Three gates, all independent
The targeted audience must clear a materiality floor, set per vertical and store size.
The effect must hold in your data: p < 0.05, with consistent direction across 28-, 56- and 90-day windows.
Revenue ranges blend your observed data with validated industry priors. Unvalidated benchmarks are refused.
A play surfaces only when all three clear. Failing any one produces a typed reason — never silence.
What you receive each month
Evidence observed on your store. Send it.
Validated industry pattern fits your data. Try it, measure it.
We looked. Here's exactly why we didn't recommend it.
Signal forming, below threshold. Tracked for next month.
Sometimes the answer is: nothing this month.
When no play clears the gates, beacon says so — with the reason for every play it held back. A system that can't say "no recommendation" can't be trusted when it says "yes."
The math under the hood
Small cohorts can't fake statistical significance.
One lucky month never becomes a strategy.
AOV changes tested without assuming clean data.
Your data + validated priors, weighted by how much evidence exists.
Ranks which customers are actually worth contacting.
Predicts when a customer is going quiet — before they're gone.
Every number on a recommendation traces back to your data, a named method, and a validated prior — auditable by your most skeptical analyst.