Product Stickiness: Calculate It Without Mistaking Activity for Value

A product can become “stickier” for the wrong reason. Imagine an operations platform that starts sending more alerts. People return every day to clear them, so daily activity rises. Yet the alerts may reflect recurring failures, not a stronger product habit. Now take the opposite case: an automated service prevents those failures and requires fewer logins. Its daily activity falls while its value improves.

That tension is the useful starting point for product stickiness. In product analytics, stickiness usually describes how frequently active users return within a longer period. The familiar DAU/MAU ratio is one way to express it. It is not, by itself, a measure of satisfaction, retention, switching costs, or commercial value.

For a B2B team, the practical question is therefore not “How do we push this ratio higher?” It is “Are the right people or accounts repeating a meaningful action at the cadence their work requires?” Answer that first, and stickiness becomes diagnostic. Skip it, and a precise-looking percentage can reward noise.

What stickiness measures—and what it leaves out#

Amplitude defines stickiness as the number of different days or weeks on which a user performs an event within a weekly or monthly interval. That definition exposes the three choices hidden inside any stickiness figure: the entity being counted, the event that qualifies it as active, and the period in which repetition is expected.

Those choices matter more than the label. “Twenty percent stickiness” could describe people logging in, workspaces completing a workflow, or devices sending an event. It could use a daily numerator against a monthly population even though the product’s core job happens each Friday. Until the definition is visible, the number is not ready for a decision.

Calculate DAU/MAU with one population and one activity rule#

The standard calculation is average daily active users during a period divided by unique monthly active users during the same period, multiplied by 100.

Suppose a product has an average DAU of 2,000 and an MAU of 10,000 under the same identity and activity definitions. Its DAU/MAU stickiness is 20%. In an aligned 30-day period, that is also equivalent to six active user-days per monthly active user on average. This is arithmetic, not a target.

Average DAU matters. Dividing one unusually busy day by the month’s MAU would let a launch, deadline, or campaign stand in for normal use. The denominator must also cover the population from which the daily users came. A regional DAU divided by global MAU, or human activity divided by an MAU that includes service accounts, produces a ratio whose parts do not describe the same group.

Identity is another quiet source of error. One person using a laptop and a phone may appear as two device identities; a shared login may compress several people into one. In a collaborative product, user activity can rise while the number of participating customer accounts falls. Mixpanel’s account-analytics guidance makes the B2B distinction explicit: teams often need to inspect individual behavior and then aggregate it at the account level, where adoption and commercial outcomes are managed.

A defensible metric definition should state the entity, qualifying event, exclusions, timezone, window, and aggregation method. For example: canonical user ID; approval completed; employees, tests, bots, and retries excluded; reporting day in the customer’s operating timezone; average daily uniques divided by unique monthly users. If the decision concerns account health, define a second view around workspaces or customer organizations rather than pretending the user ratio answers both questions.

Match the cadence and active event to the customer’s work#

DAU/MAU is most interpretable when successful use is naturally daily. Sequoia’s product-health guidance recommends choosing DAU, WAU, or MAU according to expected usage and notes that the meaning of DAU/MAU depends strongly on the business and its expected cadence. Amplitude makes the same point in its guide to product stickiness: expense reporting, enterprise planning, and other products tied to external cycles should not be judged as if they were daily-use products.

This is a measurement design, not an industry standard. Its purpose is to keep the interval faithful to the job. A monthly payroll product does not become healthier merely because administrators are forced to visit it every morning.

The event definition needs the same discipline. By default, Amplitude counts a user as active after at least one active event in the selected interval, while allowing teams to mark events as inactive. Its documentation gives push notifications and messages received as examples of events that happen without a deliberate user action; marking such events inactive removes users who triggered only those events from active-user counts.

For your own product, choose an event that signals progress in the core job: an approval completed, an analysis run, an artifact shared with a colleague, or an exception resolved. A login is usually evidence of access, not evidence that value occurred. The right test is counterfactual: if this event increased while the customer’s outcome stayed flat, would the team still call the product stickier? If not, the event is too shallow.

