Decision-making

How to make better decisions faster: classify the decision first, match the process to the stakes, put numbers on the options, stress-test the favorite, decide as a group without the classic failure modes, and judge decisions by process rather than outcome. Uncertainty, base rates and bet sizing are in probability and risk; the biases that distort choices are in cognitive biases; general tools such as inversion and second-order thinking are in thinking tools. The fast observe–orient–decide–act cycle is on its own sheet, the OODA loop.

Classify the decision first

The most common decision mistake is using the wrong process: agonising over reversible choices, or rushing irreversible ones. Before analyzing anything, sort the decision.

One-way and two-way doors

Jeff Bezos, 2015 Amazon letter to shareholders:

• Type 1: "Some decisions are consequential and irreversible or nearly irreversible – one-way doors – and these decisions must be made methodically, carefully, slowly, with great deliberation and consultation."
• Type 2: "But most decisions aren't like that – they are changeable, reversible – they're two-way doors." A suboptimal Type 2 decision can be undone: "You can reopen the door and go back through."

His warning about scale: as organizations grow they tend to apply the heavyweight Type 1 process to Type 2 decisions, and "the end result of this is slowness, unthoughtful risk aversion, failure to experiment sufficiently, and consequently diminished invention." The opposite failure (Type 2 speed on Type 1 decisions) is rarer but more expensive.

Question to ask. "If this turns out wrong, what does it cost to undo, and how long until we'd know?"

Other ways to sort

Reversibility-based delegation

Use the door test to push decisions down. The more reversible and lower-stakes the decision, the closer to the work it should be made.

Tell people which bucket a decision is in, explicitly. Uncertainty about ownership is the main reason reversible decisions get escalated.

Speed vs accuracy

Bezos, 2016 letter to shareholders: "most decisions should probably be made with somewhere around 70% of the information you wish you had. If you wait for 90%, in most cases, you're probably being slow." And: "If you're good at course correcting, being wrong may be less costly than you think, whereas being slow is going to be expensive for sure."

The 70% is a rule of thumb for Type 2 decisions, not a measured optimum. The real argument is about the value of information against the cost of delay:

Worked test. Before commissioning more research, write down what result would change your decision. If you can't name one, the research is procrastination.

Expected value and decision trees

Expected value (EV) is the probability-weighted average of the outcomes:

A decision tree lays out choices (squares), chance events (circles) and outcomes, then works backwards: at each chance node take the EV; at each choice node take the best branch. Probability foundations are on probability and risk.

Worked example: build for one big customer?

A startup has $80k of engineering time for the next quarter. Values below are in $k of gross profit over two years, net of the $80k spent.

• A: build the integration a large prospect asked for. 50% they sign: $300k. 50% they don't: the integration still has some reuse value, $40k.
• B: spend the quarter on self-serve onboarding. 30% chance of a big lift: $400k. 70% small lift: $60k.
• C: run a two-week paid pilot first (cost $10k), which reveals whether the prospect will sign. Then build A if yes, B if no.

A  build integration   -80
   |- 0.5 signs         +300   -> 220
   '- 0.5 no deal        +40   -> -40
   EV(A) = 0.5(220) + 0.5(-40)            =  90
 
B  self-serve          -80
   |- 0.3 big lift      +400   -> 320
   '- 0.7 small lift     +60   -> -20
   EV(B) = 0.3(320) + 0.7(-20)            =  82
 
C  pilot first         -10
   |- 0.5 "will sign" -> build A: 300-80  = 220
   '- 0.5 "won't"     -> do B:    EV      =  82
   EV(C) = 0.5(220) + 0.5(82) - 10        = 141

Reading the tree:

• A beats B narrowly (90 vs 82), well within the error of the estimates. Don't read much into it.
• C wins clearly (141). The expected value of perfect information is $0.5(220) + 0.5(82) - 90 = 61$: the most you should pay to know the answer first. The pilot costs 10, so it's worth it.
• Variance matters. A ranges from −40 to 220; if the company dies on a −40 outcome, EV is the wrong criterion. Survival first, then EV (see probability and risk).

