Decision Making Under Uncertainty

Founders make hundreds of decisions per week with incomplete information. Most frameworks in this wiki tell you what to decide in specific domains (pivoting, hiring, pricing). This article is about the meta-skill: how to decide well when you don’t have enough information, can’t wait for more, and have to live with the consequences. This is arguably the single most compounding skill a founder can develop — every other skill is amplified or neutralized by decision quality.

Most founder decision-mistakes are not from missing information or bad judgment. They’re from applying the wrong decision mode: deliberating slowly on decisions that should be fast, rushing on decisions that deserved thought, seeking consensus on decisions that required conviction, or agonizing over decisions that were reversible. Getting the mode right is more than half the battle.

The Fundamental Insight: Type 1 vs Type 2 Decisions

Jeff Bezos’s framing from his 1997 Amazon shareholder letter (reinforced in every subsequent letter) is the most important decision-making distinction in startup management:

Type 1 decisions are irreversible — “one-way doors.” Once made, you cannot easily walk back. Selling the company. Firing a key executive. Raising a priced round at a specific valuation. Launching a public product. Signing a multi-year contract. These decisions deserve careful analysis, multiple perspectives, and slower deliberation.

Type 2 decisions are reversible — “two-way doors.” If the decision turns out wrong, you can walk back through and try something else. Trying a new pricing experiment. Testing a new landing page. Writing a memo and sending it to the team. Most product decisions. Most hiring exploration (not final hire itself). These decisions deserve speed, not deliberation.

Bezos’s key claim: “As organizations get larger, there seems to be a tendency to use the heavy-weight Type 1 decision-making process on most decisions, including many Type 2 decisions. The end result of this is slowness, unthoughtful risk aversion, failure to experiment sufficiently, and consequently diminished invention.”

For founders, the practical rule:

Decision Type	Mode	Time Budget
Type 1 (irreversible)	Slow, multi-perspective, documented	Days to weeks
Type 2 (reversible)	Fast, delegated when possible, biased toward action	Minutes to hours

The failure mode is applying Type 1 process to Type 2 decisions. It feels responsible but it’s actually a form of procrastination that compounds into organizational sluggishness.

The 70% Rule

Bezos’s other contribution: “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.”

The math: the cost of waiting for the next 20% of information is almost always higher than the cost of being wrong with 70%. A Type 2 decision made quickly at 70% confidence and corrected if wrong beats a Type 2 decision made slowly at 90% confidence every time — because the first path tested reality while the second only reduced anxiety.

The practical version for founders: if you find yourself saying “let me think about it more” on a Type 2 decision, you’re probably at 70% and should just decide. Save your deliberation budget for the Type 1 decisions where it matters.

Expected Value Thinking

Most founder decisions aren’t “right or wrong” — they’re bets with probability distributions and outcome ranges. The discipline of expected value (EV) thinking is:

EV = (Probability of good outcome × Value of good outcome)
   − (Probability of bad outcome × Cost of bad outcome)

Two rules:

1. Good decisions ≠ good outcomes. A good decision is one where the EV was positive given the information available, even if the outcome is bad. A bad decision with a lucky outcome is still a bad decision. Founders who confuse outcome quality with decision quality slowly learn the wrong lessons. (Annie Duke’s Thinking in Bets is the canonical practitioner treatment.)

2. Think in distributions, not point estimates. “This will probably work” is weaker than “this has a 60% chance of returning 3x the investment and a 40% chance of returning zero.” The second framing forces you to face the range of outcomes instead of averaging them into a false certainty.

The founder application: before any significant decision, write down:

• The probability of success (as a number, not a feeling)
• The magnitude of the upside if it works
• The magnitude of the downside if it fails
• What evidence would change your probability estimate

If the EV is clearly positive, decide. If it’s marginal, the decision doesn’t matter much — optimize for speed and learning. If it’s negative, don’t do it regardless of how much you want to.

The Pre-Mortem

Invented by psychologist Gary Klein. Before making a significant decision, imagine it’s 12 months later and the decision has failed catastrophically. Then ask: “Why did it fail?”

The pre-mortem reverses the psychology of planning. Normal planning is optimistic — you imagine the decision working and reason forward. The pre-mortem is pessimistic — you assume it failed and reason backward to the causes. This unlocks specific failure modes that motivated optimism hides.

How to run one:

1. Describe the decision and the expected outcome (“we’re hiring a VP of Sales to double revenue in 12 months”)
2. State the hypothetical failure (“it’s 12 months later and we ended up losing customers”)
3. Each person independently writes 3-5 specific reasons for the failure
4. Compare lists — the overlapping reasons are the highest-probability failure modes
5. Adjust the decision to defend against those specific failure modes

Pre-mortems are cheap (30 minutes), work well in Type 1 decisions, and surface the actually scary concerns that get suppressed in normal planning. Most founders don’t run them. They should.

