Quick definition: A key performance indicator (KPI) is a deliberately selected metric used to assess progress toward a specific business, product, operational, or customer outcome.
What is a KPI?
A KPI is not just a number that is easy to report. It is a metric important enough to guide decisions, resource allocation, and accountability. A useful KPI has an owner, a target or decision boundary, a stable definition, and a clear connection to an objective. Revenue may be a company KPI, activation may be a product KPI, and payment-success rate may be an operational KPI. The right level depends on the decision.
KPIs differ from metrics, goals, and key results. A metric is any quantified measure. A goal is a desired direction or outcome. A key result is a measurable outcome used in an OKR system. A KPI is a continuing performance signal that an organization treats as important. The same conversion rate can be a dashboard metric for one team, a KPI for another, and a primary metric for a particular experiment.
A good KPI is decision-useful rather than merely flattering. Page views can be informative but may not indicate customer value. A subscription product might prioritize retained paying accounts, while an internal tool might prioritize successful task completion and reliability. Use leading indicators to detect movement early and lagging outcomes to validate lasting value.
KPI framework and calculation
Write a KPI specification with: objective; metric name; formula; numerator and denominator; unit; eligible population; time window; source of truth; cadence; owner; target; and exclusions. A rate has the form KPI rate = qualifying outcomes / defined opportunity population × 100. A monetary KPI might be net recurring revenue / active contracted accounts. The formula alone is insufficient: “active” and “net” require operational definitions.
| KPI type | Example | Decision it supports |
|---|---|---|
| Growth | Qualified leads per month | Is acquisition producing viable demand? |
| Product value | Weekly successful workflows per eligible account | Are customers receiving core value? |
| Economics | Contribution margin per order | Is growth profitable? |
| Reliability | Successful payment rate | Is the service meeting its promise? |
Targets should distinguish a monitoring reference from a causal threshold. A KPI target such as 95% payment success can trigger operations work, but it does not prove that a newly released checkout caused any change. To evaluate a change, compare a preplanned primary metric between randomized variants or use another credible counterfactual.
KPIs in A/B testing and product work
KPIs provide the context for choosing experiment metrics. Start from the decision and causal pathway: if the company KPI is retained revenue, an onboarding test may use activation as a near-term primary outcome only when there is evidence it predicts retention. Pair it with retention, cancellation, support, and reliability guardrails. Primary and guardrail metrics explains this hierarchy.
Do not automatically make the top company KPI the primary metric in every test. It may be too delayed or too noisy to detect a local change. Use a sensitive proximal measure with a documented relationship to the KPI, then validate the relationship through longer-term follow-up or holdouts. State the expected mechanism in the test plan; the hypothesis guide helps turn a KPI aspiration into a testable claim.
Predefine a single primary metric when possible. Testing many KPI variants, segments, and windows creates false discoveries. Report uncertainty and absolute impact, then assess practical magnitude rather than a p-value alone. Relevant guidance appears in multiple comparisons and confidence intervals.
Worked scenario
A SaaS company’s company KPI is net revenue retention, but its product team owns activation of new workspaces. It defines activation as an account completing three core setup actions within 14 days, divided by all newly created eligible workspaces. A setup assistant is tested among 8,000 workspaces per arm. Control activates 2,400 and treatment activates 2,640.
control activation KPI = 2,400 / 8,000 = 30.0%treatment activation KPI = 2,640 / 8,000 = 33.0%absolute change = +3.0 percentage points
The team checks the confidence interval, error rate, time-to-first-value, and support contacts. It does not claim a net-revenue-retention increase after 14 days. Instead, it records activation as the primary experiment outcome and follows cohorts to determine whether the lift predicts retained paid accounts. The KPI connection is a hypothesis to validate, not an automatic conclusion.
Data-quality limitations
KPI drift is common when event names, billing rules, identity resolution, or exclusions change. Version definitions, keep a change log, and backfill or clearly mark breaks in trends. A metric can improve simply because its denominator lost inactive users, a bot filter changed, or a team redefined a “qualified” lead. Audit numerator and denominator counts separately.
For experiments, verify that assignment and exposure are balanced. An unexpected traffic split or variant-specific instrumentation failure makes an apparent KPI lift unreliable; investigate with sample-ratio mismatch checks. Reconcile high-stakes financial KPIs with accounting sources rather than relying only on client events.
Common mistakes
- Calling every dashboard metric a KPI: prioritize measures tied to decisions.
- Leaving terms undefined: “active,” “qualified,” and “retained” need rules.
- Optimizing a proxy indefinitely: validate that it relates to customer and business value.
- Changing definitions silently: trend comparisons need version control.
- Using an aggregate KPI for every experiment: choose a causal and sensitive primary outcome.
- Ignoring harms: balance growth KPIs with quality, cost, and reliability measures.
Frequently asked questions
How many KPIs should a team have?
Few enough to focus action. Teams can monitor many diagnostics while maintaining a small set of accountable indicators.
Is a KPI always a percentage?
No. It can be a count, currency amount, duration, percentile, or rate, provided its definition matches the decision.
Can KPIs change?
Yes, as strategy and product maturity change. Document the reason and preserve comparable historical definitions where possible.
Should an experiment be judged by the company KPI?
Only if it can be measured with enough sensitivity and maturity. Often a proximal primary metric plus KPI follow-up is better.
Summary
A KPI is a precisely defined, decision-relevant performance indicator. Give it a formula, population, owner, source, and target; distinguish it from general metrics and experiment outcomes. In testing, connect local primary metrics to KPIs with evidence while protecting customer and operational outcomes.
Sources
- UK Government Digital Service: Choosing metrics
- Balanced Scorecard Institute: KPI basics
- NIST: Confidence intervals