Glossary

What is a KPI?

A key performance indicator is a quantifiable metric selected to evaluate progress against a campaign or business objective — a CPA target, a ROAS threshold, a retention rate.

Analytics and data

All 220 terms

What makes a KPI a KPI

The word key is doing the work. A KPI is not any metric you track; it is the small number of metrics that determine whether you are succeeding, and against which people are held accountable.

That accountability is what makes the choice consequential. People optimise what they are measured on, including in ways nobody intended, so a KPI is less a measurement than an instruction about behaviour.

The wrong KPI is worse than none

Because it will be achieved. A team measured on install volume will deliver installs, including worthless ones. A team measured on attributed ROAS will shift budget toward channels that attribute well rather than channels that work — which usually means defunding upper-funnel activity and over-investing in retargeting.

This is not a hypothetical failure mode; it is the normal outcome. Any KPI that can be satisfied without producing the underlying value eventually will be.

The defence is choosing metrics that are hard to satisfy dishonestly, and pairing every efficiency KPI with a volume or quality one so that gaming shows up somewhere.

Choosing them

  • Tie each to a decision. If no value would change what anyone does, it is not a KPI.
  • Pair efficiency with quality. CPA alone invites cheap worthless users; CPA plus retention does not.
  • Prefer metrics you can defend causally, or state explicitly that the number is attributed rather than incremental.
  • Keep the set small. A team with twelve KPIs has none.
  • Set the target with the metric. A KPI without a threshold is a chart.

Common questions

How many KPIs should a team have?

Three to five for most teams. Beyond that, priorities conflict and the set stops guiding behaviour, which was the point of choosing them.

Should KPIs change?

Deliberately and infrequently. Changing them mid-period makes performance incomparable and is often how a disappointing result gets reframed rather than addressed.

More in analytics and data

Where the numbers are stored, shaped and read.