Tokenmaxxing Is Enterprise AI’s Vanity Trap—BNY’s CFO Spills the Real ROI Playbook

Posted on by

(SeaPRwire) –

By: Oliver Hawthorne

Tokenmaxxing has become enterprise AI’s most dangerous vanity metric. Big tech firms turn token burn into a leaderboard competition, pushing engineers to rack up counts instead of delivering value. This craze masks a critical gap: AI spending is skyrocketing, but productivity gains are lagging.

BNY’s CFO Dermot McDonogh wants none of it. He says tokenmaxxing never gained traction at the bank, calling token costs “modest within modest” compared to its engineering budget. BNY built an early, deliberate AI strategy post-ChatGPT: an LLM-agnostic internal platform, partnerships with hyperscalers and model providers, and CEO-level buy-in. The bank focused on demystifying AI for staff, so employees don’t fear the technology. This cultural shift lets BNY scale AI without fixating on cost per query. Internal systems route tasks to the right models automatically, no manual prompt optimization needed. McDonogh doesn’t track prompt volumes—he tracks outcomes. Those outcomes are clear: over 40% of BNY’s code was AI-authored in Q1 2026, rising to 50% recently. Half of annual account plans use AI drafts, 25% of client onboarding gets AI support, and 70% of restricted-party payment screening relies on AI reviews. Financial results follow: revenue per employee jumped from $338,000 in 2022 to $401,000 in 2025. Pre-tax income per employee climbed from $99,000 to $143,000 over the same period. McDonogh frames these gains as capacity creation, not cost cuts—BNY hasn’t reduced its workforce, but it does more with the team it has. The bank tracks AI impact across core workflows: innovating, prospecting, onboarding, transacting, streamlining. It builds out its internal “Eliza” platform, a firm-wide context layer that improves with more data and use cases. Employees progress through three AI proficiency levels, with “pioneer” status requiring training and testing. Advanced model access is gated by expertise, ensuring quality and accountability. In finance, AI reshapes regulatory reporting, balance sheet analytics, predictive modeling, and earnings preparation.

The commercial loop here is unmissable. BNY’s approach ties AI directly to tangible business capacity, not arbitrary metrics. The industry end-game will see a split: firms clinging to tokenmaxxing will waste resources on empty signals, while those prioritizing outcome-focused AI will pull ahead. Enterprise AI’s future isn’t about how many tokens you burn—it’s about how much more you can do with the people and resources you already have.

Author bio: Oliver Hawthorne is a principal correspondent at Global Tech Insights, covering enterprise AI and financial technology trends worldwide.