Digital Investment Platform Recommendation Bias Risks

Introduction

Digital investment platforms include online brokerages, robo-advisers, investment apps, portfolio-management platforms, comparison websites, digital wealth managers, and algorithmic recommendation systems. These platforms increasingly use artificial intelligence, machine learning, behavioural data, transaction histories, and personalised interfaces to recommend securities, funds, portfolios, trading strategies, or financial products.

Recommendation bias risk arises when the platform's algorithm does not recommend investments solely because they are objectively suitable for the investor, but because of commercial incentives, conflicts of interest, proprietary products, commissions, advertising arrangements, data-driven profiling, platform economics, or algorithmic optimisation.

Competition law becomes relevant where recommendation systems can steer demand, disadvantage rival investment products, reinforce platform dominance, restrict market access, or exploit information asymmetries.

1. Meaning of Recommendation Bias

Recommendation bias occurs where an investment platform systematically gives preferential visibility, ranking, default placement, or personalised recommendations to particular financial products or providers.

For example:

An investment app controls the recommendation algorithm and recommends its affiliated mutual funds more frequently than comparable third-party funds, even though the third-party products may provide equal or better returns at lower cost.

Bias may be:

1. Commercial bias – preference for products generating higher commissions.
2. Affiliation bias – preference for products belonging to the platform's corporate group.
3. Data bias – recommendations based on incomplete or distorted datasets.
4. Behavioural bias – exploiting investor tendencies such as risk aversion or fear of missing out.
5. Ranking bias – manipulating search or recommendation rankings.
6. Default bias – placing preferred products into pre-selected portfolios.
7. Personalisation bias – using extensive individual data to steer investors.
8. Algorithmic bias – machine-learning systems producing systematically skewed recommendations.
9. Self-preferencing – giving the platform's own investment products preferential treatment.
10. Feedback-loop bias – popular recommendations generate more transactions, which create more data, causing the algorithm to recommend the same products even more frequently.

2. Why Recommendation Bias Creates Competition Concerns

The fundamental competition issue is not simply that an algorithm makes a mistake.

The concern becomes much stronger when a powerful digital intermediary controls access to investors.

The platform can simultaneously control:

Investor data → search results → product ranking → recommendation → default portfolio → transaction execution → behavioural feedback

This creates a potential bottleneck between financial-product suppliers and investors.

A dominant platform could therefore influence which products investors see and which products they never consider.

3. Self-Preferencing

A major risk is self-preferencing.

Suppose Platform A operates:

• a brokerage platform;
• a robo-adviser;
• an investment marketplace; and
• its own mutual funds or ETFs.

It could design its recommendation engine so that its affiliated products appear:

• first in search results;
• as "recommended";
• as "best match";
• in default portfolios;
• as low-risk alternatives;
• or in automatic recurring-investment plans.

Competitors may technically remain available but become commercially invisible.

This can amount to a form of algorithmic foreclosure.

4. Ranking Manipulation

Investment platforms frequently rank products according to multiple variables.

For example:

Scorei=f(Returni,Riski,Costi,Liquidityi,Commissioni,PlatformRevenuei)Score_i = f(Return_i, Risk_i, Cost_i, Liquidity_i, Commission_i, PlatformRevenue_i)

The competition problem arises if the platform secretly assigns excessive weight to:

PlatformRevenueiPlatformRevenue_i

rather than investor welfare.

A platform could therefore present a product as the "best match" even though its actual investment characteristics do not justify that ranking.

The crucial legal question becomes:

What criteria determine the algorithmic ranking, and are commercially favoured criteria being disguised as neutral investor-oriented criteria?

5. Commission-Driven Recommendation Bias

Platforms may receive:

• commissions;
• distribution fees;
• referral payments;
• rebates;
• transaction fees;
• revenue-sharing payments; or
• other economic benefits.

If those payments influence recommendations, the platform may have an incentive to steer investors toward products producing greater revenue.

Competition authorities may investigate whether such conduct:

• excludes competing funds;
• raises rivals' costs;
• reduces product competition;
• increases switching barriers; or
• exploits the platform's gatekeeping position.

6. Data Advantage and Recommendation Bias

Digital investment platforms possess unusually valuable datasets.

They may know:

• investment preferences;
• transaction histories;
• income-related information;
• risk preferences;
• search behaviour;
• portfolio composition;
• investment frequency;
• reactions to market events;
• click-through behaviour;
• rejected recommendations; and
• financial-product engagement.

A dominant platform can combine these datasets to create highly sophisticated recommendation models.

Competitors without comparable data may consequently be unable to compete effectively.

This produces a possible data–recommendation–market-power feedback loop:

More users → more data → better recommendations → greater conversion → more users → even more data.

