In most banks, judgement is still described as a human act. It sits in committees, is recorded in minutes, and is assumed to happen at identifiable decision points.
But in day-to-day operations, that picture no longer holds. Judgement has moved. Not into intelligence or intent, but into structure. It now lives inside systems.
This shift did not begin with artificial intelligence. Long before generative tools entered production, institutions had already begun delegating judgement in smaller, quieter ways. Credit thresholds replaced discretionary review. Rules engines replaced policy interpretation. Exception queues replaced escalation. Straight-through processing replaced human pause.
Each change improved speed and consistency. None of them appeared to diminish responsibility. Yet taken together, they altered where decisions are actually made. In most production environments, outcomes are determined long before anyone clicks “approve”. They are set when thresholds are defined, when rules are encoded, when workflows are designed, and when escalation paths are structured. By the time a case reaches a human, the path has largely been fixed.
Many institutions still describe these environments as human in the loop. On paper, that sounds reassuring. In practice, the human role is often procedural. Staff confirm outputs, handle defined exceptions, and carry accountability without genuine discretion.
The judgement did not occur at the moment of sign-off. It occurred earlier, during configuration.
This is why the shift feels different from traditional automation. Automation replaced effort. This replaces ownership.
When effort is automated, the organisation still knows who exercised judgement. When judgement becomes infrastructural, responsibility fragments. No single person can point to the moment where the decisive call was made. There is only a sequence of technically correct steps.
Post-incident reviews often expose this gap. Everything worked as designed. No policy was breached. Controls operated within tolerance. Yet the outcome is difficult to defend in human terms.
Artificial intelligence is frequently blamed for this dynamic. That is convenient but incomplete. AI accelerates delegation; it does not create it. The deeper issue is that institutions have allowed judgement to be absorbed into operating models without redesigning responsibility.
In older governance structures, judgement had a location. A committee room. A credit panel. A senior sign-off. It could be seen and, if necessary, challenged.
In modern systems, judgement is dispersed across configuration files, policy tables, orchestration layers, and vendor platforms. It is harder to identify and harder to question.
This becomes particularly significant in regulated environments. Regulators are not satisfied with technical correctness alone. They ask who decided. They ask why discretion was exercised or withheld. They ask whether the institution stood behind the outcome.
When judgement is embedded into infrastructure, those answers become abstract. Explanations grow technical. Responsibility appears collective. Legitimacy becomes fragile.
Practitioners recognise the moment when this tension surfaces. A decision is challenged. Logs are reviewed. Workflows are traced. Thresholds are confirmed. Everything appears compliant.
And yet the room is quiet.
No one can say clearly where judgement was exercised, or by whom.
Over time, this erodes institutional authority. Not visibly, and not immediately. Performance metrics may remain stable. Controls may appear intact. But the ability to defend decisions with confidence weakens.
Prepared institutions approach this differently. They do not abandon systems, nor do they romanticise discretion. Instead, they make explicit where judgement lives. They define which decisions are delegated, which require active human interpretation, and which can never be fully automated.
They treat judgement as a governance design choice rather than an incidental by-product of technology.
This does not slow institutions down. It clarifies responsibility. Thresholds are treated as provisional judgements rather than permanent truths. Exception processes are designed to restore discretion rather than merely increase throughput. Escalation pathways reassign authority, not just accountability.
Institutions that ignore this shift face a quieter risk. They do not immediately fail. They continue to operate and comply. Reports are submitted. Processes appear orderly.
But when confronted with scrutiny, they struggle to articulate who exercised judgement and why.
The core issue is not artificial intelligence, speed, or complexity. It is that judgement has been absorbed into infrastructure without being deliberately re-owned.
Once that happens, responsibility becomes symbolic rather than real. Institutions are left defending outcomes they no longer fully control.
The most important question for leadership is therefore not whether systems can decide. It is whether the institution still knows where judgement lives.
Because when judgement disappears into infrastructure, it does not return on its own.
Dr. Gulzar Singh, Chartered Fellow – Banking and Technology; Director, Phoenix Empire Ltd
“When financial judgement becomes infrastructure” was originally created and published by Retail Banker International, a GlobalData owned brand.
The information on this site has been included in good faith for general informational purposes only. It is not intended to amount to advice on which you should rely, and we give no representation, warranty or guarantee, whether express or implied as to its accuracy or completeness. You must obtain professional or specialist advice before taking, or refraining from, any action on the basis of the content on our site.
