Original text here from Patrice Bernard (LinkedIn)
Following the recent announcement of its partnership with Google on artificial intelligence, Lloyds Bank has joined the now-standard chorus of institutions touting the number of AI use cases they've rolled out. The bank even goes a step further: its Chief Data and Analytics Officer boldly states a goal to “enable the whole bank with AI.”
It’s the perfect topic for a Friday evening reflection: how the decades-old utopia of a fully automated bank—first dreamed up when computers began entering financial institutions—still survives today. It was revived with the rise of Robotic Process Automation (RPA) a few years ago, and now finds new momentum in the promises of agentic AI.
Subscribe to our newsletter:
But reality has always been stubborn. Spend time inside any large organization, and you'll quickly discover hidden corners where manual, low-value tasks persist—kept alive by inertia, complexity, or sheer neglect. These processes, handled by dozens of employees, continue to resist every wave of technological transformation. So why would AI succeed where its predecessors failed?
To answer that, one must analyze the root causes of past failures. Chief among them: the heterogeneity of IT systems. In many cases, the infrastructure includes platforms so outdated they require context-switching and handovers that only a human brain can manage. Integration in such environments is rarely seamless, and often impractical.
Yes, AI agents could theoretically overcome these barriers—just as RPA tools once promised to. But here lies the catch: the economic equation rarely adds up. While new solutions may be easier and cheaper to implement than older ones, operational costs (including technical debt) tend to rise in parallel, neutralizing the efficiency gains.
At the core of this dilemma lies layer upon layer of legacy IT, accumulated over decades, which limits the scope for true automation. This is where FinTech startups hold a clear advantage: by building from scratch on modern, coherent platforms, they can optimize operations with far greater agility.
Unless legacy institutions commit to replacing obsolete components, they may be forced to accept the persistence of manual tasks—not as inefficiencies to be eradicated, but as enduring realities in the process chain. In short, the dream of a fully AI-driven bank remains just that: a dream, still haunted by the past.
