Field note · March 2026
Reading activation drop curves without lying to yourself
A handsome day-1 to day-7 curve is often a sampling artefact. It keeps the people who already intended to finish and quietly drops everyone else out of the denominator. That is not Onboarding Drop-Off Analytics. That is a compliment to the product written in SQL.
Start with who is allowed to disappear
Before you plot anything, write down who can leave the cohort without counting as drop-off. Duplicate accounts? Employees? QA wallets? Invited admins who were never meant to log in? If you cannot name those exclusions in one paragraph, your curve will negotiate with you later.
In the flagship atelier we force this paragraph onto a single slide with no chart. Teams hate it. It is also the slide finance later trusts, because it shows you were willing to shrink the story.
Plot attempts, not just successes
Most warehouses store a completed KYC event and forget the started-and-abandoned twin. Without the twin, a falling line looks like “people lost interest.” With the twin, you often see interest colliding with a vendor timeout, a blurry ID photo, or a permission prompt on an older Android build — a pattern we keep meeting in Thai consumer products.
If engineering will not add the attempt event this quarter, say so in the chart subtitle. A curve that cannot see attempts should not speak in a confident voice.
Keep the quiet majority on the page
Silent accounts — signed up, no error, no ticket, no second session — are usually the largest group. Colour them. Do not park them in an “other” bucket under three percent. When we overlay interview notes from Qualitative Drop-Off Interviews, silence often turns out to be a colleague who was supposed to finish the job, or a user waiting on a salary cycle, not a UX failure you can nudge away.
A modest test for self-deception
Show the curve to someone who does not own the metric. Ask them what would have to be true for the line to look like this if the product were worse than you hope. If they cannot invent a plausible darker story, your chart is probably under-specified.
That exercise is week three of Cohort Funnel Diagnostics. You can steal it without enrolling. Please do not steal the vanity version where the darker story is never invited into the room.