Costs that matter, costs that don't

Worked example: the nine-month project. A team has spent nine months on a new data pipeline. Three months of work remain; a managed service now does 90% of what's needed for $2k/month. Sunk cost says finish. The right comparison is only forward-looking: three more months of four engineers plus ongoing maintenance, versus migration effort plus $24k a year. The nine months are gone either way. The emotional pull to justify them is the sunk cost fallacy.

The economics behind opportunity cost and marginal analysis is on microeconomics.

Time-horizon tools

Regret minimization

Bezos's account of leaving his job at D. E. Shaw in 1994 to start Amazon (Academy of Achievement interview, 2001, paraphrased): he imagined himself at 80 looking back and asked which choice would leave him with fewer regrets. He concluded he wouldn't regret trying and failing at the internet, but might regret never having tried.

Question to ask. "At 80, which choice will I regret not having made?"

Where it misleads. It is built for big, personal, identity-level choices, and it favors action: people tend to regret inactions more over the long run. It says nothing about whether you can afford the downside. Bezos had savings and a skill set that made failure survivable; check your margin of safety before applying it.

10/10/10

Suzy Welch, 10-10-10: A Life-Transforming Idea (Scribner, 2009). For a decision, ask how you'll feel about each option 10 minutes, 10 months and 10 years from now.

Worked example. Telling a well-liked but underperforming early employee that their role is changing. 10 minutes: awful, awkward conversation. 10 months: the team is shipping and the employee is either in a role that fits or has moved on with a good reference. 10 years: nobody remembers the conversation; everybody remembers whether the company survived. The short-term discomfort is the only argument for waiting.

Where it misleads. Long-horizon predictions of your own feelings are unreliable, and "it'll all look fine in 10 years" can rationalise anything. Use it to expose short-term bias, not to decide by itself.

Prioritization: the Eisenhower matrix

Sort tasks by urgency (time-sensitive) and importance (contributes to long-term goals).

The quote. Dwight D. Eisenhower, address to the Second Assembly of the World Council of Churches, Evanston, Illinois, 19 August 1954, quoting "a former college president": "I have two kinds of problems, the urgent and the important. The urgent are not important, and the important are never urgent." The line was not Eisenhower's own and he said so; accounts link it to J. Roscoe Miller, president of Northwestern University, who was present. The four-quadrant grid is a later construction, popularised by Stephen Covey's The 7 Habits of Highly Effective People (1989), whose "Quadrant II" is the important-but-not-urgent box.

Question to ask. "What's in the schedule box, and is any of it actually on my calendar?"

Where it misleads. "Important" is doing all the work and is rarely defined; without explicit goals the matrix just relabels your gut. The quote's absolutism is also wrong: plenty of things are both urgent and important (a security breach). Its value is in the bottom-right and top-right boxes: dropping and scheduling.

Weighted decision matrix

For choices with several criteria and a handful of options: list criteria, weight them (weights sum to 1), score each option 1–5 on each criterion, multiply and sum.

Worked example: which first customer segment?

Dental: $0.3(4) + 0.25(3) + 0.25(4) + 0.2(3) = 3.55$. Gyms: $0.3(3) + 0.25(5) + 0.25(2) + 0.2(4) = 3.45$.

Sensitivity check. Move 0.10 of weight from pain severity to reachability (0.20 and 0.35): dental drops to 3.45, gyms rise to 3.65. The winner flips on a small change of weights, so the matrix is really saying "dental and gyms are a toss-up; law firms are out." The right next step is a cheap test (ten customer calls in each segment), not more scoring.

Where it misleads. It gives false precision to subjective scores, and criteria are rarely independent (reachability and willingness to pay often correlate). If the result surprises you, treat that as data: either the weights are wrong or your gut is. Find out which.

Satisficing vs maximizing

Satisficing (satisfy + suffice): set an aspiration level and take the first option that meets it. Herbert Simon introduced the term in "Rational Choice and the Structure of the Environment" (Psychological Review, 1956) as part of bounded rationality: real agents can't evaluate every option, so optimizing over everything is not actually rational.

Maximizing: search for the best possible option. Barry Schwartz and colleagues (Journal of Personality and Social Psychology, 2002) found that people who habitually maximize report lower happiness and life satisfaction and more regret than satisficers. Iyengar, Wells and Schwartz (2006) followed graduating job seekers: maximizers landed starting salaries about 20% higher but felt worse about the search and the job. Schwartz's popular account is The Paradox of Choice (2004).

