Mandatory Unbundling of Personalization Engines

1. Introduction

Mandatory unbundling of personalization engines refers to a competition-law or digital-regulation remedy requiring a dominant digital platform to separate its personalization function—such as recommendation, ranking, profiling, targeting, feed optimization, or personalized search—from the platform's core service.

A personalization engine typically uses user data, behavioural signals, contextual information and machine-learning models to determine:

• what products or services a user sees;
• the order in which they appear;
• which advertisements are displayed;
• which sellers or suppliers receive visibility;
• what content is recommended;
• which search results are ranked;
• what prices, offers or terms are presented.

The competition concern arises when the same undertaking controls both the underlying platform and the algorithm that determines access to that platform. Unbundling could therefore require the platform to permit an independent or user-selected personalization layer rather than forcing users to accept the dominant firm's integrated algorithm.

The remedy is particularly significant in digital markets because personalization can create data advantages, behavioural lock-in, self-preferencing, switching costs and feedback loops.

2. Meaning of Personalization-Engine Unbundling

Unbundling can take several forms.

A. Functional unbundling

The platform remains under common ownership, but personalization is operated through a technically and organisationally separated unit.

B. Algorithmic unbundling

The dominant platform must expose standardized interfaces through which independent recommendation or ranking systems can operate.

C. Choice-based unbundling

Users must be allowed to select:

• the platform's default algorithm;
• a third-party algorithm;
• a non-personalized chronological system; or
• another permitted ranking methodology.

D. Data-portability-based unbundling

Users can transfer relevant personal or behavioural data to an alternative personalization provider.

E. Structural unbundling

In the strongest form, the personalization business itself is separated from the platform.

3. Why Personalization Engines Create Competition Concerns

3.1 Data feedback loops

A dominant platform can use its large user base to collect behavioural data.

More data → better personalization → more engagement → more users → more data.

This can create a self-reinforcing competitive advantage.

3.2 Foreclosure of competitors

An integrated platform can give its own personalization system preferential access to:

• APIs;
• user data;
• real-time behavioural signals;
• platform infrastructure;
• search history;
• transaction information;
• device-level information.

Independent personalization providers may therefore be unable to compete on equal terms.

3.3 Self-preferencing

A platform may design its recommendation system to favour its own:

• products;
• services;
• advertising inventory;
• payment systems;
• content;
• marketplaces;
• affiliated businesses.

The personalization engine consequently becomes a mechanism for leveraging dominance from one market into another.

3.4 Consumer lock-in

Personalization creates individualized environments.

A consumer who has accumulated:

• preferences;
• recommendation histories;
• playlists;
• social graphs;
• shopping profiles;
• viewing histories;
• search histories

may find it difficult to move to another service.

Unbundling can therefore reduce the importance of the platform's proprietary personalization layer as a source of switching costs.

4. Competition-Law Theories Supporting Unbundling

4.1 Abuse of dominance

Under Article 102 TFEU-type frameworks and comparable national provisions, mandatory unbundling may become relevant where personalization is used to exclude competitors.

Potential theories include:

• refusal to supply;
• discriminatory access;
• tying;
• leveraging;
• self-preferencing;
• exclusionary discrimination;
• exploitation of data advantages.

4.2 Essential-facility logic

An especially important question is whether certain personalization inputs constitute an indispensable facility.

For example, if a dominant platform possesses uniquely valuable behavioural data that independent recommendation providers cannot reasonably reproduce, competition authorities may consider whether access should be mandated.

However, courts traditionally impose a high threshold before compelling access to privately controlled assets.

4.3 Tying

The platform may effectively tell consumers:

access to the platform necessarily requires use of the platform's personalization engine.

This can produce a form of technological tying.

Unbundling would separate:

Platform service + personalization service

into two potentially contestable components.

5. Mandatory Unbundling as a Remedy

A competition authority could impose several obligations.

5.1 API access

The platform must provide APIs allowing competing personalization providers to interact with the platform.

5.2 Data access

Subject to privacy safeguards, relevant data must be made available in interoperable formats.

5.3 Algorithmic neutrality

The platform cannot disadvantage an independent recommendation engine.

5.4 Equal treatment

Third-party personalization providers must receive equivalent technical access.

5.5 User choice

Users must be able to change their personalization provider.

5.6 Interoperability

The alternative personalization engine must be able to perform the same core functions without artificial technical restrictions.

5.7 Auditability

The platform may be required to maintain logs demonstrating that it has not degraded competing engines.

6. Six Important Case Laws

The exact legal concept of "mandatory unbundling of personalization engines" is relatively novel. Consequently, the strongest authorities are cases concerning platform tying, interoperability, self-preferencing, essential facilities, data advantages and structural remedies rather than cases ordering a personalization engine to be separated as such.

Case 1: United States v. Microsoft Corp. (2001)

The Microsoft litigation is one of the most important precedents for technological bundling and platform leverage.

Microsoft integrated Internet Explorer with Windows and used its operating-system position to disadvantage competing browsers.

The case demonstrated that:

• technological integration can have exclusionary consequences;
• control of a platform can be leveraged into an adjacent market;
• apparently integrated product functionality may raise competition concerns where integration forecloses rivals.

