Agent-Based Consumer Modeling and Demand Preemption in Europe

1. Meaning

Agent-Based Consumer Modeling (ABCM) means using computer models in which individual consumers, households, firms, or other market participants are represented as separate “agents.” Each agent may have characteristics such as:

purchasing history;

income or estimated purchasing power;

location;

preferences;

browsing behaviour;

price sensitivity;

loyalty;

switching probability;

reaction to advertising;

reaction to competitors.

The system then simulates how those agents may behave in the future.

Demand preemption occurs when a business uses such predictive information to anticipate consumer demand and takes action before consumers actually express that demand.

For example, a dominant platform may predict that a consumer is likely to buy a particular product next week and then:

reserve inventory;

alter the product's ranking;

increase targeted advertising;

change the consumer's personalised price or offer;

restrict competing offers;

use the prediction to favour its own product; or

acquire or lock in the relevant consumer before competitors can respond.

There is currently no mature EU case-law category specifically called “agent-based consumer modeling and demand preemption.” The legal analysis therefore has to be constructed from established competition, GDPR, consumer-protection and digital-platform jurisprudence.

2. Why the issue is legally important

The important legal question is not simply:

“Is a company allowed to predict consumer demand?”

Generally, prediction itself is not unlawful.

The legal problem arises when predictive modeling becomes a mechanism for:

exclusion of competitors;

discriminatory treatment;

exploitation of personal data;

manipulation of consumer choice;

personalised exclusion;

artificial scarcity;

discriminatory pricing;

self-preferencing;

tying or bundling;

foreclosure;

exploitative conduct;

or unlawful automated decision-making.

A predictive model can therefore become a competitive weapon.

3. Basic example

Suppose Platform A operates a large online marketplace.

Its agent-based model predicts:

Consumer X has an 85% probability of purchasing Product Y within 10 days.

Platform A could legitimately use this information for inventory planning.

But suppose Platform A also:

suppresses competing products;

promotes its own substitute;

gives Product Y preferential placement;

increases advertising exposure only for its own product;

prevents competing sellers from accessing relevant behavioural information;

charges different consumers different prices based upon predicted willingness to pay.

The issue is no longer simply forecasting.

It potentially involves:

data → prediction → consumer targeting → market foreclosure → competitive harm.

4. European legal framework

Several areas of EU law can overlap.

A. Article 101 TFEU

Article 101 becomes relevant where businesses coordinate their conduct.

Agent-based models could facilitate:

algorithmic coordination;

information exchange;

common pricing strategies;

demand forecasting agreements;

coordinated allocation of consumers.

The crucial question is whether firms are independently using algorithms or whether their use reflects an agreement or concerted practice.

B. Article 102 TFEU

Article 102 is particularly important where the predictive system belongs to a dominant undertaking.

Potential theories include:

self-preferencing;

discriminatory access;

refusal to supply data;

leveraging;

tying;

exclusionary rebates;

margin squeeze;

discriminatory pricing;

personalised exclusion of competitors.

The Commission's 2026 Article 102 Guidelines now expressly provide a contemporary framework for assessing exclusionary conduct by dominant firms. (Competition Policy)

Importantly, dominance alone is not prohibited. The conduct must constitute an abuse.

C. GDPR

Agent-based consumer models may process:

browsing data;

purchase histories;

location;

inferred preferences;

behavioural profiles;

probabilities concerning future behaviour.

This makes profiling, lawful basis, transparency, purpose limitation, data minimisation and automated decision-making particularly important.

Article 22 GDPR may become relevant where profiling effectively produces an automated decision with legal or similarly significant effects.

D. Digital Services Act

For very large online platforms, recommender-system transparency and user choice become important.

The Commission's recent proceedings concerning Shein, for example, include scrutiny of recommender-system transparency and the requirement to provide users with an accessible non-profiling option for each recommender system. (Digital Strategy)

E. AI Act

The AI Act is increasingly relevant where agent-based consumer models are implemented through AI systems.

As of 2026, certain AI Act obligations and enforcement mechanisms are already applicable. The Act also prohibits specified AI practices involving manipulation or exploitation of vulnerabilities. (Digital Strategy)

However, not every demand-prediction model is automatically a high-risk AI system.

The legal classification depends on what the system does and the applicable AI Act category.

5. Main legal risks

5.1 Demand prediction becoming exclusionary

A dominant platform may predict demand more accurately because it possesses data unavailable to competitors.

It can then use that predictive advantage to:

anticipate emerging competitors;

identify products likely to become successful;

copy those products;

pre-empt suppliers;

reserve scarce distribution capacity;

change rankings before competitors gain traction.

This creates a possible data-driven foreclosure strategy.

