Funnels

Conversion through any multi-step flow. Chain the steps toward a goal, count how many profiles make it through each one, and see exactly where and why the rest drop off.

What a funnel is

A funnel is a goal expressed as ordered steps: the events a profile does on the way to something you care about, sign-up to purchase, install to first success. The funnel counts how many profiles did the first step, how many of those went on to each next step, and how many reached the end.

Three numbers summarize it. Entered is the profiles that did step one in the period. Completed is those that then reached the last step, in order. Conversion is completed as a share of entered. Everything else on the page explains the gap between the first two.

Each step is an event, optionally with filters (Purchase where amount is above 50). Steps can allow anything in between (sequential) or demand a strict sequence (exact), a completion window can bound how long the whole journey may take, and exclusion events can disqualify profiles entirely. A funnel can even convert by an entity other than the profile, so a team or an account converts when its members collectively finish the steps.

How a conversion is counted

The server walks each profile's event stream in order. A profile enters when it does step one; from there, each next step must occur after the previous one.

• Ordering. Sequential lets unrelated events happen between steps; the funnel only checks order. Exact stops the chain at the first non-matching event in between.
• The completion window. When set, the whole journey must fit inside it, counted from step one. Entrants that run out of time are abandoned, not completed.
• Exclusions. A profile that triggers an exclusion event is dropped entirely; it counts as neither entered nor completed.
• One entry per profile. Counts are per profile (or per the entity the funnel converts by), not per attempt, so a profile that retries five times is still one entrant.
• Caching. A saved funnel's result is reused for up to a day while Funnel caching is on, or for ten minutes while it is off, so a just-opened funnel can lag reality by that much. Editing the funnel starts fresh, and the page's Refresh recomputes on demand.

Use Cases

Find the leak

Find the step where the funnel loses the most, and size the prize.

1. Read the Biggest drop-off card. It names the transition losing the largest share of the profiles that reached it.
2. Confirm it in Funnel progress: the bars shrink at every step, and the steepest shrink is the leak.
3. Check the step table for the same transition: conversion vs previous, drop-off count, and how long the survivors took.
4. Click the drop-off count to open the profiles behind it, and read a few of their timelines to see what the leavers have in common.

Explain a step with a breakdown

A leak is a where; a breakdown is a why-shaped hint.

1. On the funnel bar of the suspect step, open the break down by control and pick a property, platform, app version, country, or any event property.
2. Read the rows under the bar: each value's share of the profiles who reached that step, sortable by largest or smallest share.
3. Break the step before it down by the same property and compare. A value whose share shrinks from one step to the next is losing people faster than its siblings, and that is your lead: one platform, one version, one country carrying the loss.
4. A profile can carry several values of a property, so rows need not add up to the step total; the note under the panel says so.

See where the leavers went instead

Profiles that dropped off did not do nothing; they did something else. The funnel can tell you what.

1. In Next event after dropping from, pick the step to inspect and press Run analysis. It computes on demand across the selected range.
2. Read the destinations: the events the leavers did instead of continuing, ranked. A support view or an error event here is a diagnosis.
3. Open the profiles behind any step or drop-off count for the individual stories.
4. Act on the audience: the preview panel lists exactly the profiles behind the number you clicked.

Judge whether a change helped

Tell whether the release, the experiment, or the redesign moved conversion.

1. Read Trends. Each point is one entry-day cohort: of the profiles that entered that day, how many completed. A change that helps shows as a step up after its ship date.
2. Mind the maturing tail: days inside the most recent completion window are still filling in, and the tooltip on the card explains which rule applies.
3. For before-vs-after or variant-vs-variant, save two funnels that differ in one step and use Compare from the list: conversion, entries, and completions side by side over time.
4. For audience effects, use the cohort comparison on the funnel page: profiles in a cohort against the rest, with the lift stated in points.

FAQs

Why is Entered lower than the first step's event count?

Entered counts profiles, not events. A profile that fired the first step's event twelve times is one entrant.

Does a profile that finishes twice count twice?

No. Within the period a profile converts once; the funnel measures how many got through, not how many times.

Why did a profile that did all the steps not complete?

Order and time. The steps must happen in order (strictly, for exact ordering), inside the completion window when one is set, and without triggering an exclusion event.

Can steps come from different sessions?

Yes by default: a cross-session funnel follows the profile across visits. A same-session funnel requires the whole journey inside one session.

Why do the numbers differ from what I computed in Drill?

Drill counts events and profiles per query; the funnel enforces order, windows, and exclusions on top. The funnel's Entered can only be smaller than or equal to the first step's unique count in Drill.

Why did the page say the funnel is running in the background?

The computation exceeded the request budget, so it continued as a background task. The page updates itself when it finishes, and the result stays available under Long runs.