Experts are supposed to know what they’re doing. Right?
I think that’s fair. They’re experts after all.
What I think we do though, is stretch that further than it should go. One way we do this is to think that ‘experts have deep knowledge and skill’ means ‘experts should know everything about their field’, even if they’re a specialist. But also, which interests me more, they should be able to recall every critical step, in sequence, under pressure, while interrupted, with incomplete information, and without looking anything up.
Those are three different claims. The first is true. The second though? Does a doctor really know everything about medicine? The third one is involved, and a happy little journey into cognitive research.
The strange and unusual
Critical events are cognitively awkward in a specific way.
They’re rare, so people get few rehearsals. They’re time-pressured. They generate arousal. They interrupt and compete with everything else happening. That’s close to a definition of the conditions under which prospective memory, remembering to do the intended thing at the right moment, becomes unreliable. The person knows the step. They just may not retrieve it when it matters. Which is why “but they were trained” can be an incomplete answer after an omission. Training puts knowledge into memory. It doesn’t guarantee retrieval under every imaginable condition. (Note: training should include a variety of conditions…)
Side note: A Professor of General Practice told me once that GPs see everything infrequently. Keeping track of a vast array of things and recognising something they come across only every so often gave me a new appreciation for how challenging a GP’s job can be.
How we can support the rare occasions
A cognitive aid is any deliberately designed external support for a task: a checklist, an emergency algorithm, a decision prompt, a status board, a structured handover tool. Used badly, they can be annoying or just clutter. Used well however, they can move critical steps out of the least reliable storage medium on hand at the time. That is, they can take what might reside in someone’s memory and make it visible
Two parts to it.
Greig and colleagues reviewed 13 randomised trials of cognitive aids in clinical emergencies, published in Anaesthesia in 2023. Most were run in simulation. Across the six trials that could be pooled, clinicians with an aid missed 11 percent of required steps. Without one, they missed 43 percent.
That’s a large effect, but as they were mostly simulations, fair to say it shows a direction of travel, not a guarantee in a specific circumstance. I say that because of a pivotal counterfinding. In two airway simulations, the aid slowed teams down.
Then there’s Urbach et al.
The Ontario scenario
Haynes and colleagues reported in 2009 that a surgical safety checklist rolled out across eight hospitals worldwide roughly halved deaths and cut complications by a third. Five years later Urbach and colleagues examined what happened when Ontario mandated surgical checklists across 101 hospitals and more than 200,000 procedures in the three months after adoption. There was no significant improvement in mortality or complications. The checklist that was effective in pilot sites did little when imposed at scale as a compliance requirement.
Together, we get the good bit.
The paper itself is not the intervention. A checklist works when it’s designed for a real failure mode, integrated into how the team operates, and used by people who understand what it’s for. Mandated tick-boxes produce tick-box behaviour. I wrote something similar about speak-up channels. You’ll likely have your own examples.
Design is important, and of more value than what the word ‘checklist’ suggests. Alfred and colleagues make this point in a 2024 review: the tool must match the failure it exists to prevent, the experience of the user, and the way the task is actually performed.
That last sentence includes fantastic advice from Alfred and co.
For example, how a job is actually performed is another way of saying ‘work as done versus work as imagined’, or how a policy or procedure might differ from the frontline reality of doing the job in real-world situations. There are significant bodies of work and literature around just that sentence.
Aids must be fit for purpose
An aid, then, must hit its targets.
A novice may need a procedural guide. An expert may need only a short verification list at a transition point. A crisis team may need an emergency algorithm that stops one rare, fatal omission. An executive team may need five questions asked before every major commitment. Calling all of these “checklists” hides the design decisions that determine whether they work. By extension, we should put effort into what we’re doing and why.
There’s how the aid is implemented, too.
Hannenberg adds the implementation piece in a 2020 review. People need to know the aid exists, where it lives, when to trigger it, and how to use it as a team rather than privately. An aid one person uses while everyone else operates from a different frame may create more problems. Good aids structure communication, not just memory.
Which brings me to the resistance
There can be status in unaided performance. Reaching for the aid can feel like admitting you don’t know the job, and in some places speed and memory can be seen as proof of mastery. That’s backwards in high-reliability work. A pilot using a checklist isn’t less expert for refusing to bet the aircraft and all the souls on board on prospective memory. The purpose of expertise is the outcome, not a memory stunt.
Senior leaders, meanwhile, routinely make complex, consequential decisions with no equivalent of a cockpit check. Picture it: a tense meeting, a long paper to finish, not enough time, and the CE wants an answer. Would five questions, used consistently, change more decisions than most governance training? Probably.
- What problem are we actually solving?
- Which assumption is carrying the most weight?
- What evidence would change our mind?
- What has changed since this paper was written?
- Who sees it differently, and have we heard them?
It’s short, consistent and tied to known failure modes. It’s the same engineering the clinical world uses.
One more thing an aid can do, and it’s the reason I use this inside my psychological safety work. A good aid gives the most junior person in the room legitimate permission to challenge. “We haven’t completed item four” is something anyone can say. “I think you’re making a mistake” is not. Good systems reduce the interpersonal courage required to prevent an error, notwithstanding that individual courage is the wrong thing to build a safety system on anyway.
What I’m not saying
I’m not saying we should externalise everything.
An aid that requires more attention than it saves is a bad aid, and no checklist can interpret the situation in front of you, for you. The candidates for external support are specific: rare steps, transition checks, stop criteria, escalation triggers, handover essentials, and the questions that precede irreversible decisions.
Real expertise knows what humans do well, pattern recognition, judgement, adaptation, and what we do less well, which is remembering every intended action in order while the world around you might be on fire.
Use the brain for what it’s good at. Put the rest where you can see it.
Photo by Austin Distel on Unsplash