5.2 Demand preemption through self-preferencing

Suppose a platform's model predicts that consumers will increasingly search for Product Z.

Before competing sellers become established, the platform:

increases visibility of its own Product Z;

reduces competitor visibility;

gives its own product preferential recommendations;

uses marketplace data to optimise its own inventory.

This resembles the economic logic examined in Google Shopping.

6. Case Law

Case 1 — Google Shopping

Google and Alphabet v Commission, C-48/22 P

This is one of the most important analogies.

The Google Shopping litigation concerned Google's treatment of competing comparison-shopping services and its own comparison-shopping service.

The Court confirmed the relevance of the way a dominant platform uses its general search infrastructure to favour its own specialised service. The underlying conduct involved algorithms that affected the visibility and traffic received by competing services. (curia)

Relevance to agent-based modeling

An agent-based demand system could similarly predict:

“Consumers are likely to prefer this product category.”

If the dominant platform then uses its algorithmic infrastructure to favour its own offering, the predictive system becomes part of an exclusionary strategy.

Principle

Prediction + algorithmic control + dominance + exclusionary effect = potential Article 102 problem.

This is an analogical authority, not a case specifically about agent-based consumer modeling.

7. Case 2 — Meta Platforms, C-252/21

In Meta Platforms, the Court examined the relationship between competition law and GDPR compliance.

The case concerned the combination of data from Facebook and other sources and the German competition authority's intervention concerning Meta's data-processing practices.

The Court recognised that a competition authority may examine GDPR-related issues when assessing an abuse of dominance, while requiring appropriate cooperation with the competent data-protection authorities. (curia)

Relevance

Agent-based consumer models often depend upon extensive data aggregation.

For example:

Facebook activity + website activity + purchasing behaviour + location + inferred preferences → consumer agent profile.

The competition-law analysis cannot necessarily ignore the data-protection dimension.

Principle

Data exploitation can have both competition-law and data-protection significance.

8. Case 3 — SCHUFA, C-634/21

In SCHUFA Holding (Scoring), the Court considered automated scoring under Article 22 GDPR.

SCHUFA generated a probability value concerning an individual's ability to meet financial obligations, which could then be used by third parties in decision-making. The Court's judgment concerned the circumstances in which automated scoring can fall within Article 22. (Infocuria)

Relevance

This is particularly important for agent-based modeling because the model may not directly make the final decision.

Instead:

Model → probability score → human/business decision.

The intermediary predictive score can nevertheless be legally significant.

Example

An e-commerce platform predicts:

“Consumer A has a 92% probability of purchasing a premium product.”

The platform then automatically determines:

which products appear;

what discount is offered;

whether a consumer receives a promotion;

which sellers are displayed.

The legal analysis must examine whether the prediction is merely analytical or becomes part of an automated decision producing significant effects.

9. Case 4 — Orange România, C-61/19

In Orange România, the Court examined whether a consumer had genuinely given valid consent to personal-data processing.

The Court rejected the idea that a pre-ticked consent mechanism could establish valid consent in the circumstances considered. It emphasised that consent must satisfy the relevant requirements of being freely given, specific and informed. (curia)

Relevance

Agent-based demand prediction frequently depends upon behavioural data.

If a company obtains the data through:

forced consent;

misleading interfaces;

bundled consent;

unclear profiling disclosures;

the predictive model may inherit a data-law defect.

Principle

A sophisticated prediction model does not cure an unlawful method of obtaining the underlying data.

10. Case 5 — Österreichische Post, C-300/21

In Österreichische Post, the Court addressed compensation under Article 82 GDPR.

The Court held that a GDPR infringement alone does not automatically establish entitlement to compensation; the claimant must establish an infringement, damage and a causal relationship between them. (Infocuria)

Relevance to demand preemption

Imagine a predictive system incorrectly categorises a consumer and consequently:

excludes the consumer from offers;

repeatedly targets the consumer with unwanted advertising;

produces a harmful profile;

causes a financial or reputational loss.

A claimant would still have to establish the relevant GDPR infringement, legally recognisable damage and causation.

Principle

Prediction error ≠ automatic damages.

The claimant must connect the unlawful processing to actual compensable harm.

11. Case 6 — Google Android, C-738/22 P

In Google and Alphabet v Commission, C-738/22 P, decided on 2 July 2026, the Court addressed Google's Android-related contractual restrictions and Article 102 TFEU.

The case concerned contractual restrictions, tying, exclusionary effects and the relevance of counterfactual analysis. (curia)

Relevance

Demand preemption can occur inside an ecosystem.

For example:

consumer prediction → operating-system-level targeting → preferred application → preferred transaction.

If control over one part of the ecosystem is used to restrict competing services in another market, the analysis can resemble the leveraging concerns addressed in Android.

