Cognitive Biases at Work: How Your Brain’s Data Filter Distorts Decisions

Most advice about better decisions assumes the problem is your reasoning. Think harder, weigh the options, be more rational. I want to start somewhere less flattering. By the time you begin to reason, the data you are reasoning about has already been filtered, weighted, and quietly rewritten — and you were not consulted.

I spend my working days looking at brain scans, and one fact organizes everything I do: the brain is not a recording device. It is a preprocessing pipeline. It discards most of what reaches the senses, amplifies a little of what remains, and hands your conscious mind a tidy summary that feels like raw reality but is nothing of the sort. In signal-processing terms, the filter runs before the analysis. Your judgment is only as good as the signal it receives, and the signal has already passed through hardware that evolved to be fast, not accurate.

Cognitive biases are the predictable distortions that filter introduces. They are not stupidity, and they are not something you can outgrow. They are the shape of the instrument. This post is about four of the most expensive ones at work — where they come from, what they cost, and the concrete correction for each. Debiasing, done properly, is just thinking like a scientist turned on your own head: treating your first read of a situation as a hypothesis, not a conclusion.

Why the distortion happens before you think

It helps to be precise about the mechanism, because the correction depends on it. Your senses take in far more information than your brain can process. So it does not process most of it. Attention, memory, and expectation act as filters — deciding what gets encoded, what gets recalled, and what gets ignored — and those filters are tuned by what you already believe, what you saw most recently, and what is easiest to bring to mind.

EVIDENCE GRADE: STRONG

That the brain aggressively selects and reconstructs rather than records is one of the better-established facts in the field. Perception is inference; memory is reconstruction, not playback. What is *not* strong is any claim that a particular bias operates identically in everyone, in every setting, at a fixed magnitude — that is where popular writing overreaches. Grade the mechanism Strong and the specific effect sizes Moderate, and you will be honest.

The practical consequence is unusual. You cannot fix a preprocessing error by reasoning more carefully downstream, any more than I can recover a signal my recording equipment never captured. The correction has to intervene at the filter — by deliberately feeding your judgment the data it would otherwise have discarded. Each fix below does exactly that.

Confirmation bias: the filter that only lets agreement through

Confirmation bias is the tendency to notice, believe, and remember evidence that supports what you already think, and to quietly discount the rest. It is the master bias, because it operates on the input stage directly. Once you hold a view, your attention starts working for it.

At work: You decide early that a new hire is a strong performer. From then on, their wins register vividly and their mistakes read as bad luck or someone else’s fault. Six months later you have a thick file of confirming evidence and a confident, possibly wrong, opinion. The file feels like data. It is a filtered sample.

The correction: Ask, in advance, what evidence would change your mind — and go looking for that specifically. In my field we call this trying to falsify the hypothesis rather than confirm it; it is the single most useful habit I know. Before the performance review, write the sentence: *”I would revise my view if I saw ___.”* Then search for that one thing. You are forcing the filter to pass the data it was built to drop.

Anchoring: the first number owns the conversation

Anchoring is the pull of the first figure you hear. Once a number is on the table, every later judgment drifts toward it, even when the anchor is arbitrary and everyone in the room knows it.

At work: A vendor opens at 50,000 euros. You negotiate hard, feel triumphant at 38,000, and never notice that the fair price was 25,000. The opening number set the gravitational center of the entire negotiation. The same happens with project timelines, salary bands, and the first estimate anyone blurts in a planning meeting.

The correction: Generate your own number before you are exposed to theirs. Decide what the work is worth, or how long it will take, from the fundamentals — then let their figure inform you rather than anchor you. When you cannot avoid hearing the anchor first, name it out loud: “That is their starting position; what does an independent estimate say?” Making the anchor visible is most of the defense.

EVIDENCE GRADE: STRONG

Anchoring is among the most robustly replicated effects in decision research — it survives warnings, expertise, and incentives to resist it. You do not get to be immune. You only get to build a procedure that runs before the anchor lands.

Availability: whatever comes to mind feels true

The availability heuristic is your brain estimating how likely or important something is by how easily examples come to mind. Vivid, recent, and emotional events are easy to recall, so they feel common and significant — regardless of the actual base rate.

At work: One dramatic client complaint reaches leadership, and suddenly the whole team is redirected to a problem that affects two percent of users, while a quiet issue degrading everyone’s experience goes unaddressed because no single story about it was memorable. The loud data point crowds out the frequent one. Recency does the same damage: last week’s incident reshapes a strategy that a year of steady evidence should have set.

The correction: Replace what is *available* with what is *counted*. Before you act on a striking example, ask for the denominator. How often does this actually happen, out of how many cases? In research this reflex is survival — I have watched a single vivid scan tempt a confident story that thirty-five patients did not support. One number, deliberately retrieved, quietly defeats a dozen memorable anecdotes.

Sunk-cost: paying more because you already paid

The sunk-cost fallacy is letting past investment — money, time, effort, reputation — drive a decision that should depend only on the future. The costs are already gone; a rational choice weighs what remains to be gained against what remains to be spent. But abandoning an investment feels like a loss, and the brain works hard to avoid the feeling.

At work: A project is eighteen months in, clearly underperforming, and the honest move is to stop. Instead the argument becomes “we have put too much into this to quit now” — which is precisely backwards. The eighteen months are spent either way. The only live question is whether the *next* six months are worth more here than anywhere else.

The correction: Reframe the decision as if you were arriving today. Ask: “Knowing what I now know, and ignoring what I have already spent, would I start this?” If the answer is no, the past investment is not a reason to continue — it is the emotion you have to overrule. It is the same clean-slate reasoning behind several of the models in the five mental models every young professional needs first: decide from where you stand, not from what you have paid.

The four at a glance

Bias	The filter error	The correction
Confirmation	Attention passes agreement, drops dissent	Name in advance what would change your mind, then seek it
Anchoring	First number sets the center of gravity	Form your own estimate before you hear theirs
Availability	Vivid and recent feel frequent	Ask for the denominator; count instead of recall
Sunk-cost	Past spending distorts future choice	Decide as if arriving today; ignore what is gone

The pattern across all four is one idea: your first, effortless read of a situation is a filtered draft, not the truth. The correction is never “try to be more objective.” It is a specific procedure that feeds your judgment the data the filter would have withheld.

Match your confidence to your filter

I cannot make you unbiased. Nobody is, and the researchers who study these effects are not exceptions — I am not one, on any given afternoon. The instrument is what it is. What you can do is stop trusting the effortless read, and build small procedures that run before the filter finishes its work: the falsifying question before the review, your own number before the negotiation, the denominator before the reaction, the clean-slate frame before the sunk cost pulls you under.

That is the entire discipline. Treat your first impression as a hypothesis, and then do the unglamorous work of testing it. The colleagues who seem to have unusually good judgment are rarely reasoning more brilliantly than everyone else. They have simply built the habit of not believing their own filter — and it is a habit, which means it is available to you too.

If you want the compact toolkit for exactly this, the models I would build first are laid out in the five mental models every young professional needs first — several of them are corrections for these biases, dressed as everyday reasoning tools you can reach for before the filter decides for you.

If this way of thinking is useful to you, the natural next step is the free guide this site is built around: [5 Mental Models to Future-Proof Your Career](/newsletter/) — five models chosen and stress-tested by a brain researcher, each with an honest grade of the evidence behind it. You’ll also get *Signal*, my monthly email: one idea from neuroscience you can use at work. No productivity spam, no AI panic.

*Mageshwar Selvakumar is a doctoral researcher in neuroscience in Erlangen, Germany, studying how chronic pain reshapes the brain using multi-parametric MRI.*