Metric

Origin & evolution

Measurement long predates software, but product practice inherited a sharp distinction from the lean movement. Eric Ries, in The Lean Startup (2011), separated actionable metrics from vanity metrics: a vanity metric (total registered users, cumulative downloads) only ever rises and flatters the team, while an actionable metric ties a cause to an effect a team can influence. He tied this to cohort👥CohortGrowthA group of users sharing a common characteristicView reference → analysis, which tracks how each weekly batch of new users behaves over time rather than blurring everyone into one ever-growing total.

Dave McClure made the categories concrete with Startup Metrics for Pirates (2007), the AARRR framework: acquisition, activation, retention, referral, revenue. It mapped the five behaviours that decide whether a business works and gave teams a vocabulary for which number belongs to which stage of the customer lifecycle.

Alistair Croll and Benjamin Yoskovitz pushed the thinking furthest in Lean Analytics (2013). They argued a good metric is comparative, a ratio or rate, easy to understand, and above all changes how you behave, and they introduced the One Metric That Matters: the single number to obsess over at your current stage, switching as the business moves through empathy, stickiness, virality, revenue, and scale. Sean Ellis's North Star Metric, coined around 2010 and later codified into a framework by Amplitude, made the durable companion idea: one metric capturing the core value the product delivers, holding steady across stages.

How it works in practice

A subscription🔂SubscriptionSales & RevenueA recurring subscriptionView reference → team reports 1.2 million registered users in the board deck, and the number always grows. It is a vanity metric: it never falls, and it hides the fact that most of those accounts logged in once and vanished. They replace the headline with a ratio defined precisely as a metric: week-4 retention, the share of each weekly sign-up cohort still active four weeks later, where "active" means at least one project edited.

The new number is comparative (this week's cohort versus last month's), it can move in both directions, and it changes behaviour. When the March cohorts retain at 22% against January's 31%, the team has a real problem to chase, traceable to a specific release🚢ReleaseProduct SpecificationA shipped version of the productView reference →. The vanity number would have kept climbing the whole time, reassuring everyone while the product quietly leaked users.

Metric vs. its neighbours

• KPI is a metric promoted to a target with accountability and a threshold attached; the same definition becomes a key performance indicator when an owner commits to moving it to a stated level. Every KPI is a metric, but most metrics are diagnostic and carry no target.
• North Star Metric is the single metric chosen to represent core customer value for the whole organisation; ordinary metrics are the many supporting and diagnostic measures that feed and explain it. The North Star is the headline, the others are the inputs.
• Guardrail metric is one you watch to ensure a push on another metric does no harm, such as protecting refund rate while optimising conversion. It is a metric playing a constraint🔒ConstraintStrategyA constraint entityView reference → role, not a goal.

In the graph

In the Unified Product Graph, metric🔢MetricStrategyA unified metric that measures progress, health, or behaviour across the product is a leaf shared between the Strategy & Outcomes🏆OutcomeStrategyA desired business or user outcomeView reference → and Analytics regions, which is why a number can be both a board-level target and an operational measure. Its richest structure is self-referential: metric_decomposes_into_metricMetricdecomposes intoMetrichierarchy builds the input tree beneath a North Star, metric_drives_metricMetricdrivesMetriccausal captures the causal links between leading and lagging numbers, and metric_guards_metricMetricguardsMetricsemantic models the guardrail relationship directly. Upward, it quantifies strategy through key_result_quantified_by_metricKey Resultquantified byMetrichierarchy and outcome_measured_by_metricOutcomemeasured byMetrichierarchy; sideways, metric_segmented_by_personaMetricsegmented byPersonacross-domain splits a number by who it describes. Because guarding, driving, and decomposing are distinct edges, the graph encodes not just what you measure but how your measures relate, which is the difference between a dashboard🎛️DashboardData & AnalyticsAn analytics dashboardView reference → and a model.