Relevance to personalization

A dominant platform could similarly argue that its personalization engine is simply an inseparable feature of its platform.

The Microsoft precedent demonstrates that technical integration is not automatically immune from competition scrutiny.

Mandatory separation may therefore be considered where integration materially forecloses competing personalization providers.

Case 2: European Commission v. Google Android (2018)

The Google Android decision concerned Google's conduct involving Android devices, Google Search and related applications.

The Commission examined practices involving:

• tying;
• licensing conditions;
• default placement;
• anti-fragmentation arrangements.

The case is particularly relevant because defaults and distribution arrangements can influence consumer behaviour even where consumers technically retain the ability to choose alternatives.

Relevance

A personalization engine may become competitively entrenched through:

• default installation;
• default activation;
• preferential access;
• contractual restrictions;
• technical integration.

The lesson is that formal consumer choice may not constitute effective competitive choice if the dominant platform controls the architecture through which alternatives must compete.

Case 3: Google Shopping — European Commission (2017)

The Google Shopping case concerned the preferential treatment of Google's comparison-shopping service within its general search results.

Google's search engine controlled an important access point for consumers, while its ranking mechanism affected visibility.

The Commission found an abuse involving preferential positioning and display of Google's own comparison-shopping service.

Relevance to personalization engines

This is particularly important for algorithmic personalization.

A dominant platform that controls the ranking mechanism can influence competitive outcomes without expressly excluding competitors.

An unbundling remedy could therefore seek to prevent the platform from simultaneously acting as:

1. infrastructure provider;
2. ranking provider; and
3. competitor.

The personalization engine becomes problematic when it functions as a competitive gatekeeper.

Case 4: Google Search (Shopping) — Court of Justice of the European Union, 2024

The CJEU's judgment concerning Google's comparison-shopping practices reinforced the significance of discriminatory treatment within a dominant platform's search infrastructure.

The broader principle is that conduct involving a dominant platform's infrastructure cannot be evaluated merely by asking whether rivals technically remain present.

The competitive question is whether the platform's conduct disadvantages competitors through the manner in which access and visibility are structured.

Relevance

For personalization engines, this supports scrutiny of:

• ranking discrimination;
• algorithmic demotion;
• preferential recommendation;
• personalized visibility;
• differential treatment of rivals.

Thus, an independent personalization engine may require non-discriminatory access to the platform's relevant infrastructure.

Case 5: Slovak Telekom v European Commission (2021)

In Slovak Telekom, the CJEU considered exclusionary conduct involving access to infrastructure and the circumstances in which refusal or restriction of access can constitute an abuse.

The case is significant for the relationship between:

• infrastructure control;
• access conditions;
• foreclosure;
• indispensability;
• competition between downstream providers.

Relevance

A personalization engine can increasingly resemble a competitive infrastructure layer.

If an independent service cannot effectively compete because the dominant platform controls the technical interface through which recommendations reach consumers, access remedies may become relevant.

The case therefore provides a useful analytical framework for determining when controlled infrastructure can become a competition bottleneck.

Case 6: Bronner v Mediaprint (1998)

Oscar Bronner GmbH & Co. KG v Mediaprint is a foundational European essential-facilities case.

The CJEU established a demanding test for compulsory access to infrastructure controlled by a dominant undertaking.

The Court emphasized factors including:

• indispensability;
• elimination of effective competition;
• absence of a viable alternative;
• practical feasibility of access.

Relevance

Bronner is especially important because mandatory unbundling is an intrusive remedy.

A competition authority cannot simply say:

"This algorithm is useful to competitors."

It would need to establish why access or separation is sufficiently necessary to preserve effective competition.

The case therefore acts as a limiting principle against excessive algorithmic unbundling.

Case 7: IMS Health v NDC Health (2004)

In IMS Health, the CJEU considered refusal to license intellectual property and the circumstances under which compulsory access could be required.

The Court identified stringent conditions for intervention, including circumstances in which refusal could eliminate competition in a secondary market and prevent the emergence of a new product or service.

Relevance

Personalization algorithms may contain:

• proprietary models;
• trade secrets;
• training methodologies;
• protected databases;
• intellectual property.

Mandatory unbundling therefore raises an important conflict between:

competition → access

and

innovation → proprietary control.

IMS Health provides an important framework for balancing those interests.

Case 8: United States v. Google LLC — Search and Search Advertising

The modern U.S. Google search litigation is particularly relevant to digital distribution and default arrangements.

The litigation examined Google's conduct concerning distribution agreements and mechanisms through which Google Search obtained privileged access to users.

Relevance

Personalization engines can be strengthened by default status.

Even where competing algorithms technically exist, the dominant platform can maintain substantial market power if:

• its algorithm is pre-installed;
• its recommendation engine is automatically activated;
• alternatives require multiple configuration steps;
• the platform controls the relevant default interface.

Therefore, effective unbundling may require meaningful user choice rather than nominal choice.

7. Structural Versus Behavioural Unbundling

A competition authority should normally consider the least restrictive effective remedy before imposing structural separation.

