Product Metrics

Analytics interviews rarely ask you to compute something. They ask you to define it, and the definition is where the judgement is.

Every metric needs a precise definition before it means anything. "Active users" is not a metric until you say what counts as active — opened the app, or performed a meaningful action? Over what window? Does a background sync count? Two teams reporting DAU from different definitions is the most common source of arguments about numbers.

DAU over MAU is the standard stickiness ratio — of the people who used the product this month, what share use it on a given day. Around 20% means a typical user shows up roughly six days a month. It is only meaningful for products people should use daily; applying it to something inherently occasional, like a tax tool, produces a number that looks alarming and means nothing.

Retention is the honest metric, because acquisition can be bought and retention cannot. N-day retention asks whether a user returned on exactly day N; rolling or unbounded retention asks whether they returned on day N or later, which is far kinder and often more appropriate for products with a weekly rhythm. Quoting one while people assume the other is a common way to mislead without lying.

Cohort analysis groups users by when they joined and follows each group over time. It separates "the product is improving" from "we acquired a different kind of user", which an aggregate average cannot do.

Funnels measure conversion between steps. The useful question is rarely the overall rate — it is which step has the steepest drop and whether that drop is the same for every segment.

Beware vanity metrics. Totals that only ever go up — cumulative signups, page views — cannot go down and so cannot tell you that something got worse. Prefer rates, ratios, and cohorts.

Every primary metric wants a guardrail. Optimising engagement without watching unsubscribes, or conversion without watching refunds, is how a team ships a win that costs money.