These notes are from people who sat the Product Signal Atelier or a clinic. We keep the awkward ones. A page of unmixed praise would be a funnel we would fail in class.
Week three of the Product Signal Atelier hurt. We had been reporting “activation” as any session longer than thirty seconds. Helena asked who had decided thirty. Nobody could say. We rebuilt the event around completing a first saved route. Leadership disliked the dip. The dip was honest. I still use the funnel contract template from module four when a new PM joins.
Amelia Hart · Head of Product, Edinburgh
★★★★
The Event Schema Craft week is the piece I quote internally. We stopped camel-casing one platform and snake-casing another. I wanted a longer lab on server-side forwarding; the notes were thin there. Still the only course that made our data engineer stay for the whole critique.
Marcus · Leeds
I took Retention Modelling Desk after two years of presenting D7 as if it were a personality trait of the app. Learning to split resurrection from genuine new-user retention changed the Q2 board pack. Finance still argues with us. At least we argue about the same curve.
Client in digital banking · London
Signal Desk suited me because I cannot do Thursday evenings. The recordings of the taxonomy critique are slower than a live room, and you miss the heckling. I used both office-hour credits on a single messy identify() call. Would I pay for Instrument Bench next time? Yes, if the diary allows. The desk is not a substitute for being in the room; it is an honest lesser option, which I appreciate.
James Okonkwo · Staff engineer, Cardiff
Anonymous because my legal team asked. The Mobile Funnel Clinic walked our growth lead through ATT and a deferred deep link that had been double-counting. We did not become “good at ads.” We stopped congratulating a campaign that never existed. That was the brief.
Client in consumer fitness · United Kingdom
A two-sided app in the Midlands arrived at the Product Signal Atelier with a north star called “orders.” Seller ops counted paid disbursements. Consumer product counted checkout success. Finance counted captured payments after refunds. All three events were named order_completed in different projects.
The work was not a new dashboard. It was a three-line contract: who fires, when, which properties are required, and which number the board will see. Week six they presented a readout with two charts labelled as such. The CEO still wanted a single line. They kept the contract anyway. Six months later the seller-ops number is the one used for bonuses; consumer product reports checkout as a supporting metric. The atelier did not make that political choice. It made the choice visible.
During Retention Modelling Desk, a media app from Glasgow brought thirteen months of weekly cohorts. D1 looked athletic. D30 was a cliff they blamed on “content gaps.” We asked for the same cohorts excluding users who arrived from a two-week free trial that auto-renewed into a paid month they immediately cancelled.
The cliff gentled. It did not disappear. The useful finding was uglier: trial users who never opened an article in week one were being counted as retained if the billing event succeeded. They rewired retention to require a content event, accepted a worse headline number, and stopped buying the same trial campaign. I would not call it a transformation. I would call it the first week they stopped arguing with a chart that had been constructed to soothe them.
