Thinking,
Fast and Slow
How intuition, attention, memory and choice shape the decisions we make, and how to build better decision environments.
Your mind is efficient, not neutral.
Most of the time, we use quick impressions and familiar patterns. That works brilliantly in everyday life, but it can also turn a guess into a belief before we notice.
Fast is the default
Intuition generates the first answer: a feeling, story, estimate or impulse.
Slow is a scarce resource
Careful reasoning can correct the first answer, but it needs attention, time and motivation.
Good systems help
Checklists, base rates, defaults and premortems make better thinking easier at the point of choice.
spotting when a confident story is built from too little evidence, then changing the decision process rather than relying on willpower alone.
Why this matters
The point isn’t to distrust every intuition. It’s to know when intuition deserves a second look.
Better decisions come from better conditions.
Roadmaps and bets
Separate what the team knows from the story it is telling about customers, competitors and timing.
Model and eval choices
Use base rates, held-out tests and explicit uncertainty instead of choosing the most fluent demo.
Evidence and replication
Expect noise in small samples. Pre-specify what would change your mind before seeing the result.
Time and trade-offs
Distinguish how an experience feels while it happens from the story you will later remember about it.
Two characters, one mind.
System 1 · automatic
- Fast, associative and effortless
- Produces impressions, feelings and first answers
- Excellent with familiar patterns
System 2 · effortful
- Attention-demanding and sequential
- Handles computation, rules and self-control
- Can check System 1, but tires easily
These are useful labels for modes of processing, not literal brain modules. The practical question is: what would make a check happen here?
System 1 proposes. System 2 usually approves.
Easy to process can feel like true.
Familiarity, repetition, clear presentation and a good mood can make an idea feel more credible or more obvious, even when they are not evidence for the claim.
Useful side
Clear language and legible design reduce needless friction. People can understand the important thing faster.
Risky side
A repeated claim can borrow the feeling of familiarity. A polished demo can borrow the feeling of competence.
Make the interface easy to use, but make the evidence hard to fake. Keep provenance, uncertainty and counter-evidence visible.
Common shortcuts, translated into plain English.
| Shortcut | What it does | Where it bites |
|---|---|---|
| Availability | Judges likelihood by how easily examples come to mind. | Recent incidents, vivid demos, news cycles. |
| Representativeness | Matches a case to a prototype and neglects base rates. | Hiring, diagnosis, user research, “this feels like a winner”. |
| Substitution | Answers an easier question than the one asked. | “How good is this strategy?” becomes “Do I like the story?” |
| Anchoring | Starts from a number and adjusts too little. | Pricing, estimates, targets and negotiations. |
A heuristic is not automatically a mistake. The risk is using a shortcut in a setting where the cues are noisy, the stakes are high or the base rate is ignored.
The first number can bend the next number.
Kahneman and Tversky found that even arbitrary numbers can pull estimates towards them. People may adjust deliberately, but they often stop too soon.
Before estimating
Write your independent range first. Separate assumptions from the number already on the table.
When negotiating
Don’t merely counter an extreme anchor. Explain why it is irrelevant and bring evidence from the outside view.
Small samples produce extreme results more often.
The smaller the sample, the more noise can masquerade as a pattern.
Kidney-cancer counties
In the book’s example, both the highest and lowest rates appeared disproportionately in small, rural counties. The shared feature was small sample size, not a rural causal explanation.
Practical implication
Treat a striking result as a prompt to ask: how much data produced this? What happens on replication? What is the comparison group?
Confidence follows the story, not necessarily the evidence.
“Our plan makes sense.”
Focuses on the particular case: our team, our product, our advantages and our intended path.
“What happened to similar cases?”
Starts with a reference class, distribution of outcomes and the obstacles that affected comparable efforts.
Make a forecast
Give a range, not just a point estimate.
Find the reference class
Use comparable projects, not only your own narrative.
Run a premortem
Assume the plan failed and write the history of how.
Record the prediction
Review calibration later. Memory edits the past.
Choices under uncertainty
We don’t evaluate gains and losses symmetrically. The frame and the reference point matter.
Losses usually feel larger than equivalent gains.
The value of an outcome is judged relative to a reference point. As the outcome moves farther away, sensitivity tends to flatten.
“Where am I starting?”
A price, current state, expectation or previous outcome can become the baseline that defines gain and loss.
“How would I feel about losing it?”