Reference Class Forecasting

Kahneman and Michael Mauboussin’s core insight: when estimating outcomes, the reference class beats the inside view.

“Inside view” thinking: “Our startup will reach $1M ARR in 12 months because our team is great and our product is unique.”

“Outside view” thinking: “Of companies at our stage with our burn rate and our go-to-market motion, what percentage reached $1M ARR in 12 months?”

The outside view is almost always lower than the inside view. This is not pessimism — it’s the base rate correcting for the optimism bias built into being the person making the decision. The inside view knows your team’s strengths; the outside view knows that every other team also thought they were strong.

Founder practice: for any major projection, find the reference class. What percentage of startups at your stage achieve the milestone you’re projecting? If you can’t find the reference class, you’re guessing with extra confidence.

The uncomfortable corollary: if the reference class base rate is 10% and you’re projecting you’ll hit the milestone, you need specific, evidence-based reasons you’re in the top 10% — not general enthusiasm about your team. “We’re special” without specifics is inside-view bias dressed up as conviction.

Disagree and Commit

Bezos again: when there’s disagreement on a team about a decision, the failure mode is usually not that the wrong decision gets made. It’s that the decision gets made but the team doesn’t commit to it. Half the team executes; half sandbags because they disagreed.

Disagree and commit is the rule: express your disagreement in full, fight for your position, and then — if the decision goes the other way — commit fully to executing the decision as if you’d agreed. Not passive compliance; active commitment.

The CEO’s job in this: make it safe to disagree explicitly, and make it non-negotiable to commit once the decision is made. “I disagreed with this plan but we’re going to execute it at 100%” is a healthy team statement. Silent sandbagging is a toxic one.

This is the antidote to the two most common team decision pathologies:

• False consensus (nobody disagreed out loud, but half the room thought it was wrong)
• Zombie decisions (everyone agreed to move forward, but nobody executes with conviction)

Believability-Weighted Decisions

Ray Dalio’s framework from Bridgewater: not all opinions are equal. Weigh inputs by the believability of the person giving them on this specific topic.

A senior engineer’s opinion on architecture is worth more than a product manager’s. A sales leader’s read on a deal is worth more than the CEO’s. The CEO’s opinion on fundraising strategy may be worth more than anyone’s. The goal is not democratic voting — it’s epistemic hierarchy by domain.

Practical founder application:

• For each type of decision, identify who on the team has the most evidence-grounded track record
• When that person’s view conflicts with yours, treat their disagreement as a high-value signal
• Don’t override believability with authority unless you have a specific reason

The failure mode: “I’m the CEO so I decide.” Sometimes true, but if you override your CTO on architecture or your VP of Sales on a deal, you’re betting your believability is higher than theirs on their domain. It usually isn’t.

Strong Opinions, Loosely Held

Paul Saffo’s phrase. Founders should have strong opinions on the important decisions — because weak opinions produce slow, hedged, cowardly action. But they should hold those opinions loosely — willing to update the moment new evidence contradicts them.

The failure modes on either side:

• Strong opinions, strongly held: cult-like certainty, ignores disconfirming evidence, expensive mistakes (WeWork Adam Neumann, Theranos Elizabeth Holmes)
• Weak opinions, loosely held: paralysis, follows the room, never commits, can’t lead
• Weak opinions, strongly held: dangerous — stubbornness without conviction, the worst quadrant
• Strong opinions, loosely held (the goal): decisive action plus intellectual honesty

This connects to PG’s “right kind of stubborn”: attached to the goal, not the specific path. Founders who confuse the two die on the wrong hill.

When to Be Fast vs Slow

A decision’s importance doesn’t determine its speed. Its reversibility does. A useful 2x2:

	Reversible	Irreversible
Important	FAST (most product decisions)	SLOW (hiring an exec, priced round, acquisition)
Unimportant	FAST (most daily decisions)	SLOW (small but permanent choices — company name, domain name)

The quadrant most founders get wrong: Important + Reversible. These feel heavy because they matter, so founders deliberate. But their reversibility means deliberation is a waste — run the experiment, get real data, iterate. Most product decisions, pricing experiments, marketing copy, feature launches, and early hiring trials live in this quadrant.

Michael Seibel’s version: the fast-moving founder beats the careful founder 9 times out of 10, because the fast-moving founder gets more attempts per month. Speed compounds in startups the way interest compounds in finance.

Pre-Commitment Devices

For decisions you know you’ll face under emotional pressure — firing a low performer, cutting a product line, walking away from a bad deal — decide the criteria in advance, in writing, before the emotional moment arrives.

Example: “If this hire is still at 50% of expected performance after 90 days, we will let them go.” Written before the 90 days. The pre-commitment removes the emotional drift that always happens in the moment (“but they’re trying so hard,” “firing them would be hard,” “maybe one more week”). Decisions made in calm are better than decisions made in crisis. Making them in advance is a form of decision arbitrage.

