Banks Prepare for AI Agents as a New Customer Interface

Banks are testing AI agents that could help customers access accounts and make payments, while institutions develop controls for data sharing, authentication and automated transactions.

Financial institutions are testing AI agents that can understand customer requests, gather information and complete tasks through connected services. In banking, an agent could compare accounts, transfer money, schedule payments or track spending across several systems.

Banks are examining how these systems can connect with account platforms and digital channels. Their work includes setting access rules, deciding which transactions need customer approval and verifying that an instruction came from the account holder.

Institutions are also creating records of an agent’s actions and limits for automated payments. They need procedures to stop activity when a request appears unusual and safeguards for personal and financial data shared with agents operated by technology companies or other third parties.

Customers could use an agent to search for a loan, explain fees or arrange a transfer instead of visiting a bank website or mobile application. The agent would communicate with the institution through approved application programming interfaces, which allow software systems to exchange information.

Banks are assessing how consent should work in agent-led services. A customer could allow an agent to view account information for a limited period, approve certain payment types or operate within a spending limit. Institutions are separating tasks that agents can complete automatically from those that require confirmation.

Privacy, fraud prevention, consumer protection and payment-access rules will govern the rollout of these services. Financial institutions may also have to explain automated decisions and identify which company is responsible when an agent makes an error.

The technology is at an early stage in financial services. Institutions are building testing environments, reviewing technology partnerships and updating risk controls before agents can perform actions on customer accounts. Initial services are likely to focus on information and customer support, followed by limited transactions with extra approval requirements.

Articles by this author