Question to ask. "What is 'good enough' here? Write the bar down before looking at options."

Sequential search with no going back (hiring, flats) has a known mathematical answer, the optimal stopping or "37% rule", covered in models from other fields.

Stress-testing a decision

Pre-mortem

Gary Klein, "Performing a Project Premortem", Harvard Business Review, September 2007. Before committing, the team imagines the project has already failed and each person writes down, independently, why. Klein cites research (Mitchell, Russo and Pennington, 1989) finding that this prospective hindsight, imagining an event has already happened, increased the ability to correctly identify reasons for future outcomes by 30%.

PRE-MORTEM (45 minutes)
1. Brief the plan (5 min).
2. "It's 12 months from now. This failed badly.
   Write down every reason why." Silent, individual (5 min).
3. Round-robin: one reason per person per turn, until
   exhausted. The leader goes last. No debate yet (15 min).
4. Cluster the reasons; vote on the 3-5 most likely or
   most damaging (10 min).
5. For each: mitigation, early-warning signal, owner
   (10 min). Revisit the list at each milestone.

It works because it legitimises dissent: finding a failure mode becomes the task, not disloyalty.

Red teams and devil's advocates

Rotate the devil's advocate role, or better, find someone who genuinely disagrees. Steelmanning is covered in thinking tools.

Group decisions

Failure modes

Janis's evidence came mainly from historical case studies (the Bay of Pigs, Pearl Harbor), and later research has had mixed success confirming the model; treat the symptoms list as a checklist, not a law.

Decision rights

Most slow decisions are unclear-ownership problems. Two common frameworks:

Rules that matter more than the acronym: exactly one decider per decision; input is not a vote; write the decision and the decider down; the decider is expected to decide by a date.

Disagree and commit

Bezos, 2016 letter: "use the phrase 'disagree and commit.' This phrase will save a lot of time. If you have conviction on a particular direction even though there's no consensus, it's helpful to say, 'Look, I know we disagree on this but will you gamble with me on it? Disagree and commit?'" He stresses that it applies to the boss too: his example is green-lighting an Amazon Studios show he doubted.

The phrase predates Amazon. It is associated with Scott McNealy at Sun Microsystems in the 1980s ("agree and commit, disagree and commit, or get out of the way") and is also widely attributed to Andy Grove at Intel.

Decision quality vs outcome quality

A good decision can have a bad outcome and a bad decision a good one. Annie Duke, Thinking in Bets (2018), borrows the poker term resulting for judging a decision by its result alone.

Question to ask. "Given what we knew and could reasonably have found out at the time, was this the right call?"

The trap is hindsight bias: after the outcome, people believe it was more predictable than it was (Baruch Fischhoff's experiments from 1975 onwards). The defense is a record made before the outcome.

Decision journals

Write down the reasoning at the time of the decision, then review it later against what happened. The practice is widely recommended in investing and was popularised for general readers by Shane Parrish's Farnam Street, which publishes a template. What to record:

• the decision and the date; the alternatives considered and why they were rejected
• what you expect to happen, with probabilities and a time frame
• the key assumptions and what would show each is wrong
• how you feel (tired, rushed, excited) and any time pressure
• a review date

Review in batches (quarterly), and score calibration and reasoning, not just hits. Calibration scoring is covered in probability and risk.

How to decide fast

HOW TO DECIDE FAST
[ ] Name the decider (one person) and the deadline.
[ ] Door test: reversible? If yes, cap discussion at
    one meeting or 30 minutes of async debate.
[ ] Can a one-way door become a two-way door?
    (pilot, flag, contractor, trial, staged rollout)
[ ] Write "good enough" criteria before looking at options.
[ ] Generate at least 3 options, including "do nothing"
    and "do it smaller".
[ ] Kill any option that fails a deal-breaker.
[ ] For each remaining option: worst plausible case.
    Survivable? If not, it's out.
[ ] Ask: what information would change the answer?
    Can we get it in a day? If not, decide without it.
[ ] ~70% confident on a two-way door? Decide.
[ ] Disagreement? Hear it once, record it, disagree and
    commit.
[ ] Set the review date and the signal that would make
    you reverse.
[ ] Announce: decision, decider, reasoning, review date.

