Every figure below is one app’s own measurement over a named window, with the basis printed under it. We name the industry and the method, never the company.
A holdout, an attributed share, and a pilot’s reply-to-pay split. Three different strengths of evidence, labelled as such.
the win-back rate of an untouched holdout. 18.6% came back against 6.9%.
Holdout measuredof a month’s revenue came through Hyzl. 255 of 431 paying customers.
Attributedof the people who reply go on to pay. The silent pay at about 10%.
Pilot analyticsThe label under each figure is the method behind it, not a grade we gave ourselves.
A slice of the same cancellations was never contacted, over the same three weeks, and counted the same way. The gap between the two arms is the claim. It is our own measurement inside one app, not a trial anyone else reviewed.
No holdout. The app reports the revenue that came through Hyzl against everything else it took in the same window, inside its own analytics.
A pilot, read from the pilot’s own numbers. Enough to show which opener worked. Not enough to claim a lift against people nobody contacted.