KPI framework
Show how operating behaviour produces the headline result.
A KPI framework is a causal hypothesis about the business. It connects outcomes to controllable drivers, diagnostics and thresholds without treating every available measure as equally important.
The actual problem
The dashboard reports outcomes but does not explain what can change them.
Organisations often call every tracked measure a KPI. The result is a flat collection of activity counts, financial outcomes and operational diagnostics with no stated relationship between them.
A useful framework imposes structure. It distinguishes the result leadership is accountable for, the mechanisms that produce it, the early signals available before the result settles and the diagnostic measures used only when investigation is required.
- Too many prioritiesEvery function has a long KPI list, making trade-offs and accountability unclear.
- Outcome-only reportingRevenue, margin or retention is visible after the period but not connected to earlier operating signals.
- Activity mistaken for valueCalls, releases, visits or tickets are reported without evidence of their effect on an outcome.
- Conflicting optimisationFunctions improve local measures while degrading a shared customer or economic result.
- No decompositionA headline change cannot be separated into price, volume, mix, conversion, retention or capacity.
- No action thresholdA measure changes colour, but ownership and the required response remain undefined.
Method
Construct the tree from economic logic and decision use.
The framework is built as a set of testable relationships, not a visual hierarchy assembled from existing dashboard fields. Each branch should explain a mechanism and identify who can influence it.
- Outcome definition
- Select the small number of strategic results that represent durable value, with population, timeframe and accounting logic stated.
- Driver decomposition
- Express the outcome through identities or defensible mechanisms such as volume x value, acquisition - loss or capacity x utilisation.
- Leading signals
- Identify measures available early enough to change action and test their historical relationship with the later outcome.
- Guardrails
- Add measures that prevent local optimisation from harming quality, risk, customer experience or long-term economics.
- Decision mapping
- Assign each KPI to a decision, owner, review cadence, comparison point and response threshold.
- Validation and pruning
- Test definition stability, controllability, explanatory value and actual use; retire measures that do not earn their place.
Evidence required
What is needed to design and test the driver logic.
Management intent alone is insufficient. The framework needs operating knowledge, historical measures and clear definitions to distinguish a plausible story from a relationship the organisation can monitor.
| Input | Purpose | Key test |
|---|---|---|
| Business model and value flows | Define how the organisation creates and retains economic value. | Can the headline outcome be decomposed without double counting? |
| Decision and accountability map | Connect measures to people who can act. | Does the owner control a mechanism rather than merely report it? |
| Historical KPI series | Test stability, lag and co-movement. | Does the proposed leading signal move early and consistently enough? |
| Segment and cohort dimensions | Reveal different mechanisms hidden by aggregates. | Does the relationship hold across relevant populations? |
| Current targets and incentives | Identify optimisation risk and conflicting behaviour. | Could improving this measure damage a guardrail? |
| Metric definitions and lineage | Ensure measures can be reproduced and governed. | Will the same logic remain valid across reports and periods? |
Outputs
A KPI system small enough to govern and use.
The output is not only a diagram. Each relationship and measure needs a contract and an operating rule.
- Outcome and driver treeStrategic results decomposed into operating mechanisms, leading signals and diagnostics.
- KPI contractsPurpose, formula, population, timeframe, source, exclusions, owner and known limitations.
- Decision mappingDecision, cadence, comparison, threshold and expected response for each priority measure.
- Guardrail setMeasures protecting quality, risk, customer outcomes and long-term economics.
- Validation recordEvidence for proposed leading relationships, including lag, stability and segment variation.
- Retirement listDuplicate, unused or non-actionable measures removed from executive attention.
Worked example
Headline growth can coexist with erosion in the installed base.
A subscription business reports that annual recurring revenue increased from $20.0m to $22.0m. The driver tree separates new acquisition from movement in the opening customer base.
| Driver | Value | Effect on ending ARR | Owner question |
|---|---|---|---|
| Opening ARR | $20.0m | Starting base | Which cohorts and segments compose it? |
| New ARR | +$3.2m | Add | Which channels produce durable customers? |
| Expansion | +$1.1m | Add | What adoption precedes expansion? |
| Contraction | -$0.6m | Subtract | Which usage or service signals precede it? |
| Churn | -$1.7m | Subtract | Where is loss concentrated? |
Net revenue retention = (20.0 + 1.1 - 0.6 - 1.7) / 20.0 = 94%
Total ARR grew by 10%, but the opening base contracted by 6%. A framework reporting only ending ARR and new acquisition would reward growth while hiding deterioration in customer economics.
The KPI tree should retain ARR growth as an outcome, separate acquisition and installed-base drivers, and test earlier adoption or service indicators that can change expansion, contraction and churn before the period closes.
Limits
A driver tree is a model of the business, not the business itself.
The framework simplifies reality so it can support action. Relationships should be challenged as the business model, market and measurement system change.
- Not automatic causalityA leading relationship can be predictive without being a mechanism that an intervention will change.
- Not one tree for every audienceExecutive outcomes, operational controls and diagnostics require different levels of detail.
- Not target setting by formula aloneTargets also require capacity, investment, uncertainty and risk assumptions.
- Not permanentDefinitions, drivers and guardrails must be reviewed when strategy or operating conditions change.