Key Points

• The token expenditure of AI agents may reach 20% of the total salary.
• Manual supervision will add considerable costs to bank automation.
• Autonomous financial assistants may threaten the profit margins of traditional banks

As the operating costs of artificial intelligence agents become higher, banks around the world are facing an unexpected financial challenge that could erode the savings that institutions hoped to achieve through automated customer service, compliance, and administrative tasks.

McKinsey warns that the expenditure on AI tokens, which are used to process information through language models, could account for 20% of the total labor costs of banks. This warning comes from the company's newly released "Global Banking Annual Review." McKinsey states that global banking profits in 2025 have reached a record $1.3 trillion.

AI The agency costs may reach 20% of the bank's labor expenditure.

Banks initially adopted the AI assistant to speed up daily tasks, including answering customer questions, reviewing documents, and assisting compliance teams.

However, autonomous agents bring about a different cost structure, as they are capable of executing multi-step workflows, repeatedly invoking language models, and communicating with other agents.

McKinsey Senior Partner Vik Sohoni explained that as agents continue to interact, the effectiveness of traditional cost control diminishes, and token consumption accumulates rapidly.

Some institutions have already established AI FinOps teams to monitor computing expenditures and redirect workloads towards more economical models.

But token expenditures are just part of the problem. A previous study by McKinsey found that in certain bank customer service workflows, manual supervision accounts for about 70% to 75% of variable costs, while tokens account for only 20% to 25%.

This also explains why banking AI regulators are placing increasing emphasis on human responsibility and transaction-level control.

The automation savings faced by banks may not necessarily translate into profits.

When employee productivity increases, but actual operating expenses do not decrease, economic benefits become more complex.

McKinsey's research found that in 7 production systems of financial and professional service companies, the AI agents reduced the task completion time by an average of about 60%.

However, a shorter task duration does not necessarily translate automatically into a smaller team or lower wage expenses.

As banks advance organizational restructuring while investing in technology, this distinction becomes increasingly important.

A bank may save two hours for each employee through automation, but it still has to pay the same salary, in addition to the additional AI subscription fees, infrastructure costs, and supervision expenses.

AI Agents May Also Threaten Bank Revenue

Financial risks are not limited to operating costs alone.

McKinsey surveyed approximately 46,000 retail bank customers and found that 23% of them are already using generative AI to handle financial tasks.

Among customers using AI, 72% indicate a willingness to delegate at least one financial task to an agent, although authorization restrictions remain a major concern.

This change may accelerate autonomous financial transactions, enabling customers to compare savings interest rates and transfer their deposits to institutions with higher rates.

For banks, this threatens a traditional source of income: low-cost customer deposits, which are used to issue interest-bearing loans and make investments.

This development also brings new risks of payment liability, especially when the software executes transactions without personal manual approval.

At the same time, the increasingly capable AI investment agents may extend automated financial decision-making beyond ordinary banking services.