Skip to main content

Jeff Bezos

Run minimum cost experiments on reversible decisions, and let the data compound.

Experimentation Thesis​

The core discipline: separate reversible from irreversible decisions, move fast on the former, move slow and careful on the latter. Customer obsession — not competitor obsession — is the compass.

Core Principles​

Outcome​

This method produces decisions taken at the right speed: fast on reversible bets, slow on one-way doors. You are done when each open decision is tagged reversible or irreversible and resourced to match.

Method​

  1. Frame the decision and name the customer outcome it serves — not the competitor it answers.
  2. Classify it: reversible (Type 2, a two-way door) or irreversible (Type 1, a one-way door).
  3. For reversible decisions, design the minimum cost experiment that resolves the biggest unknown.
  4. Ship the experiment, capture the data, and let evidence — not opinion — settle the call.
  5. For irreversible decisions, slow down, widen counsel, and disagree-and-commit before proceeding.
  6. Feed the result back so the next cycle starts from a compounding base.

Signals​

  • Measure decision latency: reversible calls that sit for weeks are a stalled flywheel.
  • Verify each experiment has a pre-named metric that would prove it wrong.
  • Observe whether customer signals, not competitor moves, are driving the queue.

Proof Of Done​

Done when every live decision carries its reversibility tag and each experiment has resolved its metric. Write the reasoning down so a later reviewer can check whether the call held.

Failure Modes​

The method breaks in familiar ways:

  • Treating every decision as irreversible — the classic failure mode. Slow one-way-door rigor applied to two-way doors kills speed and compounding.
  • Competitor obsession — optimizing against rivals instead of customers points the flywheel at the wrong axis.
  • Experiments with no falsifier — a test that cannot come back negative proves nothing; it is theater, an anti-pattern.

Changes my mind: If a domain's decisions are overwhelmingly irreversible and cheap experiments cannot de-risk them, the reversible-first speed advantage disappears and slow deliberation wins.

Next question: Which decision on your plate right now is a two-way door you are wrongly treating as a one-way door?

Context​

  • Decisions — Reversible vs irreversible as the core decision filter
  • Data Flow — How data compounds into competitive advantage
  • Platform — Infrastructure as multiplier (AWS as the canonical case)
  • Systems Thinking — Flywheel mechanics and second-order effects

Questions​

What is the minimum cost experiment that tells you whether a decision is worth making at full scale?

  • How do you distinguish a reversible from an irreversible decision when the stakes are unclear?
  • If customer obsession is the compass, what does it look like when short-term customer requests conflict with long-term customer outcomes?
  • Where does long-term thinking break down — when does it become an excuse to delay accountability?