Team Operations
KPIs: Choose Measures That Help You Decide
Choose useful KPIs, define their formulas and coverage, and turn a metric into a clear decision with a practical dictionary and worked examples.

What you will learn
- Define a KPI with its unit, denominator, period and source.
- Distinguish useful outcomes from activity and incomplete data.
- Connect a performance signal to an accountable next decision.
A KPI, or key performance indicator, is a selected measure used to assess progress toward an important objective. The word “key” matters: a business can collect hundreds of metrics without needing hundreds of KPIs. Choose the measures that would change a decision.
Start with the objective and write the definition before looking at the chart. The formula, population and period should be clear enough that a teammate can reproduce the result.
What makes a good KPI?
Tableau defines a KPI as a measurable value related to key business objectives. A count becomes useful when the team knows what it represents, why it matters and how it should be interpreted.
“Improve service” is an objective, not a measure. “Number of emails sent” is an activity count. “Share of eligible customer requests completed by the agreed date” may be a useful outcome measure, provided “eligible,” “completed” and “agreed date” are defined.
| Term | Meaning | Question it answers |
|---|---|---|
| Objective | The outcome you want to improve | What matters to the business or customer? |
| Metric | A defined observation or calculation | What happened? |
| KPI | A metric selected for a consequential objective | Are we progressing in the way that matters? |
| Target | An accepted desired level for a period | What are we aiming for? |
| Initiative | Work intended to influence the result | What will we try? |
Do not give every initiative its own executive KPI. Keep detailed operational measures where the people doing the work need them, and select a smaller set for decisions that require shared attention.
Build a KPI definition dictionary
| Definition field | Example for on-time delivery |
|---|---|
| Purpose | Understand whether customers receive accepted work when promised |
| Calculation | Eligible completed deliveries on time ÷all eligible completed deliveries |
| Unit and period | Percentage, complete calendar month |
| Eligibility | Deliveries with an accepted due date and completion recorded in the period |
| Exclusions | Canceled work, clearly documented under the same rule each period |
| Source and owner | Delivery register; operations owner responsible for reconciliation |
| Coverage and decision | Flag missing dates; investigate recurring causes before changing capacity |
Add the timezone, cutoff and treatment of reopened work when they matter. Save a version of the definition when it changes. Otherwise a chart can appear to improve because the denominator changed, not because customers received better service.
A threshold is a decision aid, not a universal standard. If the team lacks a reliable baseline, begin by measuring and checking coverage. Do not select a target because another company publishes an attractive percentage.
A KPI example with an explicit denominator
In this fictional service-business example, 24 eligible deliveries were completed during a month. Eighteen met the accepted due date and six were late. Four other accepted deliveries remained open at month-end.
On-time completed-delivery rate =18 ÷24 ×100 =75%. The four open deliveries do not belong in this particular completed-delivery denominator. They should appear in an open-work measure so the team does not hide overdue unfinished work.
| Measure | Calculation | What it does not prove |
|---|---|---|
| On-time completed deliveries | 18/24 =75% | Does not describe every request received during the month |
| Late completed deliveries | 6/24 =25% | Does not explain why the six were late |
| Open deliveries at cutoff | 4 records | Does not show their age or whether they are overdue |
If two of the 24 records lack a trustworthy agreed date, the original rate cannot simply be reported as complete. Resolve the records or state the coverage problem and the treatment used. Unknown is not the same as late, on time or zero.
Choose a balanced set, not a longer dashboard
Pair a desired outcome with a countermeasure that could reveal a harmful shortcut. Faster delivery may be useful, but not if work is returned unfinished. More leads may be useful, but not if sales teams cannot identify a relevant need. Lower support volume may reflect clearer answers or customers giving up.
- Prefer a measure close to the objective, while acknowledging delay and outside influences.
- Keep leading signals separate from realized outcomes.
- Use stable definitions and complete comparable periods.
- Inspect the underlying cases when a small sample moves sharply.
- Retire metrics that no longer support a decision.
For content-specific measures, use content performance measurement to keep impressions, clicks, visitor actions and signups distinct.
Connect the signal to a decision
KPI DECISION DICTIONARY
Objective:
Selected measure and reason:
Formula and eligible population:
Unit, period and cutoff:
Source, owner and coverage checks:
Baseline and target rationale:
Countermetric:
Signal that prompts investigation:
First question to investigate:
Person who can accept the next action:
Definition change history:Use the dictionary before the weekly business review. The meeting then deals with exceptions and decisions instead of debating what each chart means. A red status should prompt a useful question, not a demand for a confident causal story before anyone has investigated.
Copyable AI prompt
Try this with Aurora AI in Finta, Codex, or Claude to prepare your first draft.
Review [objective], [candidate KPI definitions] and [authorized aggregate records].
Build a metric dictionary with formula, denominator, eligibility, unit, period, source, owner and coverage.
Recalculate each example and distinguish activity, outcome, target and initiative.
Suggest one countermetric and one investigation trigger per selected KPI.
Mark unknowns.
Do not invent benchmarks, causal explanations or modify tracking.When a KPI moves, bring its definition and a complete comparison period to Finta’s Metric Change Investigator. Use it to prepare testable explanations and a clearer next question, with the source records beside the brief.