What is a Key Performance Indicator?

A key performance indicator (KPI) is a measurable value chosen to represent progress toward a specific goal, monitored over time, and used to make decisions: not merely a metric that exists, but one an organization has committed to act on. The "key" is the selective part. A product team can track hundreds of metrics; it can steer by a handful. A KPI therefore carries three properties an ordinary metric doesn't: a target or direction (what "good" looks like), an owner (who answers when it moves), and consequence (decisions change because of it). In research terms, a KPI is an operationalized latent construct ("product health", "customer success") reduced to a trackable indicator, with everything that reduction implies.

Leading and Lagging

Lagging indicators report outcomes after the fact: revenue, churn, renewals, NPS. They are what the business ultimately cares about and they arrive too late to steer by. Leading indicators move earlier and predict the lagging ones: activation within the first week, task success on core flows, weekly active usage of the feature that correlates with retention. The craft of a KPI system is finding leading indicators that genuinely predict the lagging outcomes (a correlational claim that deserves testing) and steering by those, while checking the lagging ones to confirm the prediction held. UX programs contribute here directly: usability metrics like task completion and SUS are leading indicators for adoption and support cost, when tracked consistently enough to trend.

Goodhart's Problem

The economist Charles Goodhart's observation, popularized in the paraphrase "when a measure becomes a target, it ceases to be a good measure", is the KPI's permanent hazard. Once a number carries consequence, the system optimizes the number by the cheapest available route, which is rarely the route the number was meant to encourage: support teams close tickets faster by closing them unresolved; onboarding completion rises because the steps were removed rather than improved; survey scores climb because the survey moved to happier moments. The KPI keeps improving while the construct it proxied deteriorates, and the dashboard reports success. Every KPI design is a bet on how the number will be gamed, and the honest ones plan for it.

Designing KPIs That Point at Reality

1. Start from the goal, then find the indicator.
"Users get value quickly" precedes "activation rate", and the readout should state why that indicator reflects that goal, the construct-validity rationale in one sentence.

2. Pair each KPI with a guardrail.
The counter-metric that catches the cheap route: ticket-closure speed with reopen rate, onboarding completion with 30-day retention, conversion with refunds. A KPI without a guardrail is an invitation.

3. Define it precisely and freeze the definition.
Numerator, denominator, window, exclusions, written down and versioned; redefinitions create false trends and quietly reset accountability.

4. Keep the set small.
Five KPIs get steered by; twenty-five get reported. Beyond a handful, a "KPI" is a metric with a title.

5. Keep a qualitative check on the construct.
Periodically ask users, in their own words, whether the experience the KPI claims to measure is actually happening; a short recurring study with open-ended video answers (the kind a Ballpark study collects alongside task metrics) catches decoupling before the lagging indicators do.

6. Test the leading-lagging link.
A leading indicator that stops predicting the outcome has become decoration; revalidate the relationship as the product and market change.

The Bottom Line

A KPI is a metric with a goal, an owner, and consequences: a proxy for something the organization cares about, chosen to be steered by. Derive it from the goal, pair it with a guardrail, define it precisely, keep the set small, check the construct qualitatively, and revalidate what it predicts. The number is never the point; it is a stand-in for the point, and it stays honest only as long as someone keeps checking that the two still agree.