Anti-Fragile take
When you score accounts on rare events (niche signals, new markets), small sample sizes produce wild scores. A company with one matching signal scores 100%; one without scores 0%. Laplace smoothing pulls extreme estimates toward the prior.
We use it in scoring models trained on fewer than 50 closed-won deals so early scores are usable instead of noisy.
Why it matters
Without smoothing, scoring models on small datasets produce false confidence that reps quickly learn to ignore.
How we build it
Smoothed probability estimates in warehouse scoring models, with prior set from overall conversion rate and updated as sample size grows.