Decision Fatigue Raises Loan Review Errors 23% After Hour 3

A credit officer at a mid-sized NBFC in Pune clears her eleventh loan file of the day. The first four took forty minutes each; this one takes twelve. The file is not obviously weaker than the ones she approved at 10 a.m., but it is thinner on promoter background, and the cash-flow statement has a small inconsistency she would have caught before lunch. She signs off. Somewhere in the next quarter's provisioning, that signature will reappear as a number. The question worth asking is not whether she was careless — she is, by every internal metric, a careful officer — but whether the quality of her judgment is a stable personal trait at all, or a resource that depletes predictably across a working day.

The Architecture of Depletion

The idea that self-regulatory capacity behaves like a finite fuel tank entered mainstream behavioral science through Roy Baumeister's ego-depletion experiments in the late 1990s. Participants who performed an initial act of self-control — resisting cookies, suppressing emotion, making a series of consumer choices — subsequently persisted less on unsolvable puzzles and gave up faster on difficult tasks. The metaphor was hydraulic: willpower as a reservoir drawn down by use and replenished by rest, glucose, and time.

That literature has since been contested. A large 2016 multi-lab replication found a much smaller effect size than the original studies suggested, and the field has largely moved away from the crude "willpower is glucose" story. But something more durable survived the replication crisis: the observation that decision quality degrades with decision volume, particularly when the decisions are frequent, low-feedback, and consequential. This is not the same claim as ego depletion. It does not require a single underlying resource. It only requires that attention, working memory, and motivation are all finite within a day, and that complex judgments consume all three.

For loan review, the distinction matters. A credit decision is not a single act of willpower. It is a bundle of sub-tasks: reconciling documents, reconstructing a promoter's history, stress-testing a repayment schedule, weighing collateral against sector risk, and then committing to a yes or no. Each of these draws on working memory. None of them has a clean stopping rule. This is precisely the profile of task where fatigue shows up not as obvious error but as simplification — the officer stops generating alternatives and starts pattern-matching to the nearest familiar case.

Why Hour Three Is the Inflection Point

The "23% after hour three" figure circulates in credit-risk circles and should be treated as illustrative rather than precise — but the shape of the curve it implies is consistent with what is known about vigilance and time-on-task. Research on sustained attention, going back to Mackworth's clock-watching studies during the Second World War and continuing through modern psychomotor vigilance testing, consistently finds that performance on monotonous monitoring tasks declines measurably within the first two to three hours, with the steepest drop occurring in the transition from a warm-up plateau to the fatigue phase.

Loan review is not monotonous in the way radar watching is. But it shares the critical feature: the stimulus is intermittent and the response is repetitive. Between files, the officer waits, or reads email, or attends a call. The cognitive set required for each file is expensive to reload. By the third hour, most officers have stopped fully reloading. They are running on the residue of the previous case.

The Variable-Ratio Trap

There is a second-order problem that makes this worse, and it comes from reinforcement learning rather than attention research. Credit approval is, structurally, a variable-ratio schedule: most approvals perform, a minority default, and the officer cannot predict which. Variable-ratio schedules produce high, steady response rates and are notoriously resistant to extinction — the same property that makes them central to behavioral psychology's account of persistent behavior under uncertainty.

In a training context, this is usually framed as a virtue: it produces engaged officers. But it also produces something subtler. Because the feedback is delayed by months and noisy, the officer never receives a clean signal about whether her third-hour judgments were worse than her first-hour judgments. The portfolio performs, more or less. The individual decision disappears into the aggregate. There is no error message. The fatigue is invisible to the person experiencing it, which is exactly why it survives.

Loss Aversion, Under Time Pressure, Does Something Unhelpful

Kahneman and Tversky's prospect theory established that losses loom larger than equivalent gains — a finding replicated across dozens of domains and cultural settings, including studies of Indian retail investors and micro-entrepreneurs. In credit, loss aversion is usually discussed as a conservative force: officers reject marginal files to avoid the pain of a default.

Under fatigue, the effect inverts in an interesting way. Loss aversion is cognitively expensive. It requires the officer to hold two possible futures in mind simultaneously — the one where the loan performs and the one where it sours — and to weight them asymmetrically. A tired reviewer does not do this. She collapses the two futures into one, usually the one most available to memory. If the last three files in her queue were approved and nothing bad has happened yet, the available future is the good one. The asymmetry disappears, and with it the very caution that loss aversion is supposed to supply.

This is the mechanism by which decision fatigue and loss aversion interact: fatigue does not simply make officers worse, it makes them less asymmetrically cautious at precisely the moment they need the asymmetry most.

A Concrete Illustration

Consider a regional rural bank in Karnataka that audited 400 agricultural loan appraisals over a six-month period, tagging each by the time of day it was completed and by whether the file was later classified as a non-performing asset. Suppose the analysis showed NPA rates of 3.1% for files cleared before noon and 4.4% for files cleared after 4 p.m. — a 42% relative increase. That specific study does not exist in exactly this form, but the pattern is common enough in internal audit data that several Indian banks have quietly moved to morning-weighted committee schedules. The mechanism is not that afternoon officers are lazier. It is that the marginal file — the one that requires one more phone call, one more document, one more question — gets cleared rather than escalated.

What Training Programs Can Actually Do

The instinct in banking L&D is to respond to findings like this with more training: modules on credit appraisal, refreshers on ratio analysis, workshops on early warning signals. This misses the point. If the error is driven by depleted working memory, additional knowledge does not help — the officer already knows the ratio. What helps is changing the structure of the decision, not the content of the knowledge.

Three interventions have reasonable support:

Hard stops and batching. Splitting the day into blocks of no more than two hours of continuous review, with a genuine break between blocks, is the most direct application of vigilance research. The break must involve a change of cognitive mode, not email.

Checklists with mandatory escalation triggers. Atul Gawande's work on surgical checklists showed that the value of a checklist is not that it teaches anything new — it is that it forces the same sequence regardless of the operator's state. A credit checklist that requires an explicit written justification for any file with a debt-service coverage ratio below a threshold does not depend on the officer noticing the problem. It depends only on the officer following the rule.

Separating the "no" decision from the "yes" decision. Some banks now route marginal files to a second reviewer whose only job is to reject or refer. The asymmetry of the role — the second reviewer is not measured on volume — restores the loss aversion that fatigue erodes.

The forward question for Indian credit institutions is not whether to adopt these. It is whether the internal audit function is currently configured to detect time-of-day effects at all. Most loan review logs record the date, not the hour. Until that changes, the 23% remains an anecdote, and the officer in Pune continues to sign.