A new study from the University of Bayreuth suggests that AI financial advice can lift decision quality dramatically - or push investors toward worse choices, depending on what the AI is programmed to recommend. The findings come from an experiment with 3,700 participants who set up fictional savings plans and chose between two fund providers tracking the same index but charging different fees.
Without any advice, about 64% of participants picked the better provider. When a neutral AI chatbot recommended the superior option, that figure rose to roughly 87%. But when the same AI was instructed to recommend the inferior fund, only about 34% of participants made the correct choice.
The study, published as a CESifo working paper, was led by Dr. Fabian Herweg, chair of international competition policy at the University of Bayreuth. The researchers controlled the advisory tools so they either gave correct recommendations or biased ones. Participants received higher payouts for better investment decisions, which incentivized them to choose carefully.
Disclosure doesn't defuse bad advice
One of the chatbots was explicitly labeled as a "bank AI," and participants were told the bank could profit from their choice. That warning didn't measurably reduce the chatbot's influence. Transparency alone, the researchers concluded, isn't enough to protect consumers.
AI also proved more persuasive than human advisers. "Whether the advice is good or bad, AI influences decisions more strongly in the intended direction than human advisers do," said Henrik Guhling, a research associate in the university's International Competition Policy group.
That persuasive power cuts both ways. Participants with larger fictional investments - who generally showed higher financial literacy and lower risk aversion - made the right call more than 87% of the time without help. But when they received a misleading AI recommendation, correct decisions fell below 50%.
"High-quality AI advice can provide enormous support for people who lack the relevant financial knowledge. At the same time, biased AI recommendations can discourage even those who would otherwise have made excellent decisions on their own from choosing the best option," said Joshua Greubel, also a research associate in the group.
What this means for finance professionals
For finance professionals, the study underscores a practical risk: clients may trust AI outputs more than they trust human judgment, even when the AI is wrong. The findings also raise questions about how firms design and audit AI tools that touch customer money.
"Simply being transparent about the use of AI is therefore not enough. What ultimately matters are the objectives built into the AI system and the economic interests behind it," Herweg said. "Providers should be required to ensure that AI-based advice is consistently aligned with customers' interests and safeguarded through appropriate monitoring and auditing mechanisms."
Finance teams evaluating AI for Finance tools should treat this as a cautionary data point: the same model that simplifies decisions for one client can steer another one wrong. For CFOs and finance leaders, the takeaway is that AI deployment needs explicit oversight, not just good intentions - a topic covered in the AI Learning Path for CFOs. The experiment's core finding is simple: AI advice is powerful, and power without alignment is a liability.
Your membership also unlocks:
