How do I analyze how an event is distributed?

Short answer

Open All-platform analytics › Behavior › Behavior analysis and choose the Distribution model. Pick what to distribute: how many times each person fired one event, active days, a numeric property, pages per session, session duration, or the participation matrix. The result is a histogram with the count and share in each bucket, plus the mean and median.

Use distribution when an average hides the story: "3.2 orders per buyer" can mean most people bought once and a few bought dozens of times.

Choose what to measure

Set the buckets

• Custom edges: type ascending bucket edges, up to 50, for example 1, 2, 5, 10, 20.
• Default for counts and pages per session: widening edges such as 1, 2, 3, 5, 10, about 12 buckets.
• Active days: one bucket per day when the range is 31 days or less, otherwise the same widening edges.
• Numeric property: equal-width buckets from the minimum to the 99th percentile; this is the only mode where automatic bins (2 to 50, default 10) apply. Larger values fall into the last bucket.
• Session duration: fixed duration edges, keeping buckets up to 6 hours.
• Participation matrix: no buckets. List up to 30 event names, or leave it empty to use the 30 custom events with the most people.

There is no breakdown and no comparison period. Identity defaults to person.

Read the result

The histogram shows each bucket's count (people or sessions) and share, with the mean and median. When the mean is far above the median, a small group is doing most of the activity: worth finding out who they are.

On the participation matrix, the horizontal axis is breadth, the percentage of active people who fired the event, and the vertical axis is frequency, the average count per person who fired it. Read it in quadrants:

• Wide and frequent: core features most people use often.
• Narrow and frequent: power-user features; consider whether more people should discover them.
• Wide and rare: used once by many, such as setup steps.
• Narrow and rare: candidates to improve or remove.

Examples

• Purchase frequency: event count of purchase over 90 days with edges 1, 2, 3, 5 shows how many customers are repeat buyers.
• Engagement depth: active days over the last 30 days shows how many users are daily, weekly or occasional.
• Order value spread: numeric property revenue on purchase shows where most orders sit and how long the tail is.

Limits

• Event count mode needs an event.
• Numeric property mode keeps only values that parse as numbers, up to 200,000 rows. For mostly text properties, use the value list in property analysis instead.
• Matrix rows are custom events; names starting with a double underscore are skipped.

Related

• Create custom metrics
• Create user segments
• Mean vs median