Principle

The important question is not merely whether the system improves the dominant firm's product, but whether the contractual or technical structure has exclusionary effects in connected markets.

12. Case 7 — Heureka Group v Google, C-605/21

Heureka Group v Google, C-605/21, concerned follow-on damages arising from Google's comparison-shopping conduct.

The Court's 18 April 2024 judgment addressed issues concerning national damages actions, limitation periods and the effect of Commission competition proceedings. (Infocuria)

Relevance

Agent-based demand preemption could eventually produce private claims where competitors allege:

lost customers;

reduced traffic;

lost sales;

lower market share;

increased acquisition costs.

Therefore, the problem is not confined to regulatory enforcement.

A competitor may potentially pursue a private damages action, subject to the applicable national and EU rules.

13. Case 8 — Google Shopping, T-612/17

The General Court's Google Shopping judgment found that Google's treatment of comparison-shopping results involved favouring Google's own specialised service and reducing traffic to competing services.

The Court examined the competitive effects of the algorithmic treatment of competing services. (Infocuria)

Relevance

This provides a useful bridge between:

algorithmic prediction → ranking → consumer attention → demand allocation.

The platform does not necessarily need to prevent a competitor from operating.

It may instead influence where consumers look and what they buy.

That is especially important for agent-based consumer modeling.

14. Direct and analogical authorities

Important: none of these cases establishes a general judicial rule specifically titled “agent-based consumer modeling and demand preemption.” They are authorities from which the legal principles can be applied to that emerging technology.

15. How demand preemption could become anti-competitive

A. Competitor pre-emption

A dominant platform predicts that a small rival is becoming popular.

It then:

copies the rival's product;

increases its own promotion;

restricts the rival's access;

offers selective discounts.

The predictive model may become evidence of a deliberate exclusionary strategy.

B. Consumer-specific pre-emption

The system predicts:

Consumer A is unlikely to compare prices.

The platform therefore shows fewer competing offers.

Consumer B is predicted to be highly price-sensitive.

The platform shows more competitive offers.

This creates a potentially important question concerning personalised treatment.

C. Inventory pre-emption

A platform predicts future demand and purchases a disproportionately large share of scarce supply.

Competitors then cannot obtain adequate inventory.

Potential issues include:

exclusion;

foreclosure;

discriminatory access;

strategic purchasing;

contractual restrictions.

The economic effect matters more than the mere existence of predictive technology.

16. Demand preemption and algorithmic pricing

Agent-based systems can estimate each consumer's:

reservation price;

switching probability;

price sensitivity;

likelihood of accepting a discount.

The company can then potentially implement personalised pricing.

This creates two different legal questions.

Consumer-law question

Was the consumer adequately informed?

Competition-law question

Is the practice part of an exclusionary or exploitative strategy by a dominant firm?

Data-protection question

Was the profiling lawful?

These questions can coexist.

17. Consumer manipulation

A particularly sensitive scenario occurs where the system does not merely predict demand but attempts to create it.

For example:

Model predicts consumer interest → system identifies psychological vulnerability → interface repeatedly targets consumer → consumer purchases product.

The distinction becomes:

Demand prediction

versus

Demand engineering/manipulation.

The AI Act is relevant to certain manipulative or exploitative AI practices, including practices that can manipulate people or exploit vulnerabilities in prohibited circumstances. (Digital Strategy)

Therefore, an agent-based model should not be analysed only as a mathematical forecasting tool when it is integrated with behavioural manipulation.

18. DSA and recommender systems

Large platforms can use agent models to determine:

which product appears first;

which advertisement appears;

which seller is recommended;

what content is shown;

how frequently an offer is displayed.

The DSA therefore becomes relevant to recommender-system transparency.

The Commission's 2026 Shein proceedings specifically identify recommender-system transparency and non-profiling alternatives as regulatory issues. (Digital Strategy)

This demonstrates the increasing legal significance of how algorithmic recommendations influence consumer choice, rather than merely whether an algorithm exists.

19. Evidence problems

Agent-based systems create unusual evidentiary problems.

A claimant may need to establish:

what data entered the model;

what assumptions were used;

how agents were classified;

what prediction was generated;

what algorithmic action followed;

whether competitors were treated differently;

whether consumer behaviour actually changed;

whether the conduct caused economic damage.

Relevant evidence may include:

model documentation;

audit logs;

source data;

feature weights;

A/B tests;

ranking histories;

pricing records;

recommendation logs;

internal communications;

counterfactual simulations.

20. Causation

Causation is particularly difficult.

Suppose a competitor loses 10% of its sales.

It cannot simply argue:

“The platform had an agent-based model, therefore the model caused our loss.”

