"Are AI-Chosen Financial Products the Best?"... Transparency in Recommendation Criteria a New Challenge

Concerns over conflict between consumer interests and platform profits
Overseas regulators stress transparency in AI recommendations
"From recommendations to sign-ups... Conflicts of interest must be managed"

As artificial intelligence (AI) moves toward choosing financial products on behalf of consumers, ensuring fairness in the recommendation process is emerging as a new challenge. If an AI platform’s partnerships or revenue model influence its recommendations, products that would be more beneficial to consumers could be passed over. Experts are calling for greater clarity around the transparency of AI product recommendations and responsibility for transactions.

According to the financial industry on November 3, four regulatory bodies, including the UK Financial Conduct Authority (FCA) and the Competition and Markets Authority (CMA), issued a joint statement on October 30 on consumer protection related to AI agents. The agencies stressed that if commercial interests are involved when AI recommends or transacts in products on behalf of consumers, users should be able to understand how those interests affect the AI’s actions. The aim is to clarify who is responsible and how consumers can seek redress when problems arise from incorrect decisions.

This issue is becoming increasingly important as AI agents evolve to make payments for products tailored to consumers’ circumstances. AI recommendations may be influenced not only by objective factors such as credit card discounts, loan interest rates and insurance premiums, but also by platforms’ revenue models and partnerships. For example, if a platform is designed to factor partnership revenue into its recommendation criteria, it could prioritize Card B, which generates more revenue for the platform, over Card A, which offers consumers a larger discount. For loans and insurance, products with less favorable interest rates or coverage terms for consumers could also appear higher in the results.

“The phenomenon in which selection criteria that should be designed around consumers’ net benefits shift in response to an operator’s economic incentives is called ‘objective function drift,’” said Yoo Jaewon, a research fellow at Toss Insight. “Internal performance indicators can create the same incentives when certain choices generate economic value for the operator, such as transaction volume, customer time spent on the platform, cross-selling results, use of affiliated financial products or data acquisition.”

The credit card industry is already concerned about risks arising from payments made by AI agents. According to a survey of 71 credit card company employees conducted by the Credit Finance Research Institute, unclear accountability was identified as the most serious risk, while platform dependency and recommendation bias were also rated as major risk factors.

Industry officials say that as AI agents become more deeply embedded in the payment ecosystem, systems are needed to verify the criteria used to recommend financial products and the scope of products compared. They say it should be possible to determine why AI selected a particular product and how economic interests, such as partnership fees, affected the result. “When recommendations are based on advertising or partnerships, users should be able to understand that fact, the scope of products being compared and the main recommendation criteria,” said Yoon Jongmun, director of the Credit Finance Research Institute.

“The issue of showing partner products first has existed for a long time, but consumers often made the final comparison and choice themselves,” a financial industry official said. “If AI chooses and pays for a product on a consumer’s behalf, or even goes as far as signing them up, a single action by AI could have a greater impact on actual financial transactions.” The official added, “A more beneficial product for the user should never be pushed down the list because of platform fees or other factors.”

This content was produced with the assistance of AI translation services.