7. Behavioural Steering

Recommendation bias can operate without explicitly misleading investors.

The platform may exploit behavioural tendencies through:

• personalised notifications;
• urgency messages;
• trending-product labels;
• social proof;
• default investment allocations;
• repeated recommendations;
• gamification;
• personalised risk framing; and
• automatic investment suggestions.

Competition concerns arise where behavioural steering systematically favours the platform's commercial interests.

This moves the analysis beyond conventional price competition toward competition over investor attention and decision-making.

8. Dark Patterns and Investment Recommendations

A platform might display:

"Recommended for you"

while hiding the fact that the product generates a higher commission for the platform.

Similarly:

"Most popular"

may actually mean:

"Most profitable for the platform."

This can transform a supposedly neutral recommendation mechanism into a commercial steering mechanism.

The relevant competition-law question is whether the design distorts competitive choice among investment products.

9. Network Effects

Digital investment platforms often benefit from network effects.

More investors attract more financial-product providers.

More providers increase product variety.

More variety attracts more investors.

Consequently:

Users↑⇒Data↑⇒Recommendations↑Users \uparrow \Rightarrow Data \uparrow \Rightarrow Recommendations \uparrow

and:

Recommendations↑⇒UserEngagement↑⇒Users↑Recommendations \uparrow \Rightarrow User Engagement \uparrow \Rightarrow Users \uparrow

A biased recommendation engine can therefore reinforce an already powerful position.

10. Algorithmic Feedback Loops

Recommendation bias can become self-reinforcing.

For example:

1. Platform recommends Fund A.
2. Investors purchase Fund A.
3. Fund A becomes highly transacted.
4. Algorithm interprets transaction volume as evidence of popularity.
5. Fund A receives an even higher ranking.
6. More investors purchase Fund A.

The initial commercial preference becomes disguised as an apparently objective market signal.

This is particularly problematic because subsequent algorithmic decisions may appear data-driven even though the original dataset was itself generated by biased recommendations.

11. Exclusion of Competing Investment Products

A dominant platform can potentially disadvantage rivals through:

A. Lower ranking

Competitors appear below affiliated products.

B. Reduced visibility

Rival products are technically available but rarely recommended.

C. API restrictions

Third-party investment providers receive inferior technical access.

D. Data restrictions

Competitors cannot access necessary investor or transaction data.

E. Higher switching friction

Investors encounter greater difficulty moving portfolios elsewhere.

F. Default exclusion

The platform's products become pre-selected while competitors require additional steps.

Together these mechanisms may constitute digital foreclosure.

12. Relevant Competition-Law Theories

A. Abuse of Dominance

Under Article 102 TFEU and comparable national laws, recommendation bias may become abusive where a dominant platform uses its position to favour its own products or exclude competitors.

Possible theories include:

• self-preferencing;
• tying;
• discriminatory access;
• exclusionary ranking;
• leveraging;
• refusal to provide data or interoperability;
• margin squeeze; and
• discriminatory trading conditions.

B. Essential-Facility-Type Concerns

If an investment platform becomes an indispensable gateway to investors, competitors may argue that access to:

• platform infrastructure;
• investor data;
• API interfaces;
• ranking mechanisms; or
• execution functionality

is indispensable for effective competition.

Traditional essential-facilities doctrine would require careful application, however, because merely being commercially important does not automatically make a digital platform an essential facility.

C. Discriminatory Treatment

A platform may give its affiliated products:

• better placement;
• faster approval;
• superior APIs;
• more favourable transaction costs;
• greater visibility; or
• privileged data access.

Where comparable rival products receive inferior treatment, discrimination becomes relevant.

13. Consumer-Welfare Dimension

Recommendation bias can harm investors through:

• higher fees;
• unsuitable products;
• reduced product diversity;
• excessive trading;
• lower investment returns;
• higher portfolio risk;
• reduced transparency;
• switching costs; and
• diminished ability to compare alternatives.

The competition-law challenge is to connect these individual harms with competitive harm to the market.

An isolated bad recommendation may primarily be a financial-regulation issue.

A systematic algorithmic preference that forecloses rivals is more clearly a competition-law concern.

14. Important Case Laws

1. Google Shopping — European Commission v Google

The Google Shopping decision is highly relevant to digital recommendation bias.

The European Commission found that Google systematically favoured its own comparison-shopping service in general search results while demoting competing comparison services.

The underlying principle is important for investment platforms:

A dominant digital intermediary should not necessarily be permitted to use control over an important ranking mechanism to advantage its own downstream service.

Applied to investment platforms, analogous concerns could arise where a dominant investment marketplace systematically gives its affiliated funds preferential recommendation or ranking treatment.