It would normally need evidence connecting:

predictive model → business conduct → consumer response → competitive harm → monetary loss.

This is similar to the broader causation requirements recognised in GDPR damages jurisprudence, where infringement, damage and causal connection must be established. (Infocuria)

21. Counterfactual analysis

Counterfactual analysis becomes extremely important.

A competition authority or court might ask:

What would consumer demand have looked like if the platform had not used the predictive system in the challenged way?

Possible comparisons include:

Scenario A

Normal independent recommendation.

Scenario B

Platform's predictive model.

Scenario C

Competitor-accessible data model.

Scenario D

Non-personalised recommendation.

The difference between scenarios can help establish competitive effects.

22. Economic dependence

A smaller seller may become dependent upon a platform because the platform possesses superior demand-prediction technology.

The seller may therefore have no realistic alternative to:

supplying the platform;

accepting its ranking rules;

sharing data;

accepting commission structures;

using its advertising system.

But commercial dependence does not automatically establish Article 102 dominance.

Market definition and dominance still have to be established.

23. Civil liability

A victim could potentially pursue different legal routes depending on the facts:

Consumer

GDPR claim;

consumer-protection claim;

unfair commercial-practice claim;

contractual claim.

Competitor

Article 102 damages;

national competition law;

contractual remedies;

tort/delict;

unfair competition.

Business partner

breach of contract;

discrimination under contractual terms;

wrongful termination;

abuse of contractual rights where recognised under national law.

24. Hypothetical example

Imagine Platform X, controlling a major European marketplace.

Its agent-based system predicts:

70% of consumers in Region A will purchase electric bicycles within six months.

Platform X then:

obtains additional supply before competitors;

uses marketplace seller data to identify successful models;

launches its own competing bicycle;

gives its own product preferential ranking;

targets likely purchasers with personalised advertisements;

reduces visibility of competing sellers;

charges different commissions to different sellers.

Possible legal questions

Article 102

Is Platform X dominant?

Exclusionary conduct

Does the conduct disadvantage equally efficient competitors?

Data

Was seller and consumer data lawfully processed?

Consumer law

Were consumers adequately informed?

DSA

Were recommender systems sufficiently transparent?

AI Act

Does the particular AI implementation fall within applicable AI Act requirements?

Damages

Can affected sellers demonstrate causation and loss?

25. Legal test

A useful European legal analysis can follow this sequence:

Step 1 — Identify the model

What exactly does the agent-based system predict?

Step 2 — Identify the data

What personal and non-personal information feeds the model?

Step 3 — Identify market power

Does the undertaking have dominance or significant platform power?

Step 4 — Identify the conduct

Is the prediction merely forecasting, or is it used to manipulate access, rankings, prices, supply or competitors?

Step 5 — Identify the mechanism

Does the model produce:

self-preferencing?

discrimination?

tying?

exclusion?

refusal of access?

personalised pricing?

strategic foreclosure?

Step 6 — Test competitive effects

Are competitors actually or potentially excluded?

EU case law recognises that an exclusionary abuse need not necessarily achieve complete exclusion before Article 102 becomes applicable. (Infocuria)

Step 7 — Examine justification

Does the undertaking have a legitimate efficiency or objective justification?

Step 8 — Examine consumer/data law

Was profiling lawful and transparent?

Step 9 — Establish causation

Can the claimant connect the predictive system to the alleged harm?

Step 10 — Calculate damages

Possible losses include:

lost sales;

lost traffic;

lost customers;

increased advertising costs;

lost market opportunities;

price discrimination losses.

26. Key distinction

27. Overall legal position

Agent-based consumer modeling is not inherently unlawful. It can produce legitimate efficiencies in forecasting, inventory management, advertising and logistics.

The legal risk arises when predictive capability is combined with market power, personal-data exploitation, algorithmic control or exclusionary conduct.

The most important legal pathway is therefore:

Consumer data → agent profile → demand prediction → automated intervention → altered consumer choice → competitive/consumer harm.

The strongest existing authorities are not cases expressly about agent-based consumer models. Rather, Google Shopping provides the strongest competition-law analogy for algorithmic allocation of consumer attention; Meta links data processing with competition law; SCHUFA addresses automated profiling; Orange România addresses valid consent; Österreichische Post addresses GDPR damages and causation; and Google Android demonstrates how ecosystem control can generate exclusionary effects. (Infocuria)

Exam Keywords

Agent-based modeling — consumer agents — demand forecasting — demand preemption — predictive analytics — profiling — personalised pricing — recommender systems — algorithmic manipulation — self-preferencing — foreclosure — data advantage — Article 101 TFEU — Article 102 TFEU — GDPR Article 22 — DSA — AI Act — causation — counterfactual — private damages — consumer autonomy — market power.