Measurement · 9 February 2026
Activation curves without self-deception
The most common lie in app analytics is not a fake number. It is an activation definition that moved after the chart looked unkind.
We have watched this sequence more than once. A team picks “completed profile” as activation because it is easy to instrument. The curve is low. Someone suggests “opened the app twice in seven days.” The curve rises. A board deck is happier. The product is unchanged.
Activation is supposed to mark the first moment a user received the value you claim to sell. That moment is usually slightly painful to measure. In ride-hailing it is closer to a completed trip than to an account create. In grocery it is closer to a delivered first order than to an add-to-cart. In audio subscriptions it might be three full tracks in forty-eight hours — not “hit play once.”
Write the definition before you plot
In the studio we freeze the sentence on day one of the retention module: “A user is activated when ___ within ___ of signup.” We date it. Changing it later requires a written reason, not a prettier line.
If you must compare two definitions, plot both, labelled, for the same cohort. Do not overwrite history. Overwriting is how a new PM inherits a miracle.
Windows are not morality
A seven-day window is not more rigorous than a one-day window. It is a claim about how your product delivers value. A slow B2B workflow may need fourteen days; a snackable consumer app that cannot activate in forty-eight hours may not have a product yet. Argue the claim. Do not hide inside a default Looker template.
Self-deception usually arrives as kindness: “Let’s give users more time.” Sometimes that is correct. Sometimes it is a way to avoid admitting the onboarding is theatre. The Telemetry Canvas health strip helps here — if identity merge is broken, your activation curve is a ghost story anyway.
Bring your current definition to Signal Architecture if you want it marked in public. Bring a thick skin.