Every idea has a hidden true probability p that a real customer converts (buys, signs up, keeps using it) — you never see p directly, only noisy evidence about it. Your belief starts at an uninformative prior Beta(2,2), centred on 50%.
Prior: Beta(a0, b0) = Beta(2, 2)
Experiment: observe k successes out of n trials (k ~ Binomial(n, p))
Posterior: Beta(a0 + k, b0 + n - k)
Mean: a / (a + b)
Variance: ab / [(a+b)^2 (a+b+1)]
This is exact Bayesian conjugate updating — each experiment's (n, k) is folded straight into the Beta parameters, so the posterior after several experiments is just the same formula applied with the totals. The 3D bars are the Beta probability-density function evaluated across viability from 0% to 100%; a taller, narrower ridge means you are more certain.
- Interviews are cheap and fast but a small sample (n=6) barely narrows the curve — a classic Lean Startup trap: people are polite in interviews, so treat a good result cautiously.
- Landing-page test measures real intent (email signups / clicks) at moderate cost and sample size.
- MVP pilot is expensive and slow but the largest sample — the strongest evidence per experiment.
- Launch / Kill ends the round: launching is scored correct if the hidden true p ≥ 50%, killing is scored correct if p < 50%. Your threshold slider is a personal risk bar shown on the chart — it does not decide for you.
Real-world relevance: this is the same Beta-Binomial model used in A/B-testing platforms and clinical-trial interim analysis to decide, under a budget and a deadline, how much evidence is "enough" before committing.