Playing it too safe can be expensive. Most AI tools sacrificed long-term growth without giving investors much extra protection. Whether AI can complement human advice and give a second opinion on how to manage your money is an important future research area.
Dr Ylva Baeckström, Senior Lecturer in Finance at King’s Business School and co-author of the study
29 July 2026
AI investment advice may sacrifice returns without reducing risk
New research found most AI tools were too cautious, whilst human adviser fees erased much of their advantage.
Investors using general-purpose AI for financial advice could miss out on up to £42,000 for every £100,000 invested over 20 years, according to new research from King’s Business School.
The study found that most of the large language models (LLMs) tested recommended safer portfolios than professional financial advisers. But this additional caution brought little reduction in long-term volatility, meaning investors could sacrifice returns without receiving a comparable reduction in risk.
Published in the Journal of Corporate Finance, the research compared recommendations from 190 professional advisers with nine LLMs, including GPT, Gemini and Claude. Each assessed the same ten fictional wealthy clients and selected from seven portfolios with different risk and return profiles.
The researchers found shortcomings on both sides. Human advisers showed evidence of projecting their own investment preferences onto clients, particularly when recommending higher-risk portfolios. AI avoided this form of projection when personal information about the adviser was removed from its prompt, but introduced distortions of its own.
Most AI configurations clustered clients into more conservative portfolios and were less responsive to differences in their circumstances. However, this was not universal. Some newer models recommended more risk than human advisers, while one GPT-5.2 configuration closely matched the overall distribution of human recommendations.
These differences could have substantial consequences. The most conservative AI configuration produced estimated terminal wealth of around £189,100 from a £100,000 investment after 20 years, compared with £230,700 under the average human recommendation, an 18% shortfall.
The comparison changed when fees were included. Applying a typical 1% annual fee reduced the estimated human outcome to around £190,300, while the AI recommendations were evaluated without a fee. The study calculated that, above an annual fee of 1.03%, even the most conservative AI configuration would produce greater terminal wealth.
The practical takeaway is that investors, advisers and regulators should not treat “AI advice” as one consistent product, or mistake caution for suitability. Models need to be tested individually and at client level, rather than judged only by their average recommendations.
The authors argue that the most promising approach is a hybrid model, in which AI produces a first draft that a qualified professional can challenge or override.
The findings relate mainly to wealthy investors and hypothetical scenarios, so should not be read as evidence that consumers should replace regulated advice with a general-purpose chatbot.
The research was co-authored with Dr Roman Matkovskyy of Rennes School of Business.
Read the full study, Financial advice behaviour: humans versus AI.
In this story
Senior Lecturer in Finance