The Clean Winner Bank: How Risk Discipline Becomes a Competitive Edge in AI & Financial Institutions

Among financial institutions, the case for responsible AI has acquired the shape of an orthodoxy. Regulatory pressure is intensifying. Reputational stakes are climbing. The consequences of governance failure, when they arrive, are not easily absorbed. From these premises follows the recommendation that prudent institutions invest accordingly.

Across the sector, AI governance has settled into a familiar shape: documentation produced, committees seated, board materials updated, without producing the deployment outcomes that bank leaders actually care about. MIT’s NANDA initiative, in its 2025 study of three hundred enterprise AI deployments and one hundred and fifty-three leadership surveys, found that 95 percent of generative AI pilots delivered zero measurable financial return despite nearly $40 billion in U.S. enterprise investment. Only 5 percent achieved scale. The barrier lay not in the model quality but in the gap between deployment and the organizational systems meant to absorb the technology.

A 2024 IBM study of five thousand C-suite executives across twenty-two industries captures the dynamic: spending on sustainability reporting exceeded spending on sustainability innovation by 43 percent. The same gap, transposed into AI, explains a great deal about why the technology keeps stalling at the pilot stage even as capital flows in. BCG’s 2025 global survey places only 5 percent of companies in the “future-built” category, where AI capabilities are scaled across functions and consistently generate substantial value.

The reason this gap persists is an incentive structure that rewards evidence of activity over evidence of value. Governance investments that produce documentation are visible immediately to regulators and boards; those that produce competitive advantage are visible only to customers, and only over time. When the visible reward sits on one side of the ledger and the compounding, structural reward on the other, the predictable outcome is a great deal of compliance theater and very little customer-facing change.

This is the problem Goutam Challagalla and Frédéric Dalsace diagnose in Clean Winners, their 2026 study of why sustainability strategy so often fails to deliver either business returns or environmental progress. Their central reframe maps cleanly onto AI regulation in finance. They argue “sustainability should be a catalyst of innovation that drives customer value.” Properly designed, governance becomes the engine of differentiation, the mechanism by which an institution moves faster and into more sensitive territory than competitors can credibly match.

The institutions pulling ahead in 2026 are those treating governance as the lever that lets them deploy with more autonomy, into more use cases, with less organizational friction.

The framework that makes this concrete is one borrowed from Challagalla and Dalsace: the distinction between Right to Play, Right to Stay, and Right to Win investments.

Right to Play. These are the investments necessary to operate at all. In finance, that means SR 11-7 model risk management compliance, the EU AI Act’s high-risk system obligations coming into force August 2026, and the U.S. Treasury’s Financial Services AI Risk Management Framework released February 2026 with its 230 control objectives. These are non-negotiable and also indistinguishable from what every other regulated competitor is doing.

Challagalla and Dalsace are precise about this category, saying, “Right to Play investments are those that keep a firm in the game.” The institutions treating governance as principally a Right to Play activity, running compliance programs, producing artifacts the examiner wants to see, satisfying the letter of the rule, are not doing anything wrong. They are simply not creating a competitive advantage. The CSO of one large German firm told the authors she spends 90 percent of her time ensuring compliance. The equivalent figure for many bank model risk officers in 2026 would likely be higher.

The danger is that the institutional reflex stops there, and the bank treats its entire AI governance budget as Right to Play spending, with no separate allocation for the categories that actually compound over time.

Right to Stay. These are forward-looking investments in resilience. They protect competitive positions against pressures not yet fully arrived. In financial services AI, some examples are continuous monitoring infrastructure, structured vendor diligence on foundation model providers, board-level AI risk reporting cadence, and the talent pipelines that allow an institution to absorb regulatory change without scrambling.

The Monetary Authority of Singapore’s November 2025 consultation paper on AI Risk Management Guidelines, and the MindForge AI Risk Management Toolkit released by a twenty-four-firm consortium in March 2026, illustrate how Right to Stay investment compounds. Singaporean banks participating in MindForge, DBS, OCBC, UOB, Standard Chartered, Citi Singapore, and HSBC, have been building AI governance infrastructure together for eighteen months. When MAS finalizes the guidelines in 2026, these institutions will already be operating inside the framework. Banks that treated MAS as a distant signal, missing its function as a leading indicator of where every major jurisdiction is heading, will absorb in 2027 the regulatory transition cost that Singaporean institutions absorbed across 2024 and 2025.

