3T Recovery Wizard

A recovery estimate on its own is half an answer: a seller sees how large the opportunity could be, but not how likely it is to pay out — or how much to trust the number.

A Bayesian recovery-intelligence app built for the Threecolts team, forecasting conservative recovery value across five channels: Amazon 1P, Amazon 3P, Walmart 1P, carrier invoice audit, and carrier contract optimization. Every parameter in the multiplicative recovery chain — exposure, base found rate, opportunity multiplier, attainment rate, policy multiplier, timing — carries a Gaussian posterior rather than a point value. Recorded outcomes conjugate-update each posterior, first-order error propagation combines them into 50/80/90/95% credible intervals, and a reliability score derived from the coefficient of variation says how much weight the estimate deserves. Outcomes land in a connected Supabase database, and a calibration step blends aggregated results back into the priors — a continuous-improvement loop where every recorded engagement tightens the next estimate. A scenario builder, a multi-year forecast with seasonality, and a channel-level uncertainty decomposition round it out.

Gives sellers and the team three reads instead of one: how much, how likely, and how much to trust it — the reliability score marks when an estimate is defensible and when it is a directional guide pending an audit. Built to sharpen as its recovery history grows.