Distribution
Distribution shows how uneven a behavior is: most people once, or a few people many times. Coverage and the value list of one property are Property analysis; the chart you can draw in the console today is Property value here.
ConsoleAll-platform analyticsBehaviorBehavior analysisHow to configure
Pick one subject: how many times each person fired one event (the event is required), active days, a numeric property, pages per session, session duration, or the participation matrix.
Behavior is the buckets. You can type ascending edges, up to 50. Without them, event counts and pages per session use widening edges such as 1, 2, 3, 5, 10, about 12 buckets. Active days are one bucket per day when the range is 31 days or less, otherwise the same edges. A numeric property is split into equal widths from the minimum to the 99th percentile, and that is the only mode where automatic bins (2 to 50, default 10) apply. Session duration uses fixed duration edges and keeps only buckets up to 6 hours. The matrix has no bins: up to 30 event names, or the 30 custom events with the most people when the list is empty.
No breakdown and no compare period. Counts, days, and properties are per subject under the current identity. Pages and duration are per session. Identity defaults to person.
How to read the result
The histogram shows each bucket’s count and share, plus mean and median. On the matrix, breadth is the percent of active people who fired the event, and frequency is the average count per person. High and wide means many people use it often.
Limits & definitions
Event-count mode fails if you do not pick an event. Property values keep only rows that parse as numbers, at most 200,000, and the automatic top edge is the 99th percentile; larger values fall in the last bucket. Active days on person identity read the daily person table; other identities count distinct local dates on events. Matrix rows are custom events and skip names that start with a double underscore. Breadth’s denominator is distinct subjects under the current filters, not a second pass with internal events removed. If a property is mostly text, use the value list in Property analysis instead of number ranges.
Interval
Interval measures the time gap, not whether people got there; that is Funnels. “Are they back on day N” is Retention.
ConsoleAll-platform analyticsBehaviorBehavior analysisHow to configure
The measurement is a start event and an end event. Both are required, written like funnel steps: events or paths, several names as OR.
Behavior: pairing defaults to the first A, then the first B after it. The other mode is every A, then the next B after that A. Scope defaults to the same person, or the same session. The maximum gap defaults to 30 days. The form accepts minutes, hours, or days, up to 180 days. Pairs beyond that are dropped. A hold property requires the same value on both sides, such as one order id.
No breakdown. Granularity drives the median trend. No compare period. Identity defaults to person; session scope pairs on the session id instead.
How to read the result
The summary is sample size, minimum, 25th, median, 75th, 90th, maximum, and average, all durations. The histogram uses fixed duration edges that stop at your maximum. A second line is the median gap inside each time bucket.
Limits & definitions
The search for B extends the query end by the maximum gap. At most 200,000 pairs are kept; the rest are cut off. B must be strictly later than A. An A with no B is not a sample, so the sample count is smaller than the count of A. In first mode, one person or session keeps only the earliest A for a given hold value. An empty hold value on both sides counts as the same value. The median trend buckets by the site timezone, using the same time buckets as event analysis.