The retained event without circularity
Stop defining retention as “people who retain.” Pick an action that is rare enough to matter and common enough to measure weekly.
Flagship programme
Retention Cohorts for Product Teams
An 11-week studio for people who already ship dashboards and still cannot agree what “came back” means. You will leave with a documented retained event, a weekly grid, and a slide that names its own limits.
Informational fee: £1,240 per seat · next studio opens in October · no payment is taken on this site.
What you should be able to do afterwards
- Write a retained-event definition that another analyst can reproduce without a meeting.
- Choose tenure weeks over marketing weeks and show why the two disagree.
- Quantify new-user mix before celebrating a “lift” in a young cohort.
- Explain a last incomplete period without hiding it in a colour scale.
- Answer a sceptical finance partner without reaching for a second north star.
Modules
Calendar weeks versus tenure weeks
Build both clocks on the same extract. The homework is a one-page note on which clock your product actually runs on.
Composition bias and new-user mix
The week Priya mentioned in the home quote. You will re-weight a fake “win” until it disappears.
Survival curves beside the classic grid
When a heatmap conceals hazard. You will plot both and decide which one a product trio can actually use.
SQL patterns for weekly grids
Denominators, late-arriving events, and the last column that is always too young. BigQuery and Snowflake dialects.
Presenting a cohort to a sceptical board
One slide, one limitation, one ask. Recorded critique from faculty, not a public speaking workshop.
Instructor
Dr Imogen Hale
Former analytics lead at a Bristol marketplace. Imogen designed Desktop Routecore’s cohort method after watching three successive “retention OKRs” collapse under mix. She still writes SQL in class; she does not outsource the messy joins to a teaching assistant.
Fee (informational)
The listed studio fee is £1,240. It covers eleven live sessions, written feedback on two assignments, and six months of template access. Travel to West Stoughton is not included. We do not process cards on this website; write to the desk if you need an invoice path.
Questions people actually ask
Do I need to be a data scientist?
No. You need to be comfortable with a warehouse or a well-structured export. If joins still frighten you, take SQL for Weekly Retention Tables first.
Will this work for a mobile game as well as B2B SaaS?
The clocks differ. The method does not. We use both kinds of example in weeks two and four.
What will this course not do?
It will not give you a causal estimate of a feature launch. Cohort grids describe who returned; they do not isolate why. If you need experiment design, this is the wrong programme, and we will say so in week one rather than stretch the syllabus.
How large is a studio?
Sixteen seats. Critique does not work when thirty people are waiting for a comment.
Notes from recent studios
The tenure-week exercise exposed that our “weekly active” metric was a marketing calendar wearing a product badge. I rebuilt the grid before the next QBR.
Helena Crowe, product manager, Edinburgh
Useful, occasionally dry. Imogen’s board-slide module is the piece I still open. The SQL week assumed more Snowflake fluency than our team had; we paired with the lab course afterwards.
Client in insurance analytics