Common pre-commitment applications:

• Firing criteria: what performance level triggers a decision, by when
• Runway triggers: at what cash balance you cut burn or fundraise
• Pivot criteria: what metrics at what time force a pivot decision
• Walk-away prices: the number below which you do not sell the company
• Customer fire triggers: which bad-fit customers you will fire and when

Write them down. Revisit annually. Trust past-you’s judgment over present-you’s panic.

Cognitive Biases Specific to Founders

Founders are especially vulnerable to:

1. Optimism bias — you wouldn’t be a founder if you weren’t. Partially useful; partially deadly. Correct with outside-view reference class forecasting.
2. Sunk cost fallacy — “we’ve invested two years in this, we can’t kill it.” Irrelevant; the past is gone. Only future expected value matters.
3. Commitment and consistency bias — “I said on stage we’d do X, so we have to do X.” The public commitment should not outweigh new evidence.
4. Availability heuristic — the last investor meeting dominates your emotional state even though it’s a sample of one. Keep a dashboard, not a feeling.
5. Survivorship bias — copying successful companies while ignoring the thousands who did the same thing and failed. Every “founders did X and succeeded” story is incomplete without the denominator.
6. Confirmation bias — seeking evidence that supports the decision you already made. Actively look for disconfirming evidence before deciding.
7. Planning fallacy — everything will take 2-3x longer than you think. Multiply all estimates.

The meta-skill isn’t avoiding biases (impossible) — it’s designing decisions to be robust against them. Pre-mortems defend against optimism. Reference classes defend against inside-view. Pre-commitment defends against availability and consistency. Written memos defend against the availability heuristic by forcing systematic thinking.

The “One-Way Door” Checklist

Before any Type 1 (irreversible) decision, run through:

• Have I written down what success and failure look like specifically?
• Have I estimated probabilities, not just directions?
• Have I run a pre-mortem?
• Have I identified the reference class and its base rate?
• Have I sought out disagreement, not just agreement?
• Have I weighed believability by domain expertise?
• Have I slept on it at least once?
• Have I checked: is this actually Type 1, or am I treating a Type 2 as Type 1?
• If this fails, what specifically will I wish I had done differently?
• Am I making this decision because the data supports it, or because I’m emotionally attached to a path?

The checklist takes 30 minutes. Skipping it on Type 1 decisions is how founders end up in the post-mortem saying “I knew something was off but I couldn’t articulate it.”

The “Two-Way Door” Anti-Checklist

Before any Type 2 (reversible) decision:

• Is this actually reversible? (Confirm in one sentence)
• What’s the smallest experiment that would give me real data?
• Can I decide in the next 10 minutes?
• If no, why not?

The Type 2 failure mode is treating it like Type 1. The anti-checklist defends against that.

Decision-Making Applied to Wiki Frameworks

• Pivoting: Type 1 (irreversible) — deserves pre-mortem, reference class forecasting, written criteria
• Senior hires: Type 1 — deserves deliberate process, pre-commitment on performance bar
• Priced rounds: Type 1 — deserves slow deliberation on terms, valuation, investor fit
• Pricing experiments: Type 2 — decide fast, measure, iterate
• Feature launches: Type 2 — decide fast, test with users
• Marketing channels: Type 2 — test fast, kill fast
• Acquisitions: Type 1 — run pre-mortems; WeWork’s WeLive, The Wing, and other acquisitions all failed for reasons a pre-mortem would have surfaced

The skill is not in applying frameworks perfectly. It’s in recognizing which type of decision you’re facing in the first place and setting the corresponding speed.

Key Takeaways

1. Type 1 vs Type 2 is the most important distinction — most founder decision mistakes are applying the wrong mode
2. The 70% rule — most decisions deserve action at 70% confidence, not 90%
3. Decision quality ≠ outcome quality — judge decisions by the EV at the time, not by what happened
4. Think in probability distributions, not point estimates — force yourself to estimate the range of outcomes
5. Pre-mortems surface concerns that planning hides — 30 minutes, high leverage
6. Reference classes beat inside-view intuition — find the base rate for your projection
7. Disagree and commit is the team decision discipline — full fight, full commitment
8. Believability-weight inputs by domain — not all opinions are equal
9. Strong opinions, loosely held — decisive + updatable
10. Pre-commit on emotional decisions — firing, pivoting, walking away — decide the criteria in calm
11. Speed is a function of reversibility, not importance — important reversible decisions should be fast
12. Design decisions to be robust against your biases — you can’t avoid them, but you can defend against them

See Also

• founders-operating-system
• execution
• focus
• founder-psychology
• wartime-peacetime-ceo
• pg-right-kind-of-stubborn
• pivoting
• hiring
• fundraising

Sources

• Peacetime CEO — Ben Horowitz
• Startup Playbook — Sam Altman
• The Right Kind of Stubborn — Paul Graham