Speed isn't the problem: Using AI in Banking

One of the main strengths of AI is speed. It enables organizations to move faster, automate tasks, and increase output. When used well, this can reduce manual workload, speed up repetitive processes, and improve overall operational efficiency. In other words, everything is happening faster, and with less direct oversight.

So, what’s the problem with the speed of AI?

In regulated environments such as banking, the role of speed is less straightforward. Systems and processes still need to meet strict requirements for security, compliance, and auditability. As the pace of change increases, it raises a different question not just how quickly things can be done, but what might change operationally once AI becomes part of the process.

Consider an operational incident where responders previously triaged, diagnosed, and documented root cause through deliberate investigation. With AI correlating logs and metrics and proposing a root cause, resolution and write-up speed up. But under DORA the resilience obligations and the integrity of the incident record sit with the organization. The challenge is if the AI's proposed root cause is plausible but wrong, does the process still force someone to test it, or does the convenient answer become the official one?

If AI does the work, who can be held responsible?

Banking and other heavily regulated industries share one common denominator; they are built and operate around accountability. If an AI-supported process leads to an incorrect decision, compliance issue, or operational failure, responsibility still remains with the organisation. Using AI for automation cannot remove process ownership. It only changes how parts of the process are executed.

AI-driven automation introduces new operational risks

Speed also scales existing weaknesses

Speed achieved through automation only solves a problem if the problem was speed to begin with. If operational issues already exist due to weak processes, unclear ownership, or inconsistent oversight, automating those processes can simply make the same problems scale faster. In regulated environments such as banking, this can become especially difficult when parts of the process previously relied on informal human oversight, people noticing inconsistencies, questioning unusual outputs, or slowing things down when something seemed wrong. As processes accelerate, those safeguards can become easier to overlook or remove entirely.

Automation can also change the process itself

If human judgement was originally part of the process, replacing parts of that judgement with AI does not simply speed up the process, it changes how the process operates. The challenge in regulated environments is that this shift may not always be obvious. What appears to be optimization can also change how control and oversight are applied in practice.

Speed requires operational clarity

‍

For AI-driven automation to work reliably in regulated environments, oversight and control mechanisms need to be clearly built into the process itself. If parts of the process instead rely on informal human judgement, experience, or people reacting when something seems wrong, those safeguards can easily disappear once the process is accelerated or automated with AI.

Implementing AI for the purpose of speed or efficiency requires organisations to understand whether critical oversight relies on informal human judgement rather than the process itself and whether that oversight can realistically be formalized.

This is precisely why we are building oversight into how AI is governed at Crosskey, through clear ownership, formalised controls, and forums whose job is to ask these questions before a process is automated. The goal is not to slow AI down, but to make sure that when a process speeds up, the oversight that kept it safe speeds up with it, by design.

‍