“I arrived convinced our onboarding funnel was a design problem. Funnel Forensics showed the permission prompt was firing twice and the second event was counted as a drop. Embarrassing. Useful.”
Voices
Not a wall of five stars.
These notes mention particular modules. A few of them mention friction. That is how we know they were written by people who actually did the homework.
★★★★☆
Posted under Storefront Experiments. “Good on listing screenshots. Thin on Android feature graphic variants. Still hired the method.”
“The packet diary habit survived the course. I open it before I open Mixpanel, which my old self would have found precious.”
Anonymous client in insurance: “Cohort Craft would not let us treat ‘engaged user’ as a feeling. We now define it as two successful claims views inside fourteen days. Legal liked that more than marketing did.”
“Taxonomy Studio is slow on purpose. If you need a certificate by Friday, look elsewhere. If you need to delete events without a mutiny, this is the room.”
“Mild reservation: the Cashel timezone for live seminars is awkward if you are in Edinburgh and already in stand-up. Recordings land the same afternoon.”
Case study 01
A grocery app that stopped counting ghosts
A mid-size UK grocer sent two analysts to Cohort Craft after a board pack claimed D1 retention of 61%. The number was true only if you accepted “app_open” from a silent refresh as a returning human.
During module three they rebuilt the returning definition around a completed basket view after a cold start. The new D1 sat at 38%. Unpleasant in the meeting. Accurate enough to stop a planned push-notification blitz that would have hammered people who had never left.
They kept the old chart in an appendix labelled “vendor default” so nobody could pretend the previous figure had not existed.
Tomáš H. · analytics manager, grocery
Case study 02
A language app and the store screenshot that lied politely
Storefront Experiments took a language-learning team through a listing test they had already “won.” Conversion to install rose. Trial-to-paid did not. The winning screenshot promised a daily streak aesthetic the first session did not deliver.
The useful output was not a new winner. It was a rule: listing claims must map to an event that fires in session one. They now keep a one-column table — claim, event name, owner — and refuse to ship creative that cannot point at a packet.
That table now lives in the same folder as their event contract from Taxonomy Studio. Two artefacts, one argument.
Aisha M. · product, education