8. Privacy Complications

Unbundling cannot simply mean handing every user's behavioural data to competitors.

Personalization data may contain:

• sensitive preferences;
• location information;
• browsing history;
• purchasing behaviour;
• inferred characteristics;
• social connections.

Consequently, competition law must interact with data-protection principles.

A viable model may involve:

User-controlled data portability → privacy-preserving interface → competing personalization engine

rather than unrestricted transfer of raw personal data.

9. Trade-Secret Problems

Mandatory algorithmic unbundling may also expose proprietary technology.

The platform could legitimately argue that disclosure of:

• source code;
• model weights;
• training datasets;
• ranking parameters;
• optimization techniques

would destroy trade-secret protection.

Therefore, regulators should distinguish between:

Disclosure of source code

Usually extremely intrusive.

Functional interoperability

Potentially much less intrusive.

Output-based testing

May allow regulators to detect discriminatory behaviour without revealing the algorithm.

Secure regulatory access

Auditors can inspect algorithms under confidentiality protections.

This makes functional separation and API interoperability potentially preferable to wholesale source-code disclosure.

10. Economic Effects

Positive effects

Mandatory unbundling can:

• lower entry barriers;
• increase algorithmic competition;
• reduce platform self-preferencing;
• improve innovation;
• reduce lock-in;
• increase consumer choice;
• encourage competing recommendation models;
• weaken data-network effects.

Possible negative effects

It may also:

• reduce personalization quality;
• increase cybersecurity risks;
• create privacy risks;
• weaken incentives to develop algorithms;
• increase regulatory costs;
• facilitate manipulation of recommendation systems;
• fragment platform functionality.

Therefore, unbundling should not automatically be presumed beneficial.

11. The "Personalization Bottleneck" Theory

A useful competition-law concept is the personalization bottleneck.

The theory can be represented as:

Large user base

↓

Massive behavioural dataset

↓

Superior personalization model

↓

Higher engagement

↓

More users and transactions

↓

More behavioural data

↓

Greater algorithmic advantage

The resulting feedback loop can make the personalization engine a source of durable market power.

If competitors cannot reproduce the data and feedback necessary to challenge the engine, the personalization layer can become a competitive bottleneck.

12. When Should Unbundling Be Ordered?

A strong legal framework would require several findings:

1. Dominance

The undertaking must possess substantial market power.

2. Strategic importance

Personalization must materially influence competition.

3. Foreclosure

The integrated personalization system must disadvantage competitors.

4. Causation

The harm must result from the integration or discriminatory operation of personalization.

5. Insufficient alternatives

Ordinary behavioural remedies must be inadequate.

6. Proportionality

Unbundling must be reasonably necessary to restore effective competition.

7. Privacy safeguards

The remedy must comply with applicable data-protection requirements.

8. Innovation safeguards

The remedy should avoid unnecessarily destroying legitimate incentives to develop better algorithms.

13. Possible Regulatory Model

A sophisticated regulatory regime could establish a Personalization Neutrality Framework:

1. Users own/control portable preference data.
2. Platforms provide standardized APIs.
3. Independent algorithms can access permitted data.
4. The platform cannot discriminate against competing engines.
5. Users receive a genuine choice of personalization provider.
6. Personalization providers are independently auditable.
7. Sensitive data remains protected.
8. Algorithmic manipulation is prohibited.
9. The platform must maintain technical logs.
10. Persistent violations may justify functional or structural separation.

14. Key Legal Principles Emerging from the Case Law

The cases collectively suggest six important principles:

Principle 1 — Integration is not automatically unlawful

Microsoft demonstrates that technological integration must be examined through its competitive effects.

Principle 2 — Defaults can determine competitive outcomes

Android demonstrates the importance of default arrangements.

Principle 3 — Algorithms can constitute competitive infrastructure

Google Shopping demonstrates that ranking and visibility mechanisms can materially affect competition.

Principle 4 — Access remedies require justification

Bronner limits compulsory access to situations satisfying demanding conditions.

Principle 5 — Intellectual property cannot automatically defeat competition law

IMS Health demonstrates that proprietary rights may sometimes be subject to competition-law intervention.

Principle 6 — Infrastructure discrimination can generate foreclosure

Slovak Telekom illustrates the importance of access conditions and infrastructure control.

15. Conclusion

Mandatory unbundling of personalization engines represents a potentially powerful next-generation digital competition remedy.

Its central premise is that a dominant platform should not necessarily be permitted to control simultaneously:

the marketplace + the user data + the recommendation algorithm + the competitive visibility of rivals.

However, complete structural separation should generally be reserved for situations in which less intrusive measures—such as interoperability, data portability, API access, non-discrimination, algorithmic choice and independent auditing—cannot restore effective competition.

The most important legal challenge is therefore to establish when personalization ceases to be merely a product feature and becomes a competitive bottleneck.

The combined lessons of Microsoft, Google Android, Google Shopping, Slovak Telekom, Bronner and IMS Health provide a useful doctrinal foundation for that analysis, even though none of these cases involved an order literally requiring a dominant platform to spin off a "personalization engine."