The paper highlighted today provides an analysis of 36 of the largest Bank Holding Companies (BHCs) in the U.S. and finds a one standard deviation increase in AI investment produces a 24% increase in quarterly operational losses. Ouch!
The more banks have incorporated AI into their operations the higher risk they’ve exposed themselves to and the greater subsequent losses they’ve incurred. It gets worse. Not only are these losses of the trivial and recurrent variety they also encompass novel tail-risks introduced in the process of AI adoption.
The work has authority as it’s based on confidential reporting banks must make to the Federal Reserve which is not available to the general public.
Ping McLemore, a Senior Financial Economist at the Federal Reserve Bank of Michigan and Atanas Mihov, an Associate Professor of Finance at the University of Kansas have used the Fed data together with employment databases to match up the number of qualified AI-capable employees a bank has taken in to produce their results.
Analysts, the keen and other practitioners can read the paper in full via this link AI and Operational Losses: Evidence from U.S. Bank Holding Companies.
For the time-poor I’ll offer my two pennyworth summary here. AI bad, have nothing to do with it? No. AI new, experimental and in need of active husbandry. The paper highlights that problems were more acute in banks that already had a poor culture of risk management. Sloppy managers who think it offers plug-and-play solutions have been most negatively affected, as they would be with any other technological advance.
The key, in my opinion, is not to see a one-size-fits-all problem solver in AI applications. Good managers can, do, and will learn from mistakes and apply the technology more effectively going forward.
Understanding risks better, which is what the paper aims to help do, is the way forward for more effective AI application; not the wholesale abandonment of an infant technology currently generating mixed results.
Happy Sunday.