Banks Face Rising AI Agent Costs As McKinsey Warns Of Growing Financial Burden

Banks worldwide are reassessing their use of artificial intelligence agents as the cost of operating autonomous systems threatens to become a significant expense, according to findings highlighted by consulting firm McKinsey.

In a report published on November 10, 2026, Reuters said banks are increasingly concerned that spending on the digital tokens used to operate AI agents could eventually reach 20% of their total labour costs. The estimate, discussed in McKinsey’s Global Banking Annual Review, reflects the financial challenges emerging as lenders expand their use of automated technology.

The findings suggest that banks are moving beyond experimentation with AI and examining whether the technology can deliver enough measurable financial benefits to justify its growing operating costs.

Why AI Agents Are Becoming Expensive For Banks

AI agents are software systems capable of carrying out tasks with limited human intervention. Unlike conventional chatbots that primarily respond to questions, these systems can perform multiple connected activities, communicate with other agents and complete complex workflows.

Banks are exploring their use in customer service, financial analysis, compliance monitoring and routine administrative operations.

However, every interaction with an AI model requires computing resources. These interactions are generally measured in tokens, which represent units of information processed or generated by the system.

Technology providers commonly charge customers according to the number of tokens consumed, the model used and other computing requirements.

The cost becomes more complicated when several AI agents work together.

A single customer request may trigger multiple automated processes, including retrieving information, analysing data, communicating with other systems and checking the final response.

Each additional process can generate more token usage and increase operating expenses.

According to Vik Sohoni, a senior partner at McKinsey, traditional methods of controlling AI spending become less effective when agents communicate with one another and generate additional processing demands.

The concern is particularly relevant for banks, where automated systems may be required to handle sensitive financial information accurately while operating at considerable scale.

Banks Establish Teams To Monitor AI Spending

Financial institutions are beginning to create specialised teams responsible for managing the cost of artificial intelligence.

Some banks refer to these functions as AI FinOps, a term describing the management and optimisation of AI-related financial expenses.

Initially, lenders attempted to control costs by directing simpler employee requests towards less expensive AI models and reserving more advanced systems for complicated tasks.

Such arrangements can help reduce unnecessary computing expenses when employees interact directly with AI tools.

However, the approach becomes more difficult when autonomous agents communicate with other agents and perform tasks without continuous human supervision.

As a result, banks are examining the financial performance of individual AI applications more closely.

Rather than deploying AI across every available business function, institutions are increasingly concentrating on activities expected to produce stronger returns.

Productivity gains remain an important consideration because banks can measure reductions in processing time and manual workloads.

However, McKinsey’s analysis suggests that lenders are also looking for evidence that AI can generate additional revenue, rather than merely reduce existing costs.

This shift could influence which AI projects receive further investment and which remain limited to testing.

Customer Adoption Creates New Challenges For Banking

While banks are attempting to manage AI expenses, customers are adopting the technology at an accelerating pace.

McKinsey expects AI agents to become widely used within approximately three years, considerably faster than the adoption of mobile internet, which took between five and ten years to achieve broad acceptance.

A survey involving approximately 46,000 retail banking customers worldwide found that consumers are already using AI tools to compare financial products, identify savings opportunities and explore ways to manage debt.

Among respondents using AI, 72% indicated that they would allow a banking agent to perform at least one financial task on their behalf.

However, many remained cautious about giving automated systems unrestricted authority over their money.

Respondents who expressed willingness to permit autonomous financial transactions generally preferred approval limits of $500.

The findings demonstrate a growing willingness to use AI for financial decisions alongside continuing concerns about security and control.

For banks, this creates another challenge.

AI agents may help customers identify better interest rates, compare competing financial services and move money towards more attractive products.

Such capabilities could intensify competition between financial institutions and put pressure on traditional sources of banking revenue.

What This Means For You

The increasing use of AI agents could eventually change how customers interact with banks, from comparing savings products to managing routine financial activities.

However, wider adoption does not necessarily mean that banks will immediately allow autonomous systems to move money or approve transactions without supervision.

Many financial institutions continue to restrict external AI agents from authorising transactions because of security, compliance and liability concerns.

For Indian banks and financial technology companies expanding their AI capabilities, the findings highlight the importance of balancing automation with operating costs, cybersecurity safeguards and customer trust.

Financial institutions will need to demonstrate that AI investments deliver measurable benefits while retaining adequate oversight of systems handling sensitive customer information.

The challenge is no longer simply whether banks can deploy increasingly capable AI agents, but whether they can operate them economically and securely at scale.