How to choose the five KPIs that actually matter

The usual KPI process is additive: every department nominates measures, nobody removes any, and the result is a dashboard with forty tiles that no one uses to decide anything.

Run it subtractively instead. Start from decisions.

Step 1: name the decision the number should change

For each candidate measure, finish this sentence: *if this number moves the wrong way for two periods, we will ______.*

If you cannot finish it, the measure is diagnostic. Keep it in the metric library; do not promote it to a KPI. This single test usually removes half the candidates.

Step 2: one lagging, one leading, per goal

• Lagging — the outcome itself. Honest, but late: retention rate, reserve ratio, median permit time.
• Leading — an earlier signal you can influence this month: onboarding completion, share of applications submitted online, pipeline coverage ratio.

The pair is what makes a goal manageable. Lagging alone means finding out too late; leading alone means feeling productive while nothing changes.

Step 3: define it in writing, including what is excluded

Each definition needs:

• The formula, in words.
• The unit and direction (higher is better / lower is better).
• Exclusions — the part everyone forgets, and the part that causes two teams to publish different numbers.
• The population or scope.

Example. Weak: "customer retention." Defined: "the percentage of clients active on the first day of the quarter who are still active on the last day, excluding clients whose contract was scheduled to end in that quarter."

Step 4: name the owner and the source

One person. The person closest to the data, not the executive who reports on it. And a source: the system, the report or the query the number comes from, plus how often it is refreshed.

Measures pulled automatically from a source system get updated. Measures that depend on someone remembering to type a number get updated for two months. Where a number can come from a spreadsheet or a data warehouse on a schedule, wire it up — the reliability difference is larger than any dashboard design choice.

Step 5: baseline before you target

Record where you are, dated. Then set the target and date that too. Plans that set targets without baselines produce a predictable argument at the first review about whether things were always like this.

If a measure has never been collected, that is a legitimate baseline: "not previously measured; first value to be recorded in October."

How many is right

One to three per goal. With nine to fifteen goals that is roughly fifteen to thirty KPIs across the organization — but only three to five belong on the executive summary. Choose those by asking which numbers, taken together, would let you tell whether the year is working.

A worked example

Goal: residents renew routine permits online in under ten minutes.

• Lagging KPI: median online renewal completion time. Source: permitting system report P-14, monthly. Owner: Development Services Director. Baseline 22 minutes (Sep 2026). Target 10 minutes (Sep 2027).
• Leading KPI: share of renewals started online rather than at the counter. Source: same report, monthly. Owner: same.
• Decision rule: two consecutive months without improvement triggers a review of the online form's failure points.

That is a complete measurement design in five lines, and it can be reviewed by someone who has never seen the plan before.

Review your KPI set annually

At the annual refresh, ask three questions of each KPI: did it change a decision this year? Is the definition still accurate? Is the owner still the right person? Retire the ones that failed the first question — a KPI nobody acted on is costing update effort and attention for nothing.