Most banking enterprises now have an AI strategy. What’s the gap between AI spend and realized value? It isn’t the tech. It’s the absence of an economic discipline for intelligence. That discipline has a name: tokenomics.
What every bank must know about AI economics right now
AI spending in banking exceeded $73 billion in 2025, yet only four of the top 50 banks reported realized ROI from use cases.
Banks still manage AI through vocabulary built for FTEs and licenses, hiding what tokens realistically cost or produce.
Tokenomics is the discipline of designing, optimizing, and governing the economics of intelligence inside a financial institution.
It introduces five primitives that translate AI consumption into a governable, finance-grade discipline for boards and operators.
Institutions that build this discipline early will define the cost, speed, and safety frontier of banking for the next decade.
The AI economics gap, in four numbers
$73B+
Banking AI spend in 2025
95%
GenAI implementations stuck in pilot
4 of 50
Top banks reporting realized ROI
Aug 2026
EU AI Act high-risk provisions apply
Is there an ROI paradox at the heart of banking AI?
Investment is at record highs and adoption is near universal. On every meaningful measure of value realization, however, the industry remains in early innings. McKinsey’s Global Banking Annual Review 2025 estimates AI adoption could trim banking industry costs by up to 20 percent. Generative AI alone, by McKinsey Global Institute estimates, could create $200 billion to $340 billion in annual value in banking, equivalent to 9 to 15 percent of operating profits. The execution picture is more sobering. Industry analysis indicates that 95 percent of generative AI implementations in financial services remain in pilot rather than scaled production. Only four of the 50 largest banks reported realized ROI from AI use cases in 2025. 70 percent of financial services organizations are deploying or exploring agentic AI, but only 14 percent have achieved full-scale implementation. The conventional answer points to data quality, talent gaps, regulatory uncertainty, and integration complexity. All of those are real. None is the deepest cause.
The deepest cause is structural. Banks are attempting to manage the economics of intelligence using a financial and operational vocabulary that was built for a different era.
That vocabulary, built around FTEs, projects, software licenses, and transaction volumes, has no clean place to put what AI actually is, what it costs, or what it produces.
Three forces making tokenomics a board-level consideration
A token, in the language of large language models and agentic systems, is the atomic unit of work performed by an AI system. Every prompt, every retrieval, every model call, every agent action, every drafted credit memo, every triaged complaint is metered, priced, and consumed in tokens. Tokens are to the agentic bank what kilowatt-hours are to a manufacturing economy and what API calls were to the early cloud. They have pricing, demand elasticity, utilization patterns, capacity constraints, and value attribution that must be managed deliberately, or they will manage the institution. Three forces make this a board-level concern, not a future one.
Force one: The scale of spend has crossed a threshold
The AI in banking market is forecast to reach $45.6 billion in 2026 and $143.6 billion by 2030, at a compound annual growth rate exceeding 30 percent. JPMorgan Chase reportedly operates hundreds of AI models enterprise-wide, with approximately 150,000 employees using large language models every week. Bank of America has invested several hundred million dollars across 20 AI projects. Goldman Sachs has identified six core operating processes targeted for AI-driven reshaping. This is no longer an IT line item but one of the largest discretionary spend categories on the bank’s P&L, and the trajectory is steepening.
Force two: Consumption volatility breaks traditional forecasting
AI spending does not behave like any spend category most CFOs have managed before. IDC’s FutureScape 2026 warns that by 2027, large enterprises will face up to a 30 percent rise in underestimated AI infrastructure costs. The cause is not excess spending but systemic under-forecasting and the opacity of consumption models. A single agentic workload deployed at scale can generate thousands of dollars of token cost in minutes. Multiplied across the dozens of agents a bank may deploy in origination, servicing, surveillance, and middle-office processes, the financial control problem becomes structural.
Force three: The EU AI Act is closing the compliance window
The EU AI Act’s high-risk provisions apply to financial services from August 2026. Credit scoring, automated lending, and AML risk profiling systems must meet strict requirements around transparency, human oversight, auditability, and bias detection. The Act has extraterritorial reach. Any institution serving the EU market is in scope regardless of headquarters location. Non-compliance penalties reach up to 7 percent of global annual turnover. Over half of organizations still lack a systematic inventory of the AI systems they operate, which is the fundamental prerequisite for compliance. Getting tokenomics wrong is a regulatory liability.
What the full article covers
This article sets the strategic case for tokenomics. The PDF delivers the complete operating blueprint for banks moving from concept to execution. Inside: the six-stage maturity model, the 90-day sprint that produces live results in one quarter, the four-tier governance structure with a board-ready RACI, the workforce transformation roadmap for oversight and advisory roles, and the quarterly workshop program that sustains tokenomics literacy across the enterprise. Also included are implementation timelines, decision templates, and diagnostic frameworks that make the discipline immediately actionable for boards, CFOs, and transformation leaders.
Three questions every banking tech leader should ask
What does a token produce, not just cost?
Cost per token misses the value question. Outcome density per million tokens is the metric that separates AI activity from AI value.
Who owns AI economics?
Diffuse ownership is why 95% of pilots stay pilots. Tokenomics requires a single accountable executive with a clear remit.
Is the platform funded, or the project portfolio?
Project mode pays for the same plumbing dozens of times. Platform mode compounds. The decision sets bank AI economics for a decade.
Is tokenomics just repackaged FinOps with a new name?
The principles overlap, but banking adds asymmetric downside, model risk, and regulatory liability that generic FinOps frameworks were never engineered to govern.
How banking leaders should activate tokenomics now
Reframe the AI conversation at the board level. Move it from “how much are we spending on AI?” to “what is our outcome density per million tokens?” within the next reporting cycle.
Name a single accountable executive for AI economics. If no single name exists, tokenomics is aspiration, not discipline.
Stop negotiating per-token prices as the lead strategy. Build the inventory, instrumentation, and outcome attribution that make price negotiations evidence-based rather than reflexive.
Fund the platform, not the project portfolio. The cost-curve advantage compounds, and it cannot be recovered later.
A series for the agentic banking era
This article opens a seven-part series on tokenomics for banks. The articles that follow explore unit economics for the bank CFO, outcome density as the metric that matters, the platform versus project decision, the 90-day implementation sprint, the governance and ownership question, and the workforce and culture shift the discipline demands.
Tokenomics: Good to know
Tokenomics is the discipline of designing, optimizing, and governing the economics of AI inside a financial institution, using the token as the atomic unit of measurement, value, and accountability.
Banks should measure AI ROI through outcome density: the business value produced per million tokens consumed, expressed in the currency of the workload's outcome. Loss avoided per thousand tokens for fraud triage. Regulatory escalations avoided for complaint handling. Time saved for advisory copilots. Cost per token is the wrong lens.
AI economics requires a single accountable executive supported by a cross-functional Tokenomics Council spanning Finance, Risk, Technology, and business lines. Ownership diffused across an IT center of excellence, a FinOps team, and a model risk committee is the structural reason most AI programs stall between pilot and production.
The EU AI Act's high-risk provisions apply to financial services from August 2026, covering credit scoring, automated lending, and anti-money laundering (AML) risk profiling. Non-compliance penalties reach 7% of global annual turnover. Any bank serving EU customers must maintain a systematic AI workload inventory, which most institutions currently lack.
Forward-looking thoughts and compelling stories
Case Study
Banking and Financial Services
Auto lender cuts wait times by 30%, lifts velocity by 75%