The emotional impact of giving something up can outweigh the pleasure of receiving the same thing.
When introducing change, understand the user’s reference point. A “benefit” can still feel like a loss if it removes a familiar capability or control.
Same facts. Different frame. Different choice.
“90% fat-free”
Highlights what remains as a positive attribute.
“10% fat”
Highlights what is given up or at risk.
What choice is actually being made?
Write the same option as gain and loss.
Would a neutral description change your preference?
Explain the trade-off in plain language.
The self that experiences is not the self that remembers.
“How does it feel now?”
Tracks the moment-by-moment experience. Duration matters because moments accumulate.
“How was it overall?”
Builds a compact story, strongly influenced by the peak and the end. Duration can disappear from the summary.
A good ending can improve the memory of a long experience.
What the experience contains
A whole curve: high and low moments, plus the time spent in each state.
What the memory keeps
A compressed judgement that overweights the most intense moment and the ending.
Before improving an experience, decide which outcome matters: lower total burden, a better memory, or both. They can point to different interventions.
Use the model as a prompt, not a verdict.
System 1 and 2 aren’t brain boxes.
They are shorthand for automatic and effortful operations. The book says this explicitly in its conclusion.
Intuition isn’t always a bias.
In stable environments with repeated practice and fast feedback, expert intuition can be valuable.
A checklist can still fail.
Good process reduces predictable error; it doesn’t remove uncertainty, incentives or bad data.
Ask what is directly supported by the source, what is a reasonable interpretation, and what is a recommendation for your context.
A decision hygiene protocol
A lightweight routine for slowing down only where slowing down is worth the cost.
Define the decision before discussing the story.
Write the question
“Which option should we choose?” not “Do we like option A?”
Name the stakes
What is reversible, what is costly and what has a deadline?
Set the reference point
What is the baseline, current alternative or default?
Record the first estimate
Capture your range before other people’s numbers become anchors.
Make the hidden shortcuts visible.
What happened in comparable cases?
How much data supports this pattern?
What else could explain the same evidence?
Would the decision survive a gain/loss rewrite?
Ask for a “consider-the-opposite” pass before asking for consensus. It is easier to notice a blind spot before the group has publicly committed.
Use a premortem to release dissent.
Imagine the plan has failed one year from now. Each person has five to ten minutes to write the history of the failure.
“It failed because…”
What obstacle, dependency, incentive or assumption did we underestimate?
Risk register, not theatre
Turn each credible failure mode into an owner, an early signal and a mitigation.
Keep a record so confidence can meet reality.
Prediction log
Record the date, forecast, range, assumptions, reference class and confidence.
Outcome: ______ Range: ________ Confidence: ___% Base rate: _____
Review loop
- Compare outcome with forecast.
- Separate luck from process quality.
- Update the reference class.
- Adjust the next estimate.
Keep the original prediction. Retrospective coherence is one of the reasons overconfidence survives.
Turn the book into an AI decision checklist.
| Decision moment | Likely trap | Countermeasure |
|---|---|---|
| Model selection | Fluent demo becomes “best model”. | Blind task set, fixed rubric, cost/latency and failure taxonomy. |
| Launch forecast | Inside view and competitor neglect. | Reference class: comparable launches, adoption funnel and disconfirming signals. |
| Eval result | Small sample or vivid success story. | Power/sample check, held-out cases, confidence range and replication. |
| UX change | Loss aversion and default bias. | Frame flip, user baseline, opt-out test and reversible rollout. |
When the decision matters, ask these seven questions.
- What is the actual question?
- What first impression or number is anchoring me?
- What is the base rate and how large is the sample?
- What evidence would disconfirm my story?
- What does the outside view say?
- How does the frame change the choice?
- Will I later judge the experience by the peak and the end?
Don’t aim to think slowly all the time.
Aim to recognise the moments when a quick answer is expensive, when uncertainty is being hidden, or when the environment is nudging the choice for you.
Slow down selectively
Use effort where the stakes, uncertainty or irreversibility justify it.
Design the process
Base rates, defaults, logs and premortems beat heroic vigilance.
Stay humble
Confidence is a feeling about the story. Calibration is a relationship with outcomes.
Share this deck with anyone who makes decisions under uncertainty. Source: Daniel Kahneman, Thinking, Fast and Slow, 2011. This deck is a concise educational synthesis, not a substitute for reading the book or checking original studies.