Relevance: self-preferencing, ranking bias, platform leverage and foreclosure.

2. Amazon Marketplace — European Commission

The European Commission's Amazon investigation concerned the use of non-public marketplace seller data and Amazon's potential competitive advantage as both marketplace operator and retailer.

The case demonstrates the significance of dual-role platform conflicts.

For an investment platform, the analogous structure would be:

platform operator + investment-product provider + recommendation engine.

The platform's access to competitor information and investor behaviour could potentially create a structural advantage for affiliated financial products.

Relevance: data advantage, vertical integration, platform neutrality and conflicts of interest.

3. United Brands v Commission

In United Brands, the Court of Justice established important principles concerning abuse of dominance and discriminatory conduct.

Although the case predates digital platforms, its broader significance lies in examining how a dominant undertaking can use market power to impose commercially disadvantageous conditions.

For investment platforms, the case provides conceptual support for analysing whether dominant intermediaries use their position to treat market participants unequally.

Relevance: dominance, discriminatory conduct and exploitation of market power.

4. Bronner v Mediaprint

Oscar Bronner GmbH & Co. KG v Mediaprint is a foundational EU case concerning access to infrastructure controlled by a dominant undertaking.

The Court established a demanding test for when refusal of access can constitute abusive conduct.

Its relevance to digital investment platforms is particularly important where a platform controls infrastructure necessary for competitors to reach investors.

However, access to a platform would not automatically satisfy the strict Bronner criteria.

Relevance: access, indispensability, infrastructure and refusal to deal.

5. Slovak Telekom v Commission

The Slovak Telekom litigation concerned exclusionary conduct involving access to telecommunications infrastructure.

It is useful for understanding how a dominant vertically integrated undertaking can use control over an upstream infrastructure layer to disadvantage downstream competitors.

An analogous investment-platform structure could be:

investment infrastructure → recommendation interface → execution → affiliated investment product.

Relevance: vertical foreclosure, infrastructure control, access discrimination and leveraging.

6. Intel v Commission

Intel is significant for the analysis of exclusionary conduct and the economic assessment of conduct capable of foreclosing competitors.

The case demonstrates the importance of considering the actual competitive effects of conduct rather than relying solely on formal characterisation.

For algorithmic investment recommendations, authorities could therefore examine:

• scale of the platform;
• duration of preferential treatment;
• percentage of investors exposed;
• availability of alternatives;
• switching costs;
• rival foreclosure;
• efficiencies; and
• actual or likely effects.

Relevance: foreclosure analysis, economic evidence and effects-based assessment.

7. Servizio Elettrico Nazionale v Autorità Garante della Concorrenza e del Mercato

This judgment is particularly relevant to the concept of leveraging historical or structural advantages.

The Court examined whether a dominant undertaking could use advantages associated with its established position to extend dominance into related markets.

For digital investment platforms, a similar concern could arise where a platform uses:

• accumulated investor data;
• established user relationships;
• default status;
• technological infrastructure; or
• proprietary recommendation technology

to expand the market position of affiliated investment products.

Relevance: leveraging, data advantages and expansion of dominance.

15. Case-Law Comparison

16. Algorithmic Transparency

Competition authorities may increasingly need to examine the architecture of recommendation systems.

Relevant evidence may include:

• algorithmic ranking criteria;
• training datasets;
• weighting variables;
• commission data;
• product-affiliation information;
• A/B testing;
• recommendation logs;
• user segmentation;
• click-through rates;
• conversion rates;
• rejected recommendations;
• default settings;
• changes to recommendation algorithms.

A central forensic question is:

Would the platform have recommended the same product if the platform received no financial benefit from that product?

If the answer is consistently negative, the commercial neutrality of the recommendation engine becomes questionable.

17. Explainability and Competition Enforcement

Algorithmic recommendations can be difficult for authorities to investigate because the platform may claim that:

"The algorithm independently selected the product."

That statement is insufficient by itself.

The relevant inquiry is how the algorithm was designed.

For example:

Recommendation=f(UserRisk,Return,Cost,Liquidity,Commission)Recommendation = f(UserRisk, Return, Cost, Liquidity, Commission)

If the platform gives:

Commission=40%Commission = 40\%

of the recommendation weight, while:

Risk=10%Risk = 10\%

then describing the recommendation as "personalised" does not make it commercially neutral.

18. Auditability

Effective competition enforcement may require:

Algorithmic logs

Records of what products were recommended.

Version histories

Evidence of changes to recommendation models.

Counterfactual testing

Determining what would have been recommended without commercial incentives.

Independent audits

External assessment of ranking and recommendation systems.

Data-access protocols

Allowing authorities to verify relevant datasets.

