BNY ignores tokenmaxxing, measures AI by business outcomes instead

The tokenmaxxing trend emerged as a status symbol at major tech firms, where engineers were pressured to climb leaderboards by consuming more AI tokens—a practice critics argue misaligned incentives and masked a gap between AI spending and real productivity gains. BNY Mellon took the opposite path. Rather than adopt external benchmarks, the bank's leadership, led by CEO commitment and CFO Dermot McDonogh, built an internal LLM-agnostic platform and invested years in cultural adoption after ChatGPT emerged. The bank was deliberate: it routed tasks to appropriate models without requiring employees to manually optimize prompts, and it measured success by business outcomes, not token volume.

That discipline produced tangible financial results. Revenue per employee climbed from $338,000 in 2022 to $401,000 in 2025, a 18.6% gain over three years; pre-tax income per employee jumped from $99,000 to $143,000 over the same span—a 44.4% increase. These gains reflect what McDonogh calls "capacity creation": the bank did not shrink its workforce, but AI allowed it to accomplish more with the same headcount. The platform, called "Eliza," serves as a firm-wide context layer, improving as it ingests additional data and use cases. BNY's structured approach to employee adoption—progressing staff through proficiency levels to a "pioneer" designation requiring formal training—reinforces both quality outcomes and organizational accountability, setting it apart from the undisciplined token-chasing culture that emerged elsewhere.