Bank of New York Mellon has not followed the "tokenmaxxing" practice that was once popular in the enterprise AI field. The bank's chief financial officer, Dermot McDonogh, said that internal AI evaluations focus more on business results than on the number of prompts, token consumption, or agent deployment scale.
Do not treat token usage as a core metric
McDonogh stated that token costs represent a small percentage of the bank's overall engineering budget, so management did not prioritize it. Even as this metric gained traction in external discussions, BNY ultimately concluded that it did not adequately reflect the true value of AI.
He stated that since the emergence of ChatGPT, BNY has been building its internal AI platform for several years and collaborating with multiple cloud service providers and model providers. This platform is not tied to a single large model; the internal system assigns tasks to appropriate models, eliminating the need for employees to manually optimize prompts repeatedly around cost considerations.
The code, account opening, and payment processes have been implemented.
BNY disclosed that AI has been integrated into several core business processes, and its effects are beginning to be quantified.
- In the first quarter of 2026, over 40% of code was generated by AI.
- This proportion has recently risen to approximately 50%.
- About half of the annual account plans are drafted by AI.
In addition, approximately 25% of customer account opening processes are now AI-enabled, and about 70% of restricted party payment screenings are reviewed by AI. McDonogh stated that these changes have not directly resulted in a reduction in staff size, but have improved the processing capacity of existing teams.
Financial indicators improved
From a financial perspective, BNY's revenue per employee has increased from $338,000 in 2022 to $401,000 in 2025; during the same period, pre-tax profit per employee has increased from $99,000 to $143,000. McDonogh defines these changes as "capacity expansion" rather than simply cost savings.
BNY is currently tracking the impact of AI across key areas such as innovation, customer acquisition, account opening, trading, and process streamlining, and is continuously improving its internal "Eliza" platform. This platform serves as the company-wide context layer and will iterate as data and usage scenarios increase.
In terms of employee management, BNY has also established a three-tiered AI proficiency system. Employees can only advance to a higher level and gain access to more advanced models after completing training and testing. The bank stated that this is done to balance usage quality and accountability.
McDonogh also mentioned that within the finance department, AI has begun to be used in regulatory reporting, balance sheet analysis, predictive modeling, and financial statement preparation, including summarizing analyst expectations and preparing potential investor questions in advance. According to him, the key to measuring AI productivity is not "how much is used," but "how much more the organization can do as a result."
