Banks are navigating the integration of agentic AI, balancing potential customer demand with significant operational and regulatory hurdles. According to a McKinsey survey of 46,000 retail banking customers, 72% of AI users would permit an agent to perform financial tasks, provided autonomous approvals remain under $500. Analysts expect widespread adoption of these agents within three years.
Regulatory compliance remains a primary concern, particularly regarding transaction authorization and customer consent. WilmerHale highlights that authenticating an agent is distinct from verifying its authority. Banks currently face challenges in defining the limits of delegated actions, such as permitted amounts or counterparties, which complicates existing anti-money-laundering and sanctions monitoring protocols.
Internally, banks are grappling with the financial impact of AI processing costs. McKinsey warns that token consumption for internal agents could reach 20% of total labor costs. To manage these expenses, some institutions have established AI FinOps teams to oversee spending, while others are becoming more selective about deployment to ensure that productivity gains and revenue justify the investment.
The combination of customer-facing agents and internal automation creates potential margin pressure. As external agents may shift deposits toward higher-yielding products, banks must simultaneously absorb the costs of their own AI infrastructure. Consequently, firms are focusing on establishing clear boundaries for delegation and improving recordkeeping to handle potential transaction disputes.
