Norman Cross studio · scoring lab
Give each tap a weight before you call it engagement.
Most product scores treat a push open like a completed job. Source Stackhub trains analysts and PMs to build App Engagement Scoring models that punish empty motion and reward work that actually finishes.
Flagship preview
Signal Weighting Lab
Twelve weeks of live critique. You arrive with an event stream; you leave with a score that can be explained to a sceptical head of product. We do not ship a universal formula. We teach you how to defend every weight in writing.
Next UK evening cohort sits Tuesdays. Seats are capped so each scorecard can be read aloud.
What people actually said after the worksheets
All voicesWe had been reporting “engaged users” as anyone who unlocked the app twice in seven days. The Signal Weighting Lab made us split unlocks from completed jobs, then again from jobs that survived a week. The scoring model is fussier than I wanted. Our growth deck got shorter, which the commercial team disliked, but the score finally moved when we shipped a real workflow, not a badge.
★★★★☆
Clear on weighting. Dry on visualisation. Bring your own charting taste; they will not pick a BI tool for you.
The quiet-week module saved our B2B score. Sparse events were being treated as churn. They are not always churn.
What we insist on teaching
Event dignity
An open, a hover, and a paid conversion do not share a moral universe. We show how to write weights that a new hire can reconstruct from the comment column.
Session depth
Duration is a lazy proxy. Depth is a sequence: which jobs were attempted, which were abandoned, which were repeated because the UI lied.
Score humility
A single number will be misused. We teach confidence bands, exclusion lists, and the sentence you put next to the number in the board pack.
A lab week, without the theatre
Monday brief
A short memo lands with one flawed scorecard and the event dictionary it claims to use. You annotate the gaps before class.
Live weighing
Tuesday evening we argue weights in the open. If you cannot say the weight aloud, it does not enter the model.
Rewrite
You resubmit by Friday with a change log. We mark the log, not the polish of the chart.