Do not compare stickiness across periods until identity, activity, exclusions, timezone, and window definitions are identical or the effect of the change has been recalculated.

Turn a stickiness movement into a diagnosis#

A higher ratio means that the numerator grew relative to the denominator. It does not reveal why. Average DAU may have increased because established users returned more often, because automation started firing an event, or because a workflow acquired extra steps. MAU may have fallen after occasional users disappeared. All four situations can raise DAU/MAU, but they lead to different product decisions.

The average also hides distribution. Mixpanel demonstrates this with several worked examples: one user active on one day in a seven-day period and 1,000 users all active on the same single day both produce 14.29% DAU/WAU. In another example, an aggregate 20% ratio combines a highly frequent cohort with a much less frequent one even though no individual user has 20% stickiness. The ratio reports average frequency, not scale, timing, or the shape of behavior underneath it.

Use the movement as the start of an investigation:

1. Reproduce the definition. Confirm the exact event, identity, exclusions, day boundary, date window, and averaging method. A historical change in any one of these can create a break in the series without a change in customer behavior.

2. Separate numerator and denominator. Plot average DAU and MAU beside the ratio. A rise driven by stronger daily participation is different from one driven by a shrinking monthly population. Look at daily counts as well: a deadline spike and evenly repeated use can share an average.

3. Split the aggregate where behavior could genuinely differ. Start with use case, role, account maturity, plan, acquisition cohort, and platform. The purpose is not to produce every possible dashboard. It is to find whether the headline movement belongs to the customers, jobs, or environments affected by the decision in front of you.

4. Move from people to accounts. For B2B products, inspect both the user distribution within each account and the account-level result. Ten frequent users inside one large customer do not carry the same renewal implication as one active user in each of ten customers. Breadth matters: an account dependent on a single champion may look sticky at user level while remaining fragile.

5. Pair frequency with depth, retention, and outcome. Frequency says how often the qualifying action occurs. Depth says whether the workflow reaches a meaningful point. Cohort retention tests whether the same eligible population returns over time. The outcome measure asks whether the customer’s work improved. Contradictions are informative: rising frequency with lower completion may indicate added friction, while fewer sessions with stable automated outcomes may reflect a better product.

This sequence also changes how a team tries to improve stickiness. If the core event is healthy but concentrated in one role, the next question is whether the workflow should spread to collaborators—not how to manufacture more visits from the existing champion. If retained accounts repeat a workflow less often because they complete it faster, reducing friction may be more valuable than restoring the old activity level. If new cohorts visit once and vanish, investigate activation and the first completed job before adding reminders.

There is no context-free “good” rate. A published benchmark can be a useful prompt only when its products share your cadence, event definition, identity unit, platform mix, and customer lifecycle. Sequoia explicitly qualifies DAU/MAU by product and expected usage, while Mixpanel’s examples show that identical ratios can arise from materially different populations and timing. A stronger target is internal and conditional: for a defined cohort that has reached a defined stage, does repeated meaningful use predict a customer result the business cares about?

That target still needs testing. Compare cohorts or accounts with different frequency patterns and examine later workflow completion, expansion, renewal, or another supplied business outcome. An association does not prove that more frequent use caused the outcome; stronger customers may simply use the product more. It does tell the team whether the ratio has earned a place in the decision, rather than inheriting one from convention.

Make the metric answer one real decision#

Before putting stickiness on an executive dashboard, write the decision beside it. “Should we invest in weekly collaboration for mature customer accounts?” can support a coherent metric: eligible accounts, a collaboration event, active weeks per month, breadth across relevant roles, and a retained customer outcome. “Is engagement good?” cannot.

The most useful first move is to recalculate the last few periods with one meaningful event and the natural customer cadence, then split the result by account and one consequential segment. If the conclusion changes, the old ratio was averaging away the behavior you needed to see. Keep the qualified view. The headline number has no claim to survive its own diagnosis.

Frequently asked questions#