Event frequency analysis

Event frequency analysis is the study of how often users perform a specific action, expressed as a rate per user over a defined period. If 4,000 users triggered a key event 12,000 times in a week, the average frequency is three times per user per week. The analysis then asks how that rate is distributed, how it shifts between cohorts, and whether it is rising or falling over time. The average is the start, not the answer.

The reason frequency matters more than a raw count is that habits live in cadence. A user who performs a core action once and never returns is very different from one who performs it every day, even though both register as having done it. Frequency separates the two. It is why the same event can look healthy by total volume and unhealthy by frequency, when a small group of heavy users masks a long tail who barely return.

This ties directly into stickiness and retention. The frequency of a core action is one of the strongest leading indicators of whether users will stay, which is why it underpins metrics like retention rate and stickiness ratios built on daily active users. Watching frequency by cohort shows whether new users are building the habit the product depends on, long before retention numbers confirm it.

The base measure is total event occurrences divided by active users in the period, giving an average frequency per user. The analysis then extends this across distribution, cohorts, and time, because the average alone rarely tells the real story. Each extension uses the same simple ratio applied to a narrower slice.

1. 1

   Choose the event and the period

   Pick a core action that reflects real value, not a passive event like a page view. Fix the period, usually a day, week, or month, and keep it consistent so frequencies are comparable across time.

2. 2

   Compute the average frequency

   Divide total occurrences of the event by the number of active users in the period. This is the headline rate, but treat it as a starting point rather than the finding.

3. 3

   Build the frequency distribution

   Bucket users by how many times they performed the event, for example zero, one, two to five, and six or more. The shape of this distribution reveals whether the habit is broad or concentrated in a few power users.

4. 4

   Compare frequency across cohorts

   Measure frequency by signup cohort, plan, or segment. A new cohort with rising frequency in its first weeks is forming the habit. A cohort whose frequency falls after onboarding is slipping toward churn.

5. 5

   Track frequency over time

   Follow the average and the distribution week over week. A stable average can hide a shrinking core of frequent users being replaced by occasional ones, which the distribution will expose.

The combination of average, distribution, cohort, and trend is what makes this an analysis rather than a single metric. Each view answers a different question: how much, how concentrated, who, and which direction. The metric tree below turns these into branches so that a change in any one of them points to a specific owner and a specific cause.

A metric tree decomposes event frequency into the factors that drive it, then connects each factor to the team that can move it. This is what turns a frequency chart into a set of owned levers rather than an interesting observation.

The first level splits frequency into the questions that determine it: how many users are even reachable to perform the action, what share of them adopt it at all, how often adopters return to it, and what friction limits each repeat. Each of these decomposes further. Reach splits into active base and notification deliverability. Adoption splits into onboarding exposure and first-use success. Repeat cadence splits into the trigger that brings users back and the value they get each time.

The decomposition matters because a falling frequency can come from very different places. Fewer active users to perform the action is a top-of-funnel problem. Adopters performing it less often is a value or friction problem. Reading frequency as a single number hides which it is. Reading it as a tree shows whether the fix belongs to growth, onboarding, or the team that owns the core loop.

Absolute frequency benchmarks are product-specific, because the natural cadence of a daily tool differs from a monthly one. The benchmark that travels is the shape of the trend and the share of users who reach a habit-forming frequency. The ranges below describe how to read frequency health rather than a target number.

A practical rule is to define a habit threshold for your product, such as performing the core action at least three times in the first week, and track the share of new users who clear it. That share is more actionable than the raw average, because it ties frequency to whether the habit is taking hold. Reading it alongside activation rate and feature adoption rate shows whether users both reach the action and keep returning to it.

Improving the analysis means making the signal trustworthy and then acting on the part of the distribution that matters. Cleaner instrumentation and a focus on distribution rather than averages sharpen the read. Acting on the right branch moves the number. Both are needed.

The metric tree approach starts at the branch with the largest gap between current and achievable frequency. If adopters are returning less often, the repeat cadence branch is the priority and acquiring more users will not help. If few users adopt the action at all, the adoption branch comes first.

KPI Tree attaches RACI ownership to each branch, so the return-trigger work has an accountable owner separate from the team that owns onboarding exposure, rather than frequency landing on product as a whole. When the frequency of a core action moves, the change is pushed to the owner of the branch that caused it. And because the platform checks whether an intervention actually shifted the number, you learn whether a new trigger genuinely raised cadence or merely moved existing activity around.

1. 1

   Reporting only the average

   A mean frequency hides whether the habit is broad or carried by a few heavy users. The distribution is where the real signal lives, and ignoring it leads to false confidence in stickiness.

2. 2

   Counting passive events as engagement

   Page loads, automatic refreshes, and background pings inflate frequency without reflecting intent. Track actions that signal real value, or the number measures noise rather than habit.

3. 3

   Mixing period lengths

   Comparing a weekly frequency against a monthly one produces meaningless conclusions. Fix the period and hold it constant so trends and cohorts are genuinely comparable.

4. 4

   Ignoring active-user denominators

   Dividing total events by all registered users rather than active users masks declining engagement. As the registered base grows, frequency per registered user falls even when real usage is flat.

5. 5

   Treating a stable average as good news

   A flat average can hide a shrinking core of frequent users replaced by occasional ones. Without watching the distribution over time, a healthy-looking number can mask a churn risk forming underneath.