- Key insight: The biggest barrier to effective adoption of AI within banks is not the capability of the AI systems themselves, but the failure to recognize that the technology is only as good as the underlying systems it helps automate.
- Supporting data: In a recent Capgemini report on financial services, only 10% of companies reported being able to roll out AI agents at scale.
- Forward look: The winners will be those whose operations, people and decision-making combine to create an integrated human-AI model.
Conversations about
The challenge is operational.
Almost every large bank today is running AI pilots. They have
The instinct is to question the technology first. This threatens to mask a bigger problem. The banks seeing the most success with the technology today are not those
Too much of the industry's AI debate treats operational readiness as a checkbox within the technology itself. This is a mistake. Operational readiness is key to success and is the product of trusted data, connected processes, integrated technology, accountable governance and empowered teams. Only when these elements come together can banks deliver on the promise of AI.
AI is much closer to being an MRI than a magic wand. It exposes structural weaknesses most banks avoid addressing because they've spent years working around them. With AI, challenges such as siloed teams, fragmented handoffs, poor-quality data and slow processes become even more pronounced.
Give AI clean, consistent data within well-run processes, and it will return actionable decisions. Unleash that same intelligence across fragmented processes shaped by legacy handoffs, and it simply scales the inconsistencies.
A classic example is a credit decisioning engine trained on incomplete or outdated data. It may process faster, but lending decisions won't suddenly improve. Deploying AI on weak operational foundations doesn't merely fail to add value; it amplifies the cost and risk already entrenched within the process.
Discussions on AI readiness should begin with data. Too often, they end there. When handled responsibly, AI can also help banks on their path to clean and structured data. The data isn't ready when it's perfect; it is ready when it can be trusted.
However, AI doesn't operate on data alone. It operates within processes, governance structures and decision networks. All too often, AI takes over the silos already present. Lending, servicing, fraud and compliance optimize their individual workflows, unaware that the customer experiences a single institution rather than a handoff between tasks.
The real transformational opportunity lies in redesigning the value chain, rather than continuing to optimize individual components. Take customer onboarding. The customer doesn't experience identity verification, KYC, document validation and risk assessment as separate steps. Unless AI supports them throughout the journey, banks risk improving processes without fundamentally shifting the needle for the customer.
This is where the real opportunities start. With these foundations in place, AI can take over routine operational work, allowing people to focus on delivering valuable advisory support.
Read more:
Should a bank ever be liable when a customer gets scammed ?OpenAI models' Hugging Face breach is a red flag for bankers The top-performing banks with $2B to $10B of assets in 2025 Inside Upstart's journey to get a bank charter approval
One assumption I would challenge is that speed matters more than sequence. Companies often rush to introduce AI before simplifying the work around it. In reality, the best results often come when processes have already been standardized and digitized, and when the technology fits the work being done. Routine and rules-based work often benefits more from intelligent automation and document processing, while AI delivers the greatest value where judgment and context are needed.
Even the strongest AI solutions fail when they do not connect cleanly to the underlying systems, leaving the output marooned without being embedded in the decision engines and transaction systems where all value lies.
Most AI failures are designed into the system. The culprit is not the model or some post-deployment governance oversight, but a flaw introduced at the system's conception that later manifests itself during operation. Regulations, policies and customer behavior are in constant flux. Thus, models require perpetual monitoring and retraining. AI at scale is not a one-time implementation exercise. It is an ongoing operational discipline.
Governance is often portrayed as something that slows AI innovation. I believe the opposite is true. Explainability and accountability are not barriers to innovation; they are core requirements for AI-driven decision-making to become trustworthy and repeatable.
As AI increasingly performs routine maker-checker activities, oversight moves away from duplicated manual checks to an exception-based approach. AI can trigger new cases, validate documents and approve transactions, while people remain accountable for scenarios that require judgment. The future banker will not be replaced by AI, but will devote less time to interpreting data and more to applying thought and judgment — improving overall outcomes for the institution.
That is when AI stops being a technology project and starts becoming business transformation. Automation eradicates chores; augmentation improves decisions.
The debate about AI is no longer about its adoption, but about why there continue to be challenges to realizing value at the enterprise level.
The answer lies less in access to better AI and more in operational discipline. What stands out among high-performing banks is the confluence of business alignment, strong data foundations and solid operational execution. That operational discipline will become even more important as copilots, integrated customer data and automated onboarding become the industry standard.
Multi-agent systems are making their presence felt across the banking ecosystem in financial crime, credit and servicing. Scaling autonomous AI in these areas will require careful consideration of use cases, in which low-risk, repetitive, high-volume tasks will help banks realize value, while leveraging the human workforce as the last check in high-risk scenarios.
AI is available to everyone but building an institution ready to harness it will be key. The winners will be those whose operations, people and decision-making combine to create an integrated human-AI model.
