Decisions read at a glance. If a stakeholder needs the legend, the slide is doing too much.
Three serif numerals, two sans weights, six saturated colours. Anything else is a special case.
Lead with form. Use weight, scale, and silhouette before reaching for icons or imagery.
Reserve full-bleed black for moments that deserve a beat. Never as wallpaper.
Three behavioural signals in the first 72 hours predict 18-month retention better than any feature-usage metric we tracked.
The first three days are the only window where customers are both paying attention and willing to write back. Every interaction here does the work of roughly four interactions in week three.
The cost of getting this wrong is not refunds — it is the long, quiet churn of an account that never returns to the inbox.
Cohorts that received a written welcome and a human reply within 24 hours retain at roughly 2× the rate of the templated cohort, and the gap holds steady for ninety days.
Translate the insight into a single behavioural hypothesis we can falsify in a sprint.
Smallest end-to-end change that lets the hypothesis be tested cleanly in one cycle.
Ship to a 50/50 holdout in one segment. Hold the line for two cycles before reading.
Pre-registered metrics only. Kill, scale, or extend — not three of three.
Promote to the default surface and retire the legacy path inside the same release.
Cohort with written welcome + human reply, vs. templated control. Sustained through D90.
Modelled on the current quarter's signup cohort, holding all other inputs constant.
Behaviour after the first 72 hours predicts 18-month retention better than any feature metric.
Three days in, someone wrote me a real sentence. I'd been a customer of theirs for nine months before I noticed I'd never been a customer anywhere else again.
Three pilots in eight weeks. We'll bring back evidence the quarter after, and the question won't be whether to ship — it'll be which two to default.
Confirm two of three by Friday. Owners named in the same conversation; calendars cleared.
Lock the metric, the holdout, and the kill criteria before any code ships. Analytics writes it; both sides sign.
One slide each pilot, every Friday, until the bet defaults or dies. No exceptions, no makeup decks.