Reproducibility

Ability to recreate historical recommendations.

This converts an opaque algorithmic dispute into an empirically testable competition question.

19. Dynamic Competition Risks

Recommendation bias may produce long-term structural effects.

If investors repeatedly encounter the same products, rival investment providers may lose:

• customers;
• transaction volume;
• brand visibility;
• distribution opportunities;
• data;
• economies of scale.

Eventually, competitors may leave the platform.

That can reduce:

Number of competing productsNumber\ of\ competing\ products

and increase:

Platform market power.Platform\ market\ power.

The platform may then have even greater ability to control recommendations.

This creates a foreclosure feedback loop.

20. Remedies

Competition authorities could consider several remedies.

Structural remedies

In extreme circumstances:

• separation of platform and affiliated products;
• divestiture;
• ownership restrictions.

Behavioural remedies

More commonly:

• non-discrimination obligations;
• transparent ranking criteria;
• prohibition of undisclosed self-preferencing;
• equal API access;
• data portability;
• interoperability;
• audit requirements;
• algorithmic monitoring.

Consumer-choice remedies

Platforms may be required to:

• disclose commercial relationships;
• explain recommendation criteria;
• provide alternative rankings;
• permit investor-controlled sorting;
• make default choices reversible.

21. Competition-Law Test for Recommendation Bias

A useful analytical framework is:

Step 1 — Define the market

Possible markets include:

• digital investment platforms;
• online brokerage services;
• robo-advisory;
• investment-product distribution;
• fund-distribution marketplaces.

Step 2 — Establish platform power

Consider:

• users;
• assets under management;
• transaction volumes;
• network effects;
• switching costs;
• data advantages;
• technological barriers.

Step 3 — Identify recommendation bias

Determine whether the platform favours:

• own products;
• affiliated products;
• high-commission products;
• preferred commercial partners.

Step 4 — Identify competitive harm

Ask whether rivals experience:

• reduced visibility;
• reduced transactions;
• increased acquisition costs;
• exclusion from default portfolios;
• restricted access to investors.

Step 5 — Assess efficiencies

Possible legitimate explanations include:

• better product quality;
• lower risk;
• lower cost;
• superior liquidity;
• improved investor protection;
• genuine personalisation.

Step 6 — Apply proportional remedies

The remedy should correct the competitive distortion without unnecessarily destroying useful personalised investment services.

22. Distinguishing Legitimate Personalisation From Anticompetitive Bias

Not every personalised recommendation is unlawful.

A platform can legitimately recommend Product A because:

• it has lower fees;
• it matches the investor's risk tolerance;
• it has better diversification;
• it provides superior liquidity;
• it has lower volatility;
• or it otherwise better satisfies the investor's stated objectives.

The problem arises where:

Commercial Incentive>Investor SuitabilityCommercial\ Incentive > Investor\ Suitability

and particularly where the platform deliberately hides that relationship.

Thus, competition law should avoid converting algorithmic personalisation itself into an offence.

The concern is strategic manipulation of personalisation to exclude competitors or exploit platform power.

23. Relationship With Financial Regulation

Recommendation bias frequently sits at the intersection of:

• competition law;
• securities regulation;
• fiduciary principles;
• investor-protection rules;
• data protection;
• consumer protection;
• algorithmic governance.

A recommendation that is unsuitable for an investor may principally raise a financial-regulatory issue.

But where millions of investors receive systematically biased recommendations because a dominant platform is favouring affiliated products, the issue can simultaneously become a competition problem.

24. Emerging AI-Specific Risk

Generative AI and autonomous investment agents make the issue more complicated.

An AI investment assistant may decide:

"This ETF is the best choice for you."

The investor may not know that:

• the ETF belongs to the platform;
• the platform receives a fee;
• the algorithm has been trained using platform-specific commercial objectives;
• competing products were excluded;
• or the recommendation was influenced by advertising relationships.

AI therefore creates the possibility of hidden commercial influence embedded inside apparently neutral financial advice.

The competition concern becomes particularly serious where one platform becomes the principal interface through which investors interact with financial markets.

Conclusion

Digital investment platform recommendation bias represents a developing form of algorithmic competition risk.

The central concern is not merely inaccurate financial advice. It is the possibility that a powerful digital intermediary can transform its control over data, rankings, recommendations, defaults and investor attention into market power over investment-product suppliers.

The most important competition-law risks are:

1. self-preferencing;
2. commission-driven steering;
3. algorithmic ranking discrimination;
4. data-enabled foreclosure;
5. vertical leveraging;
6. exclusion of rival investment products;
7. network-effect reinforcement;
8. behavioural manipulation;
9. reduced investor choice; and
10. algorithmic opacity preventing effective enforcement.