One-page decision template

DECISION: _______________________  DATE: __________
DECIDER: ________  INPUT FROM: ____________________
DEADLINE: ________  REVIEW ON: ____________________
 
1. TYPE
   Reversible? (Y / N / partly)   Cost to undo: _______
   Stakes: low / medium / high    Repeated? Y / N
 
2. CONTEXT
   Problem in one sentence:
   Why decide now (cost of delay):
 
3. GOOD ENOUGH
   Must-haves (deal-breakers):
   Nice-to-haves (weighted):
 
4. OPTIONS (at least 3, incl. "do nothing")
   | Option | Upside | Downside | Worst case | Reversible |
   | A      |        |          |            |            |
   | B      |        |          |            |            |
   | C      |        |          |            |            |
 
5. KEY UNCERTAINTIES
   Assumption | Confidence % | How we'd know it's wrong
 
6. EXPECTED VALUE / SCORING (if useful)
   Estimates as ranges, not points.
 
7. PRE-MORTEM
   "It failed. Why?" Top 3 reasons + mitigations.
 
8. SECOND-ORDER EFFECTS
   And then what? Who reacts?
 
9. DECISION AND REASONING
   Chosen option:
   Why this over the next best:
   Dissent recorded:
 
10. PREDICTION
   What we expect by the review date (with %):
   Tripwire that reopens the decision:
 
11. STATE
   Rushed? Tired? Emotional? Anyone with an incentive?

Common mistakes

References

• Jeff Bezos, 2015 Letter to Shareholders (PDF) (opens in a new tab): Type 1 and Type 2 decisions, one-way and two-way doors
• Jeff Bezos, 2016 Letter to Shareholders (opens in a new tab): high-velocity decisions, the 70% rule, disagree and commit
• Jeffrey P. Bezos, Academy of Achievement (opens in a new tab): 2001 interview including the regret minimization framework
• Dwight D. Eisenhower, Address at the Second Assembly of the World Council of Churches (1954), American Presidency Project (opens in a new tab): the urgent-vs-important quote in context
• Quote Investigator: "What Is Important Is Seldom Urgent…" (opens in a new tab): tracing the quote's origin
• Stephen R. Covey, The 7 Habits of Highly Effective People (Free Press, 1989): the four-quadrant time-management grid
• Suzy Welch, 10-10-10: A Life-Transforming Idea (Scribner, 2009)
• Herbert A. Simon, "Rational Choice and the Structure of the Environment", Psychological Review 63(2), 1956 (PDF) (opens in a new tab): satisficing
• Barry Schwartz et al., "Maximizing versus Satisficing: Happiness Is a Matter of Choice", Journal of Personality and Social Psychology 83(5), 2002; Barry Schwartz, The Paradox of Choice (Ecco, 2004)
• Sheena Iyengar, Rachael Wells and Barry Schwartz, "Doing Better but Feeling Worse", Psychological Science 17(2), 2006: the job-seeker study
• Gary Klein, "Performing a Project Premortem", Harvard Business Review, September 2007 (opens in a new tab)
• Irving L. Janis, Victims of Groupthink (Houghton Mifflin, 1972); 2nd ed. Groupthink (1982)
• Jerry B. Harvey, "The Abilene Paradox: The Management of Agreement", Organizational Dynamics, 1974 (opens in a new tab)
• Paul Rogers and Marcia Blenko, "Who Has the D? How Clear Decision Roles Enhance Organizational Performance", HBR, January 2006 (opens in a new tab): RAPID
• Atlassian Team Playbook: DACI (opens in a new tab): DACI roles and template
• Ronny Kohavi, "The Origin of HiPPO" (opens in a new tab): history of the term
• Disagree and commit (Wikipedia) (opens in a new tab): Sun Microsystems and Intel origins
• Annie Duke, Thinking in Bets: Making Smarter Decisions When You Don't Have All the Facts (Portfolio, 2018): resulting, decision quality vs outcome quality
• Baruch Fischhoff, "Hindsight ≠ Foresight: The Effect of Outcome Knowledge on Judgment under Uncertainty", Journal of Experimental Psychology: Human Perception and Performance 1(3), 1975
• Farnam Street: Decision Journal (opens in a new tab): a widely used template