A national telephone triage service logged more than 230,000 real medical judgments, and researchers used the near-random way shifts were scheduled to test one of the most repeated ideas in modern management: that the quality of your decisions decays as you make more of them. The result was not a small effect. It was no effect. Across every main test, the Bayesian models returned a Bayes factor above 22 in favor of the null โ strong evidence that decision fatigue did not degrade judgment quality at all (Nature Communications Psychology, 2025).
That matters because decision fatigue is not just a wellness talking point. It is a load-bearing assumption inside how most operations teams are designed. If you schedule hard approvals for the morning, cap how many sign-offs a manager handles in a day, or delegate to "protect the decision budget," you are running your process on a finding that just failed its largest field test to date. The throughput drag you are managing around is real. The story you have been told about its cause is probably wrong.
The Assumption Buried in Your Approval Process
Walk through almost any mid-market operation and you will find decision fatigue quietly baked into the workflow. Escalations are routed to be resolved before lunch. Approval gates are justified as a way to keep a senior manager from "burning out their judgment" on low-stakes calls. Delegation rules are framed as budgeting a finite daily supply of willpower. None of these are crazy. They all rest on the same premise: that a decision-maker is a depleting battery, and that the tenth hard call of the day is worse than the first because the reservoir is running low.
The premise is intuitive, which is exactly why it is dangerous. It feels true, so it rarely gets tested against your own data. And because it feels true, it justifies process overhead โ extra hand-offs, artificial time windows, deferred decisions โ that carries a real cost in cycle time. When a Head of Operations at a 50-to-500-person company adds a day to an approval because "we don't want fatigued sign-offs," that day is a measurable drag defended by an unmeasured assumption.
What 230,000 Triage Calls Actually Showed
The 2025 study is the strongest evidence yet against that assumption, and its design is what makes it hard to dismiss. Most prior decision-fatigue research is observational, retrospective, and non-preregistered โ you look back at outcomes and infer that ordering caused them. This study did the opposite. Because the triage service scheduled clinicians in a way that made their position in a sequence effectively random for parts of the data, the researchers could isolate the effect of "how many decisions you've already made" from the effect of "which cases happened to land in front of you" (Nature Communications Psychology, 2025).
They ran Bayesian generalized mixed models across more than 230,000 judgments and found one-sided Bayes factors above 22 for all main tests โ meaning the data were more than 22 times more consistent with no decision fatigue than with the effect. The authors are careful, as good researchers are: they do not claim decision fatigue can never exist in any context. But they conclude it does not hold up as a domain-general effect for sequential decisions, which is precisely the version that operational design relies on. If it does not reliably degrade the judgment of triage nurses making high-stakes calls at volume, the burden of proof shifts to anyone claiming it reliably degrades approvals in your workflow.
The Hungry Judges Were Never Hungry
Almost every article you have read about decision fatigue traces back to one 2011 study of Israeli parole boards. It reported that the share of favorable rulings started a session around 65% and collapsed toward zero before a food break, then rebounded afterward โ a graph so clean it became the canonical proof that tired deciders make worse decisions (Danziger et al., PNAS, 2011).
The problem is that the graph has a second explanation that requires no fatigue at all. A later analysis showed that the same dramatic pattern can be reproduced by a purely rational judge who simply schedules cases in a particular order โ for instance, taking cases likely to be quick or unfavorable as a session runs long, and clustering represented or stronger cases early. Once you model realistic case ordering, the magnitude of the "fatigue" effect is substantially overestimated, and much of it dissolves into an artifact of sequencing, not willpower (Glรถckner, Judgment and Decision Making, 2016). The judges were not depleting. The docket was structured.
That is the through-line worth holding onto. The most cited evidence for decision fatigue may be measuring the order cases arrive in, and the largest field test finds no depletion effect once you control for that order (Success, 2026). Both point at the same culprit, and it is not the decider's stamina. It is the structure of the work in front of them.
If It's Not Willpower, What Is Dragging Your Throughput?
Here is the part that should change how you run your process. The slowdown you observe late in a decision cycle is real โ approvals do get slower, calls do get worse under load. What the evidence reframes is the cause, and the cause points to levers you actually control:
- Case complexity. Hard decisions are slow and error-prone because they are hard, not because they come tenth. A cluster of complex cases back-to-back will degrade throughput whether it lands at 9 a.m. or 4 p.m.
- Ambiguous defaults. When there is no clear "what happens if we do nothing," every decision is rebuilt from scratch. Ambiguity, not depletion, is what makes the tenth call feel heavier than the first.
- Interruption load. Context-switching and fragmented attention degrade judgment directly. A manager pulled across twelve threads is not fatigued from deciding โ they are paying a switching tax on every call.
None of these three respond to a nap, a snack, or a protected morning block. They respond to redesign. That is the practical difference between the fatigue model and the evidence: the fatigue model tells you to ration decisions, and the evidence tells you to restructure them.
But We've All Felt Decision Fatigue
The fair objection: this contradicts lived experience. Everyone has felt worse at deciding by the end of a grinding day. Isn't a large null result just averaging away something that is obviously real?
Two things reconcile the feeling with the data. First, the study does not claim you never feel tired โ it claims that decision count itself is not what's degrading your judgment quality in sequential decisions. What you feel at 5 p.m. is far more plausibly the accumulated weight of complexity, ambiguity, and interruption than the mere tally of choices made. Second, "I felt it" is exactly the kind of retrospective, uncontrolled evidence the field has leaned on for a decade โ and it is precisely what the preregistered, near-random field design was built to correct. The point is not that end-of-day fog is imaginary. It is that blaming it on a depleting willpower reservoir sends you toward the wrong fix. You protect the calendar when you should be redesigning the docket.
What Mid-Market Ops Should Change This Quarter
The lever here is not your team's stamina. It is the sequencing and default structure of your approval process โ and unlike willpower, both are things you can edit. Three concrete moves, none of which require new tooling:
1. Re-sequence approvals by complexity, not by time of day. Stop defaulting hard calls to the morning on the theory of a fresh battery. Instead, batch by difficulty: cluster simple, high-default decisions so they clear fast, and give genuinely complex ones dedicated, unfragmented blocks whenever they occur. You are optimizing for the actual driver โ complexity and attention โ not for a phantom reservoir.
2. Pre-set default rules for the ambiguous cases. For every recurring decision type, define the "if we do nothing" outcome and the threshold that triggers an exception. Most of what feels like fatigue is the cost of re-deriving a judgment that could have been a default. A clear default converts a decision into a check.
3. Attack interruption load on your decision-makers. Protect decision blocks from context-switching, not from the clock. Fewer threads during an approval window will do more for judgment quality than any amount of rescheduling around an imagined depletion curve.
Then do the thing the researchers would respect: test it against your own approval-cycle data. You already have the timestamps. Look at whether error rates and cycle times actually track time-of-day and decision count โ or whether they track case complexity and interruption instead. The answer is measurable inside your own operation, and it will tell you which of your current process rules are earning their overhead.
The One Decision for This Quarter
Decision fatigue is one of those ideas that is too useful to be false โ it explains a real feeling, licenses reasonable-sounding process, and rarely gets checked. The largest field test to date checked it, across 230,000 judgments, and it did not survive. That does not mean your late-cycle slowdown is imaginary. It means you have been managing the wrong variable.
So the concrete decision for this quarter is narrow: pick one high-volume approval process, stop rationing it by time-of-day, and re-sequence it by complexity with a pre-set default for the ambiguous cases โ then check the change against your own cycle-time data. If throughput improves, you were never fighting fatigue. You were fighting a docket you had the power to redesign all along.
