Banks Deploy AI in Financial Decision-Making Without Redesigning Core Systems

Banks Layer AI Onto Legacy Systems Without Fundamental Redesign

For decades, banking technology followed a straightforward pattern: customers made financial decisions, and banks executed them. Artificial intelligence now threatens to reverse that sequence entirely. Instead of waiting for human instructions, AI in banking can interpret customer goals, understand financial context, and determine next steps autonomously. It could detect an impending cash shortfall, identify solutions, and potentially execute appropriate actions without prompting1

1

McKinsey estimates that generative AI could generate $200 billion to $340 billion in annual value for the banking industry. Yet most institutions are adding artificial intelligence to systems designed for the old model rather than rebuilding around AI's capabilities. The problem extends beyond technology—AI inherits existing product silos and process boundaries, creating another layer without enabling the cross-system financial decision-making needed to realize its full potential1

1

Source: TechRadar

AI Deployment Outpaces Governance Maturity

While 54% of banks now deploy live AI models, barely 12% report having mature governance and approval systems in place, according to the 2026 ProSight CRO Outlook Survey conducted with Oliver Wyman. AI deployment is scaling faster than oversight capabilities. When flawed data, hidden bias, or hallucinated outputs infiltrate live workflows, algorithmic errors escalate into regulatory inquiries or public relations liabilities before anyone identifies who possessed authority to intervene2

2

Most financial institutions treat this as a model-risk or IT issue. In reality, it represents a management delegation problem. When banks organize AI governance by functional silos—finance, HR, risk, operations—they miss a crucial detail: individual decisions within the same department carry radically different operational impact. The critical question shifts from "How much can we automate?" to "How much decision-making authority should we delegate to an algorithm?"2

2

Current AI Use Cases Focus on Process Improvement

The most visible applications of artificial intelligence in banking remain the easiest to deploy. Banks use AI agents to improve customer service, automate fraud detection, personalize recommendations, summarize documents, and accelerate credit decisions. Lloyds Banking Group rolled out more than 50 AI use cases across the group in 2025, generating around £50 million in value, with more than £100 million in additional value expected in 20261

1

However, McKinsey observed that simply adding AI on top of existing processes will not produce transformational change and can instead create another layer of technical debt. The difference between AI-native architectures and a chatbot attached to existing systems can be tested with three questions: Can the AI choose and coordinate actions within defined guardrails, trusted data sources, and approved tools? Is the process designed for the agent to act with human oversight and every decision recorded? Are permissions, monitoring, and regulatory controls built into the workflow, particularly for compliance and fraud prevention?1

1

From Product-Centric to Outcome-Centric Systems

Significant transformation begins when banks start with customer objectives rather than banking products. Consider a customer wanting to maintain liquidity while earning maximum returns on excess cash. An intelligent banking system could continuously monitor balances, upcoming payments, income, available credit, and other relevant information, then determine whether money should remain liquid, be invested, or used to reduce borrowing1

1

This approach doesn't require customers to surrender control immediately. Adoption can begin with low-risk actions like moving excess cash into savings or setting aside VAT for future tax payments before expanding into more consequential human decision-making. Deutsche Bank deployed an agentic AI system for third-party risk management where several AI agents retrieve relevant controls, analyze documentation, and propose assessment outcomes. Human assessors remain responsible for reviewing or overriding recommendations1

1

Related Stories

Financial Institutions Face Rising Pressure to Strengthen AI Governance Amid Regulatory Concerns

AI's Transformative Impact on Banking and Financial Oversight

AI Revolutionizes Banking: Enhancing Customer Experience and Risk Management

AI Agents Trigger Security Crisis: 65% of Enterprises Report Incidents Amid Scaling Challenges

Three-Mode Framework for AI Authority

The answer to delegation hinges on two variables: operational impact of wrong decisions and ease of reversal. This yields three distinct modes for financial AI: Act, Recommend, or Advise. In "Act" mode, AI automates reversible analysis like payroll-tax forecasting and headcount expense calculations. Human oversight focuses on exception handling rather than manually recalculating every spreadsheet row2

2

In "Recommend" mode, as consequences grow and reversibility becomes difficult, the model's mandate shrinks. AI can analyze operational capacity, attrition rates, and hiring timelines, then suggest reallocating headcount. However, algorithms miss critical factors invisible to models—pending regulatory commitments, skill gaps, or strategic growth programs. Leadership retains full veto power, and organizations should track how often managers override recommendations2

2

In "Advise" mode, decisions regarding equity incentive architecture, organizational restructurings, and major strategic shifts belong strictly to human leadership. AI tools can simulate dilution impacts and model retention probability faster than legacy spreadsheets, but algorithms cannot shoulder accountability for how modified structures change human behavior or whether staff view compensation changes as fair2

2

Infrastructure Shifts Toward Agent-Initiated Transactions

Banks are evolving into continuous decision systems that intelligently manage financial activity to accomplish defined objectives. This structural shift extends beyond traditional banking. Visa and Mastercard are both building infrastructure for AI-initiated payments, allowing agents to act on behalf of consumers and businesses. Visa Intelligent Commerce is designed to let AI agents find and purchase products on users' behalf, with tokenized credentials, authentication, and spending controls built into the payment flow1

1

As agents move from recommending actions to executing them, banks must ensure transactions remain within customer intent and risk tolerance. Like a new employee, an AI agent should receive defined permissions. Transparent activity logs, alerts, approval thresholds, and override capabilities make this principle visible in products and enforceable by regulators. The institution remains responsible for keeping agents within established guardrails1

1