Fictura
Insights & sessions

Building insights

Funnels, trends and retention over any event or action, with property filters, breakdowns, conversion windows and an audience filter.

Dashboard → Insights. Pick a kind, a date range and what to count, then press Run. The answer appears under the builder; the address bar updates with a ?q= link that opens the same query for anyone on the app's team (Copy link).

Three kinds:

KindAnswersDraws
FunnelHow many users went through these steps, and where they dropped off.A bar per step with the share of the first step, the share of the step before, and the median time to reach it.
TrendsHow a number changed over time.A line per series, plus a table with the sum over the range and the latest value.
RetentionOf the users who did X, how many came back to do Y in each later period.A cohort grid, one row per start period, each cell shaded by the share that came back.

What to count

Every step, series, start and return is either an Event or an Action.

  • Event takes a name. The picker lists the canonical names first (app_open, paywall_shown, trial_started, purchase_completed and the rest of the event dictionary), then your own events. Names merged into another, and the built-in old names such as purchase_succeeded, are left out: querying the canonical name already counts them (aliases).
  • Several names separated by commas (paywall_shown, paywall_screen_viewed) count any of them as the step.
  • Action picks one of the app's actions: a named group of events matched on screen, element text and properties. Actions are made on Dashboard → Actions.

Each one can be narrowed with Where…: property filters that apply to that step or series only.

Filters

Filters under Filters on every event apply to the whole query. Each filter is a property, an operator and (for most operators) a value.

OperatorValue
is, is not, containsText.
is one ofComma-separated values, up to 100.
matches regexA pattern, up to 200 characters.
>, ≥, <, ≤A number.
is set, is not setNone.

A property name is letters, digits and _ $ . -, up to 64 characters. Prefix it with person. to read the user's attributes instead of the event's properties: person.plan is pro.

Only users in limits the query to one audience.

Breakdown

Funnels and trends can be split by one property (store, product_id, placement, person.country), or by $audience to split by membership in the app's audiences. The top values are shown, 10 by default and at most 25. For a funnel, value at first step reads the property where the user entered the funnel; value at last step reads it where they finished.

Funnel options

  • Steps: 2 to 10.
  • Order: In order (other events may come between steps), Strict order (each step is the very next event), or Any order (every step inside the window).
  • Conversion window: how long a user has from the first step to the last, in minutes, hours or days.
  • First time only on step 1: count a user only on the first time they ever did that step, so a funnel of new users is not inflated by returning ones.
  • Exclusions: drop users who did an event between two steps (for example, paywall_dismissed between step 2 and step 3).
  • Series: up to 10. Each has a measure: total count, unique users, daily / weekly / monthly active users, or the sum, average, median or 90th percentile of a numeric property (price_usd).
  • Interval: hour, day, week or month. Hourly is for ranges up to 31 days, and not for weekly or monthly active users.
  • Compare to the previous period: draws the period just before the range as a dashed line in the same colour, lined up bucket by bucket, and adds the change to the table.

For unique-user measures the table's sum counts a person once per bucket, so it is not the number of distinct users over the range. Read the chart for those.

Retention options

  • Users who did, optionally for the first time (a new-user cohort).
  • and came back to do: the return event or action. Each cell counts returning users, or, with Sum a property per user, adds up a property of the return events (price_usd) and divides by the cohort's users: revenue per install by cohort.
  • Period: day, week or month; periods shown: up to 30.
  • Counts as returned: On that period (came back in exactly that period) or On or after (that period or any later one).

The Average row weights each cohort by its size and only includes cohorts that have reached the period.

Presets

The row above the builder holds ready questions about subscriptions, served by the backend: trial to paid by product, store, paywall and placement, revenue per install cohort by week, and same-session paywall conversion. One click loads the query into the builder and runs it; change anything and run again.

Limits

  • Date ranges are at most 366 days. Dates are days in your own time zone.
  • A query that runs longer than 10 seconds is stopped and the page says so. Shorten the range, remove a step or add a filter.
  • When the engine stops scanning at its cap, the result says the numbers may be low.

Those limits are enforced by the server; the builder checks the same ones before sending so you see why straight away.

On this page