Week four was the first time I had to present a chart I disliked. Harriet would not let me replace it with a prettier window. I still show the ugly one in our Monday pack.
Flagship · 11 weeks
Retention Architecture
For the person who owns the graph the executive team screenshots. You will rebuild how activation is defined, choose cohort windows that match how the product is used, and learn where resurrection campaigns quietly poison the series.
Modules
Naming the leak
Before charts: a written account of who leaves, when, and which event would have to fire for you to notice. We mark the writing, not the screenshot.
Windows that match the product
Seven, fourteen, and thirty are calendar habits. You will defend a window against how your users actually return — including apps with weekly rituals rather than daily ones.
Activation versus habit
A first-value moment is not a habit loop. We separate them on paper so marketing cannot borrow the halo of one for the other.
Resurrection without costume
Win-back campaigns inflate returning-user counts. You will learn to isolate them so the core graph remains readable.
Board-ready without vanity
A pack that survives a sceptical CFO: who is missing, what you tried, what you will not claim.
Instrumentation repair clinic
Live work on the schema you brought. Faculty will not rebuild your pipeline for you; they will refuse to let you hide behind “the tool did that.”
What you should be able to do afterwards
- Write an activation definition a new analyst can implement without a meeting.
- Choose a cohort window and document why a neighbouring window was rejected.
- Spot a resurrection campaign inside a retention heatmap.
- Retire a north star in writing, with a successor and a sunset date.
- Explain an empty cell without widening the window.
- Keep a measurement diary that would embarrass a decorative dashboard.
Instructor
Harriet Voss
Harriet spent fourteen years in product analytics, lastly as measurement lead at a United Kingdom payments firm that had to explain a bundle-ID migration to a listed parent. She designed Retention Architecture after noticing that vendor academies taught click-paths and left the politics of metrics to chance. She still marks diaries by hand.
Informational fee
The published figure is £2,480 for the eleven-week Cohort Atelier seat, excluding VAT where it applies. That is the same number as on the fees page. There is no checkout here. Seats are offered after a written note to the desk. Observatory teams sometimes send a named seat; that is a different contract.
Reviews from this programme
The instrumentation clinic found that our “habit” event was a silent background refresh. Fixing it made retention look worse for a month. The diary is the only reason I still have political cover.
I wanted more Mixpanel recipes. That is not this course. Once I accepted that, the board-pack week was the most useful sitting I have paid for.
Questions we are asked
Do you certify Amplitude or Mixpanel click-paths?
No. Tool interfaces change quarterly, and a certification in last year’s navigation bar is a souvenir. Retention Architecture will use whichever stack you bring. If you need a vendor badge for a job description, this is the wrong room — we will say so in the first reply rather than take the fee.
What is a real limitation of the programme?
We do not implement tracking plans for you, and we do not staff an engineering function. If your events are broken in production, faculty will help you see it; they will not open a pull request. Teams without anyone who can change the schema stall in week six. That stall is on us to name early, and on you to staff.
Can I join from outside the United Kingdom?
Yes. Evening sittings are on UK time. We will not shift the clock for a single seat. Recordings exist for the lecture half only; critique is live.
What if I miss a Thursday?
You may miss two sittings without conversation. A third means we pause your diary marking until the next cohort — fees are not automatically refunded; see refunds.