Funnel drop-off rate

Direction is published as neutral because the shipped renderers disagree with the arithmetic. More drop-off is plainly worse, but neither shared oracle knows the field: one finds no classifying word in its name and falls through to its higher-is-better default, and the other classifies it as a plain count with no unit at all, so a rising leak can be painted as an improvement, and a fraction can be rendered as though it were a number of things. Never let the automatic colouring or formatting speak for this field.

This is a macro-funnel over aggregate counts, not a per-user event sequence. Drop-off is the relative decline between consecutive totals and is directional, not a per-user retention rate. The tool carries this caveat in its own payload.

Steps can span two different traffic scopes. Behavioural steps are read on organic search traffic while conversion and transaction steps are read across all traffic sources, mirroring the product's own card defaults, so a mixed funnel compares populations, and an all-source conversion count can legitimately exceed an organic session count above it. The tool discloses this when it happens.

There is no ordered step or event dimension in the model, so the caller supplies the order. A different step order produces different drop-offs from the same data, and the tool does not validate that the order is a real funnel.

Steps are drawn from a fixed list of count metrics, sessions, users, conversions, transactions, unique pageviews and events, and nothing else can be a step.

Because the step counts come from differently constructed measures, some approximate distinct counts and some plain sums, a drop-off between two of them mixes estimator behaviour with real attrition.

A step with no data is kept as absent and excluded from the arithmetic rather than counted as zero; fewer than two steps with data returns an explanatory note.

Whole-domain only; the result is stamped observational.