A decision framework will not pick your answer. It forces you to classify the decision before you analyze it, and the classification is where almost all the value sits. The research splits into three questions: how to sort a decision into a type, when to trust a fast gut answer over slow analysis, and whether the techniques sold as debiasing tools actually work. My Decisions note is the personal toolkit. This one traces where the tools come from and how well they hold up.
Reversibility is the cheapest first cut
Jeff Bezos split decisions into two kinds in his 2015 shareholder letter. Type 1 is consequential and nearly irreversible, a one-way door. Type 2 reverses cheaply, a two-way door. His worry was not that Type 2 calls get neglected. It was that big companies run every decision through the slow Type 1 process, and pay for it in “slowness, unthoughtful risk aversion, failure to experiment sufficiently and diminished invention.”
The mechanism is asymmetric cost. On a two-way door, a wrong call costs about what it takes to walk back through it, so deliberating past a low bar is waste. On a one-way door, you pay for the mistake forever, and deliberation is cheap insurance. Once you know which door you face, the frame has done its job. Its only job was to make you check before you analyzed, because the default move is to skip the check and spend the same effort on everything.
That sort is a leverage-point judgment: tuning a reversible parameter versus changing something you cannot easily put back.
Cynefin: the problem decides the process
Reversibility sorts by the cost of being wrong. Dave Snowden’s Cynefin framework, built at IBM in 1999, sorts by the kind of cause and effect you face, and each kind demands a different process. The five domains:
- Clear: cause and effect are known and stable. Sense, categorize, respond. Apply best practice.
- Complicated: cause and effect exist but need an expert to see. Sense, analyze, respond. There is a right answer; you pay to find it.
- Complex: cause and effect connect only in hindsight. Probe, sense, respond. Run small safe-to-fail experiments and amplify what works.
- Chaotic: no usable link between cause and effect. Act, sense, respond. Stabilize first.
- Disorder: you don’t yet know which domain you’re in. The most dangerous place to stand.
Picking the wrong domain is the root failure. Treat a complex problem like a merely complicated one, solvable by enough analysis up front, and you get the signature error of consultants and central planners. The sharpest edge runs between Clear and Chaotic. Grow complacent in the “obvious” domain and you can fall off a cliff into chaos with no warning, because you stopped watching a system you assumed was safe. Cynefin makes sense of a situation. It does not run one. Critics note it resists the kind of test that would show whether two people sort the same case the same way.
When to trust the fast answer
Classical decision theory says list the options, score each by odds and payoff, take the maximum. Almost nobody under time pressure does this. Gary Klein studied fireground commanders and found that in about 80% of cases they decided in under a minute, and denied “deciding” at all. His Recognition-Primed Decision model describes what they do instead. They match the situation to past cases, generate one workable action, run it through a quick mental simulation, and act if it holds. No list. No comparison. The first option that survives the simulation is the one taken.
Daniel Kahneman started from the opposite belief, that intuition is riddled with systematic error. The two views look irreconcilable until you read the paper Klein and Kahneman wrote together, “Conditions for Intuitive Expertise: A Failure to Disagree” (2009). They agreed on a boundary. Skilled intuition is real, but only where the environment is regular enough to hold valid cues and gives fast, clear feedback so a person can learn them. Firefighters, anesthesiologists, and chess players get both. Stock pickers, clinicians predicting violence, and long-range political forecasters get neither, and there the confident hunch is an illusion the expert cannot tell from the real thing. So the frame reduces to one question: is this a regular environment with fast feedback? If not, distrust the gut and force the analysis.
Debiasing the forecast, and what the evidence shows
The best-evidenced part of the field targets the planning fallacy, which Kahneman and Tversky named in 1979: we underestimate time, cost, and risk even after living through similar projects that ran long. Bent Flyvbjerg put numbers to it. Across hundreds of transport projects, nine in ten run over budget, and rail overshoots by about 45% on average. The errors lean one way. That skew is the fingerprint of bias, plus optimism baked into the bid.
Two tools have a real track record.
Reference class forecasting. Stop estimating from inside the project. Find a set of similar finished projects and anchor on how they actually turned out. Planners first used it in 2004 on a line of the Edinburgh tram. The punchline writes itself: the tram that got built came in £300m over budget and three years late. A forecasting method cannot save a process that ignores the forecast.
The premortem. Before you commit, assume the project has already failed and have everyone write the story of why. It runs on a 1989 Wharton and Cornell finding by Mitchell, Russo, and Pennington: imagining an outcome as already settled, “prospective hindsight,” lifts the number of correctly named reasons by about 30%. Calling failure certain rather than possible gives people permission to voice the doubt that group optimism would otherwise bury.
One caveat, stated plainly. The underlying optimism bias is not as universal as the popular story claims. Reviews find mixed evidence, and some projects come in under their estimates. Flyvbjerg argues that much overrun is not honest bias but strategic misrepresentation: planners lowball on purpose to get a project approved, and no debiasing trick touches that. The distinction is the whole game. The same overrun needs a different cure depending on whether the cause is a fooled brain or a rigged incentive. Here decision frameworks shade into risk and boundary questions, where the issue is no longer “what is the right answer” but “whose interests drew the frame.”
What frameworks do not do
None of these computes the decision. They redirect attention. Reversibility forces a sort you would skip. Cynefin forces a read of the causal structure. The Klein and Kahneman boundary forces honesty about whether your gut has earned trust here. Reference classes and premortems drag the outside view and the buried doubt into the open. The standard failure is using a framework as theater: running the premortem, then ignoring it; naming the door, then deliberating anyway. A framework changes nothing when the organization rewards the behavior it was built to interrupt.
Try it
Reference-class your own estimates (1-2 hours, any spreadsheet). Pull your last 8 to 10 finished tasks where you wrote down a time or cost estimate. Put the estimate next to what it actually took and divide. Take the median ratio. That number is your personal planning-fallacy multiplier, and for most people it sits above 1.3. Apply it to your next estimate before you commit, and notice how much it stings. Watch for the one-way skew. Honest noise would scatter around 1.0; your ratios will cluster above it. That asymmetry, in your own data, is what Flyvbjerg measured across a nation.
Run a two-person premortem (30 minutes). Take a decision you are about to make. Each person writes, in past tense, the story of how it failed a year out. Compare the lists. Count the failure modes that neither of you raised in normal discussion. The past-tense framing does the work, and the gap between the two conversations is the 30% effect made visible.
See also
- Decisions — the personal, prescriptive toolkit; this note is its research backing
- Systems Thinking — why intuitive interventions backfire, and where structural leverage actually lives
- Marginal Risk Framework — a decision frame for a narrow case: how much danger a thing adds over the baseline, rather than whether it is dangerous
- Critical Systems Heuristics — the boundary-judgment questions that decide what a decision frame leaves out
Sources
- Jeff Bezos, 2015 Letter to Shareholders — the Type 1 / Type 2 distinction in the original.
- Cynefin framework (Wikipedia) — Snowden’s five domains and the prescribed action for each.
- Klein & Kahneman, “Conditions for Intuitive Expertise: A Failure to Disagree,” American Psychologist (2009) — the negotiated boundary on when intuition is trustworthy.
- Flyvbjerg, “From Nobel Prize to Project Management: Getting Risks Right” — cost-overrun data and reference class forecasting.
- Reference class forecasting: promises, problems, and a research agenda — the mixed evidence on optimism bias.
- Pre-mortem (Wikipedia) — Klein’s technique and the 30% prospective-hindsight finding.