The same logic applies to the EU AI Act’s August 2026 high-risk obligations. Institutions that built data governance, model documentation, and human oversight capabilities ahead of the deadline will treat compliance as a documentation exercise. Institutions that waited will treat it as a re-engineering exercise. The cost differential between the two is substantial and entirely a function of when the investment was made.

Broadly, Right to Stay investments share a structural feature: they are invisible until they are not. The bank that built continuous behavioral monitoring in 2024 looks indistinguishable from the bank that did not, until an agentic system drifts mid-week and one institution catches it before the regulator does. At that moment, eighteen months of Right to Stay spending pays for itself, and the question of who funded what at what time becomes very visible.

Right to Win. Challagalla and Dalsace describe Right to Win investments as “genuinely optional investments made by firms that believe they can win through sustainability.” The translation into finance is direct: investments made because the firm believes it can win through governance and simultaneously create value for customers. The scale of the opportunity is now quantifiable. BCG’s November 2025 From Branches to Bots report projects $370 billion in annual additional profit for retail banks by 2030, with AI-first banks achieving cost bases 30 to 40 percent lower than peers and profit gains of 30 percent or more. McKinsey’s parallel analysis estimates $200 to $340 billion in annual value for banking, which is 9 to 15 percent of operating profits.

The clearest 2026 illustration is Goldman Sachs’ six-month embedded engineering arrangement with Anthropic, which CIO Marco Argenti described to CNBC in February: Anthropic engineers worked alongside Goldman teams to build autonomous agents for trade accounting, transaction reconciliation, and client onboarding. Goldman deployed these agents because its governance program was sophisticated enough to absorb agentic systems without producing the kind of organizational anxiety that elsewhere keeps similar deployments stuck at the proof-of-concept stage. By May 2026, Goldman, Blackstone, and Hellman & Friedman had committed $1.5 billion alongside Anthropic to a forward-deployed engineering venture aimed at mid-market firms, with Goldman’s asset management arm contributing $150 million. JPMorgan, Citi, AIG, and Visa appeared on the same stage at Anthropic’s May briefing as production deployments.

JPMorgan’s own published figures show the trajectory. Its LLM Suite has been deployed to 230,000 employees across the firm, with 450 AI use cases in production and a path to more than 1,000 by year’s end. Coach AI improved advisor response times by 95 percent during market volatility and contributed to a 20 percent gross sales lift in asset and wealth management. CIO Lori Beer has tied engineer performance reviews to AI tool adoption for 65,000 staff, after measuring 10 to 20 percent productivity gains on the coding assistant.

The Right to Win move is to invest such that governance becomes the engine of speed. Goldman’s embedded engineering model produces customer-facing improvements, faster onboarding, trade accounting, and compliance review, that the institution can credibly defend to regulators because its underlying monitoring infrastructure was built to handle agentic systems from the start.

The deeper point, and the one most likely to be missed in board discussions of AI investment, is that Right to Play, Right to Stay, and Right to Win are not three options to choose between. They are three categories that must each be funded separately, with different decision rules and against different time horizons. The bank that funds all three as a single undifferentiated budget will systematically underinvest in Right to Win because the most visible reward sits with Right to Play.

The remedy is structural. Boards should ask their executive teams to separate the AI investment line into three buckets, each with its own metric. Right to Play measured against regulatory expectation; Right to Stay against resilience scenarios twelve to twenty-four months forward; Right to Win against specific customer outcomes that competitors cannot match without similar governance investment.

The firms outperforming in AI in 2026 are those that have understood, before the field as a whole, that governance is the catalyst of innovation that drives customer value. The incentive structure rewarding documentation over innovation can be redesigned only at the level of the board’s investment decision. The institutions that redesign it first will compound the advantage for years before the sector catches up.

The remaining question is whether bank boards will recognize this in time, or whether they will continue to fund AI governance as Right to Play spending alone, leaving Right to Win on the table for the small group of institutions already moving.