Banks are adopting AI faster than they are redesigning the processes around it. Accenture research cited in the article found that 84% of banking leaders have seen at least moderate business value from AI, but only 20% say their investments are producing broad and sustained value. Michael Abbott, Accenture’s global banking lead, says much of the industry is essentially putting AI on top of existing workflows because of competitive pressure and fear of missing out.
The bigger opportunity is parallelizing work that was traditionally sequential. Instead of moving a mortgage application step-by-step through document collection, financial analysis and credit evaluation, separate AI agents could work on different parts simultaneously and communicate with one another. The same idea could apply to customer-service operations: rather than building one universal AI agent, banks could organize systems around individual customer intents—such as a lost card—and make those capabilities available across mobile apps, online banking, call centers and branches.
The article also makes an important distinction between frontier models and specialized AI. Banks do not necessarily need the most powerful large language model for every task. For deterministic or narrowly defined work, smaller specialized models can be cheaper, more predictable and less prone to hallucination. That matters particularly in banking, where outputs need to be documented, controlled and understandable to regulators.
The broader takeaway is that AI's biggest banking gains may come from organizational redesign rather than better chatbots. Simply automating individual steps can produce little improvement—or even reduce productivity if the overall workflow remains unchanged. The real transformation happens when banks rethink the entire process around what AI agents can do in parallel, while humans provide oversight, judgment and the ability to recognize when an AI system is wrong.
