaller competitors cannot reproduce.

16. Data as the Source of Pricing Power

Hyper-personalized pricing can therefore be understood as a three-stage system:

Stage 1 — Data accumulation

The platform collects:

clicks + purchases + searches + location + device + behavioral history.

Stage 2 — Behavioral prediction

AI estimates:

willingness to pay + urgency + switching probability + price sensitivity.

Stage 3 — Commercial extraction

The platform determines:

individualized price + discount + ranking + promotion.

Thus:

Data advantage → prediction advantage → pricing advantage → market-power reinforcement.

This feedback loop can make the pricing system itself a source of durable dominance.

17. Feedback Loops and Entrenchment

A dominant platform can become increasingly powerful because every transaction creates more data.

The cycle becomes:

More users
↓
More behavioral data
↓
Better price prediction
↓
Higher monetization
↓
Greater investment in personalization
↓
Better consumer targeting
↓
More users

This creates a potential data-driven network effect.

Competitors with fewer users may not possess sufficient data to reproduce the same level of pricing precision.

18. Discrimination Against Competitors

Personalized pricing does not only affect consumers.

A dominant platform might personalize:

• commissions;
• access fees;
• advertising prices;
• logistics charges;
• ranking;
• API fees;
• marketplace terms.

For example, a platform could identify which merchants have no viable alternative distribution channel and charge them higher commissions.

This transforms consumer-level personalization into business-partner discrimination.

Article 102(c)-type concerns become especially relevant where equivalent transactions receive discriminatory conditions that create competitive disadvantages.

19. Self-Preferencing Through Personalized Prices

A vertically integrated platform could potentially manipulate personalized prices in favor of its own downstream services.

For example:

Third-party seller's product → ₹1,000
Platform's competing product → individualized ₹850 offer.

If the platform controls:

• consumer data;
• search ranking;
• pricing;
• recommendation;
• marketplace access;

the individualized price can become part of a broader self-preferencing strategy.

20. Algorithmic Collusion Risk

Hyper-personalization also creates an unusual possibility involving multiple firms.

Suppose competing AI pricing systems observe market conditions and independently learn that:

raising prices produces higher margins without causing substantial switching.

If algorithms repeatedly respond to one another, prices could converge toward supracompetitive levels even without a conventional human agreement.

The legal issue would depend heavily on evidence of:

• communication;
• coordination;
• algorithm design;
• information exchange;
• conscious parallelism;
• predictability of competitor responses; and
• human involvement.

The mere fact that algorithms independently produce similar prices is not automatically proof of collusion.

21. Transparency Problem

Hyper-personalized pricing creates three different transparency questions.

1. Price transparency

Does the consumer know that different consumers receive different prices?

2. Data transparency

Does the consumer know what information was used?

3. Algorithmic transparency

Does the consumer understand why the system selected that price?

A system can therefore be formally transparent about the final price while remaining opaque about the process producing the price.

22. Proof and Evidentiary Problems

Competition authorities may face significant evidentiary challenges.

They may need to establish:

1. the relevant market;
2. dominance;
3. the pricing mechanism;
4. the variables used;
5. the causal relationship between variables and price;
6. discriminatory treatment;
7. competitive disadvantage or exploitative effects;
8. foreclosure or consumer harm; and
9. absence of legitimate objective justification.

AI systems make this difficult because models may be:

• continuously retrained;
• proprietary;
• probabilistic;
• adaptive;
• distributed across several systems.

23. Defences Available to Firms

A firm accused of discriminatory personalized pricing may argue that differences reflect legitimate economic factors.

Possible justifications include:

• genuine differences in costs;
• demand fluctuations;
• inventory management;
• fraud prevention;
• loyalty programmes;
• promotional campaigns;
• different service levels;
• consumer acquisition costs;
• legitimate risk differences.

The existence of personalization therefore does not automatically establish an antitrust violation.

The legal analysis must distinguish legitimate differentiation from abusive discrimination.

24. Regulatory Tests for Hyper-Personalized Pricing

A competition authority could examine the following questions:

A. Market power

Does the undertaking possess substantial and durable market power?

B. Data advantage

Does the undertaking control data competitors cannot reasonably replicate?

C. Individualization

Are prices genuinely individualized?

D. Discriminatory variable

Are protected characteristics or strong proxies involved?

E. Competitive effect

Does the pricing system disadvantage competitors or trading partners?

F. Exploitation

Does the system systematically extract excessive consumer surplus?

G. Switching effects

Are prices strategically adjusted to prevent consumers from switching?

H. Transparency

Can consumers realistically understand and compare offers?

I. Objective justification

Can the undertaking demonstrate legitimate economic reasons for the differentiation?

25. Relationship With GDPR and Data Protection

Hyper-personalized pricing may also intersect with data-protection law.

The competition-law question is:

Does the pricing practice harm competition or constitute abusive discrimination?

The data-protection question may instead be:

Was personal data lawfully collected, processed and used for this purpose?

These are different legal inquiries.

A pricing model can therefore potentially comply with one body of law while creating problems under another.

This produces an increasingly important intersection:

Competition law + consumer protection + data protection + AI governance.

26. Theories of Harm

The principal theories of harm can be summarized as follows:

27. Six Core Cases at a Glance

28. Emerging Legal Concept: Algorithmic Price Discrimination

The traditional model of price discrimination was:

Seller knows something about the consumer.

The emerging model is:

AI predicts something about the consumer.

This distinction is crucial.

The system may never know a consumer's actual income or willingness to pay. It may simply predict it with sufficient accuracy.

Consequently, future competition law may increasingly need to examine inferential discrimination, rather than merely discrimination based on information explicitly supplied by consumers.

29. A Four-Layer Analytical Framework

A useful framework for analysing hyper-personalized pricing is:

Layer 1 — Data

What information does the undertaking possess?

Layer 2 — Inference

What characteristics does the algorithm predict?

Layer 3 — Pricing

How are those predictions converted into prices?

Layer 4 — Market effect

What happens to:

• consumers;
• competitors;
• suppliers;
• entry;
• switching;
• innovation;
• market transparency?

The fourth layer is particularly important for competition law.

30. Conclusion

Hyper-personalized pricing is not inherently unlawful. Firms may legitimately use data to provide discounts, tailor offers, manage inventory and respond to consumer demand.

The competition-law problem emerges where personalization becomes a mechanism for exploitation, exclusion, discriminatory treatment, or reinforcement of market power.

The most significant transformation is that AI can move price discrimination from broad customer groups toward individual-level continuous optimization.

The central legal concern can therefore be expressed as:

When a dominant undertaking possesses sufficiently detailed behavioral data to predict each consumer's willingness to pay, the pricing algorithm can transform information asymmetry into individualized market power.

The traditional doctrines developed in United Brands, MEO, British Airways, Intel, Hoffmann-La Roche, AKZO and Deutsche Telekom provide important foundations, but hyper-personalized AI pricing raises questions that those cases did not confront directly: proxy discrimination, continuous experimentation, opaque inference, behavioral vulnerability, data-driven dominance and individualized foreclosure.

Accordingly, future competition-law analysis is likely to move beyond the simple question “Are prices different?” toward the more sophisticated questions: