The ongoing evolution of artificial intelligence’s role in corporate finance brings with it many questions. For one: Who should be responsible when a financial model developed with the assistance of AI produces an error?
If anyone should have a definitive answer, it would be financial modelers, right? Well, maybe not.
A newly formed group of expert modeling professionals created by the Financial Modeling Institute, a provider of accreditation programs, appears to be nowhere close to a consensus point of view on the question.
The 63-member Financial Modeling Global Leaders Council agreed almost unanimously that when an AI tool generates or substantially modifies a financial model, a human review is essential before the model can be used for decision-making.
Yet, when asked who should bear primary responsibility when such a model leads to a significant error or financial loss, the council’s leading response — the human modeler — garnered just a 25% share.
Almost the same proportions said there should be shared accountability (22%), that the organization itself should be responsible (21%) and that the model owner/sponsor should be most accountable (19%). Another answer option, the reviewer or approver of the AI-assisted model, drew 13% support.
“The most experienced modelers in the world have not resolved the fundamental question of accountability in an AI-assisted environment,” FMI wrote in its research report. Further, the report opined, “If they cannot agree, it is unlikely that the organizations they serve have figured it out either.”
Analyzing the council’s choices, FMI noted that those who selected the human modeler argued that a modeler can’t abdicate responsibility just because a tool was involved. Those who advocated for shared responsibility pointed out that AI-assisted models can introduce systemic risks, the report said. And those who said the model owner/sponsor, who commissioned the model and used its outputs in making a decision, is the ultimate bearer of responsibility.
“If accountability is unresolved, it creates ambiguity in governance frameworks, audit processes and professional liability,” the report said.
The council members were similarly disparate in their views on what information should be disclosed to a model’s users in cases where the model was substantially built or updated using AI tools.
That scenario is far from rare, with 86% of council members saying they had used AI for modeling tasks in the previous year.
However, for a majority of the expert modelers, AI is not a major element of their work. Almost half (43%) said AI assists 0% to 25% of their workflow, and 27% said it affects 11% to 25% of their workflow.
Additionally, they were at odds in their perceptions of the direction of the financial modeling profession. Almost half (48%) of the modelers saw downward pressure on modeling as a career, while 37% saw upward opportunity.
The expert financial modelers on the council consisted of practitioners (29%), trainers/educators (24%), consultants/advisers (24%), association leaders (21%) and academic/others (3%).
