The third function is also the hardest to automate: accountability.
AI can generate hypotheses, challenge assumptions, and stress-test investment cases. It cannot assume fiduciary responsibility or explain a disappointing outcome to a client.
After 25 years in asset management, I can say that those conversations define the profession. During one discussion following several years of underperformance, what mattered was not model output but explaining which assumptions had failed, when they failed, and why we chose not to abandon the investment process under pressure.
AI can prepare that conversation. It cannot replace it.
Current industry practice reflects this reality. A 2024 Bank of England and Financial Conduct Authority survey found that three-quarters of responding UK financial firms already use AI, yet only 2% of reported use cases involve fully autonomous decision-making.
AI does not eliminate accountability. It changes how accountability is organized.
- Who validates models and data quality?
- Who determines whether an AI-generated signal is investable?
- Who manages dependence on external models and vendors?
- Who explains the resulting decisions to clients?
These remain investment decisions, not merely compliance exercises.
Regulators and practitioners are moving in the same direction. IOSCO’s AI/ML guidance emphasizes senior accountability, testing and monitoring, skills, third-party controls, disclosure, and data quality. The CFA Research Foundation volume AI in Asset Management, edited by Joseph Simonian, frames the issue more broadly: AI should strengthen, not supplant, human judgment, trust, and fiduciary responsibility. Gennaioli, Shleifer, and Vishny model trust as central to investment delegation.
In an AI-driven investment process, trust is earned through decisions clients can challenge and revisit.
