According to latest reports, AI has moved from conversation to action, entering a phase of operational maturity where agents can interact with systems, query data, use tools, and make decisions under supervision. Businesses are no longer prioritizing merely chatbot rollouts but seeking to integrate cognitive systems into their value chain and optimize entire workflows. This evolution is particularly impactful in banking, where AI agents can help prepare reports, review data, generate commercial proposals, analyze risks, coordinate schedules, and automate internal processes. The third layer of AI strategy in banking - focusing on institutional operations including treasury management, compliance, and operations - may create the most durable competitive advantage as these applications require institution-specific work using their own data, workflows, and accumulated judgment. As Dave Murphy from Publicis Sapient explains, "If you get the context right, the workflow itself changes fundamentally. Rather than simply having an agent assist you, you as a human are instructing a set of agents to do the work."
AI in banking operates across three distinct layers with varying impact potential. The first layer handles customer-facing interactions through chatbots and personalized financial recommendations, where AI has moved fastest in major digital banks. However, as simpler interactions are automated, remaining complex tasks like disputes, exceptions, and reassurance cases become increasingly difficult to automate, limiting incremental gains. The second layer focuses on fraud detection and credit decisioning, where traditional fraud systems relied on fixed rules while newer models examine behavior across accounts and process thousands of signals simultaneously. JPMorgan reports that AI has enabled its transaction-screening operation to review more than twice the volume while cutting manual checks in half, though this creates an ongoing arms race with fraudsters using similar tools. According to Publicis Sapient, gen AI as autocomplete could provide up to a 15% improvement to workflows, though this represents only marginal gains compared to the transformative potential of agentic AI.
Credit decisioning represents another major AI use case at the transaction layer. Lenders can enhance static bureau scores with current views of income, cash flows, spending, and other signals to assess borrowers with thin credit histories. However, richer data does not automatically produce fairer decisions, as opaque models can make decisions harder to understand and exclude some borrowers more aggressively. This layer requires careful balance between improving decision accuracy and maintaining fair lending practices. The challenge lies in capturing undocumented institutional knowledge that experienced analysts possess but struggle to explain, as this undocumented expertise is crucial for AI systems to operate effectively at scale. As Dave Murphy notes, "The most important and powerful thing is the context you can provide to these agents." Fintechs that become genuine experts in their field and deliver improved solutions will be able to offer real added value that customers can't easily replicate themselves.
The third layer focuses on institutional operations including treasury management, compliance, and operations, which may create the most durable competitive advantage. In treasury operations, banks continuously decide liquidity requirements, deposit behavior, and capital deployment. A survey by BCG found that more than 80% of the largest global banks were using AI in some form within treasury, compared with around half across the sample as a whole. These applications include deposit modeling, liquidity forecasting, and cash-flow analysis, where marginal improvements in forecasting processes can significantly impact capital deployment decisions. The key difference between traditional assistants and AI agents lies in their ability to plan, retrieve information, use tools, link steps together and complete complex tasks - opening huge opportunities for organizations to integrate AI as an intelligent layer within day-to-day operations. As Dave Murphy explains, "This is an interesting time to be a fintech. You have to separate the hype from the reality." The most valuable AI applications require institution-specific work using their own data, workflows, and accumulated judgment, where shared computing capacity and sector-specific models cannot replace the need for context-specific expertise.
The banking sector has experienced unprecedented digital adoption acceleration, with digital adoption among high street banks rising from just 35-40% in 2014-2015 to 80-90% today. According to Publicis Sapient, this transformation was fostered by collaborative business communities and socio-economic events like the pandemic that changed operational models. As Dave Murphy notes, "For organisations that have been around for 350 years; these are long-standing institutions. They have become digital businesses whether they liked it or not – their customers pushed them into that model." Being digitally native comes with risks, as businesses feel vulnerable without human touchpoints like branches or telephones. However, the shift to agentic AI represents a fundamental change from historical AI approaches, where machine learning models focused on personalization and data monetization to agentic AI that enables true automation and workflow transformation. This evolution requires careful implementation of guardrails and platforms, with banks starting from back-office operations before introducing AI to customer-facing channels.
