Why Clinicians Are Natural Bayesians
A short BMJ editorial that argues doctors are already doing Bayesian reasoning — just informally, through pattern recognition and base rates — and would benefit from doing it explicitly.
The piece is a clear, short introduction to Bayes for practitioners who need to update beliefs in the face of noisy evidence.
For product direction it transfers almost directly: every meaningful product metric is noisy, every interpretation is an update on prior beliefs, and pretending otherwise is what Gill calls "the pre-Bayesian mode" of reasoning.
A four-page paper; worth reading twice a year.
Central argument
Gill argues that clinicians already reason in a Bayesian fashion — they unconsciously apply prior probabilities (base rates, prevalence, pattern recognition) when diagnosing patients — but because they do so implicitly, they are prone to systematic errors that explicit Bayesian thinking would prevent. The central finding is that the 'pre-Bayesian mode' treats each piece of evidence as if it speaks for itself, ignoring the prior, which leads to overconfidence in test results and poor calibration when evidence is noisy or rare conditions are involved. Formalising the intuition, Gill contends, is not a foreign imposition on clinical practice but a discipline that completes what clinicians are already doing.
Critique
The paper assumes that the primary obstacle to better clinical reasoning is conceptual — that once practitioners understand Bayes, they will apply it — but the cognitive science literature on dual-process thinking suggests the real obstacle is that explicit probabilistic calculation competes poorly with fast, heuristic System 1 reasoning under time pressure and cognitive load. Gill offers almost no account of how Bayesian updating would be embedded in clinical workflows, which makes the argument feel aspirational rather than actionable. A short editorial cannot do everything, but the gap between 'you should think this way' and 'here is how to actually do it in a busy emergency department' is large enough to matter.
Why it matters for product
Product leaders face the same structural problem Gill diagnoses: they treat metrics as if the data speaks for itself, ignoring the prior — the conversion rate dropped, so the feature is bad, without accounting for seasonality, sample size, or what was already believed before the experiment ran. Gill's framing gives CPOs a precise label for a chronic team failure: running A/B tests or reading dashboards in 'pre-Bayesian mode', where each data point resets the conversation rather than updating a calibrated prior. More concretely, it argues for building explicit prior-setting rituals into discovery and experimentation processes — stating what you believe and why before results come in, so interpretation is honest rather than post-hoc.