Marketing Data Joined to Clinical Records, Countably
In healthcare the measurement problem is identity, governance, and the fact that most of the interesting data cannot leave the building. Samuel Martin builds systems that are auditable at every join and publishable without exposing anything.
This surface re-sequences the same four case studies and swaps the sentence above it. No proof point here is unique to this page, and none of the four surfaces is the real one.
The closed loop first, then measuring an AI surface honestly.
Nine months later, the real gain was 7.0 points
Measuring an AI chat feature on whether people stayed rather than whether they tried it. Novelty and self-selection inflate every early number on a new AI surface, and the nine-month answer was 7.0 points: real, and much smaller than the launch enthusiasm.
Financial reporting went from weeks to minutes
Marketing API streams fused to CRM leads and Electronic Medical Record encounters in one closed-loop warehouse. Volume was never the constraint; identity was. Each join emits a survival count, so the 38% loss at the CRM join is a fact anyone can check rather than a suspicion.
One average became four peer groups
Clustering a book of 750+ enterprise clients on mixed-type attributes, and giving up presentation precision on purpose. The review line says median of its cluster rather than a percentile, because eight to fifty accounts do not support a percentile.
A $99 experiment that made outbound a revenue product
Randomized measurement inside a healthcare operation, reported at intent-to-treat on 4,812 contacts, with the unit economics attached: $99 cost per acquisition against $549 of first-year value.
What Samuel Martin would own here.
The measurement architecture across protected data, the joins that make it auditable, and the honest evaluation of AI features that arrive with their own novelty premium.
Figures across this site are redrawn on synthetic data, no client is named, and the method and reasoning are exact.