Bayesian Thinking

Bayesian Thinking is a reasoning framework rooted in Bayes' Theorem — a mathematical formula for updating the probability of a hypothesis as you receive new evidence. Named after Reverend Thomas Bayes, it formalizes what good thinkers do intuitively: start with a belief, encounter new information, and adjust accordingly. The key insight is that your prior beliefs matter, and evidence should shift them proportionally to its strength.

1. Start with a prior belief — What do you currently believe, and how confident are you? Express this as a rough probability (e.g., "I think there's a 60% chance this feature will increase retention").
2. Encounter new evidence — Gather data, run experiments, or observe outcomes. What new information have you received?
3. Assess the evidence's strength — Ask: "How likely would I be to see this evidence if my belief were true? How likely if it were false?" Strong evidence is much more likely under one scenario than the other.
4. Update your belief — Shift your probability based on the evidence. Strong evidence → big shift. Weak or ambiguous evidence → small shift.
5. Repeat — Each piece of evidence refines your belief. Over time, your confidence converges toward reality.

• Don't ignore your prior knowledge — it's valuable context
• Don't cling to priors when strong evidence contradicts them
• Extraordinary claims require extraordinary evidence
• Update incrementally, not in leaps

• Prior: 50% confident users will adopt (new category, uncertain)
• Evidence 1: Competitor launched similar feature, saw 30% adoption → Update to 65%
• Evidence 2: Internal survey shows 70% of users find current search frustrating → Update to 75%
• Evidence 3: Beta test with 100 users shows only 15% switched to AI search → Update down to 45%
• Evidence 4: Post-beta interviews reveal users didn't know the feature existed (discovery problem, not value problem) → Update to 55%

Bayesian Thinking helps you make better decisions under uncertainty by systematically updating beliefs with evidence. It prevents both stubbornly ignoring new data and overreacting to a single data point.