skills/design-marketing-measurement/references/measurement-principles.md

# Marketing Measurement Principles

## Contents

1. Purpose
2. R, KGI, KPI, and operational metrics
3. Equation types
4. Evidence levels
5. Shared guardrails

## 1. Purpose

Measurement exists to improve a decision. It should identify the intended return, observe progress and failure, constrain causal claims, and specify what action follows. A report that cannot change a decision is descriptive overhead.

Marketing communications normally own contact, attitude change, recall, trust, and the path to action. Attribute sales directly only when a defensible counterfactual and control of material non-communication variables exist.

## 2. R, KGI, KPI, and operational metrics

- **R:** the primary return sought from the intervention.
- **Business KGI:** the organization-level result.
- **Intervention KGI:** the state or result the intervention can reasonably own.
- **KPI:** an intermediate variable connected to the KGI by an explicit hypothesis.
- **Operational metric:** delivery, process, or quality health.
- **Signal:** evidence that a hypothesis may be strengthening or weakening.
- **Guardrail:** the loss, lower bound, or condition that forces review, restoration, reduction, or withdrawal.

If a KPI reaches target but the KGI does not move, do not defend the KPI. Revisit the metric or causal chain.

## 3. Equation types

- **Meaning equation:** expresses responsibility or causal framing. Do not calculate it.
- **Accounting equation:** adds or subtracts countable components.
- **Probability equation:** represents conditional stages. Define population and period.
- **Hypothesis tree:** maps possible drivers. Do not calculate until units, overlap, saturation, and coefficients are specified.

Standard incremental definitions:

```text
Incremental effect = observed result − counterfactual result
Incremental CPA = additional advertising cost ÷ incremental conversions
Incremental ROAS = incremental revenue ÷ additional advertising cost
```

## 4. Evidence levels

| Level | Basis | Permitted claim |
|---|---|---|
| A | Controlled experiment or strong counterfactual | State a conditional incremental or causal effect |
| B | Multiple independent sources point the same way | State a strong hypothesis |
| C | Observational data, correlation, or a single source | State consistency, indication, or hypothesis |
| D | Experience, analogy, or insufficient evidence | State an exploratory hypothesis and limit action |

Separate facts, interpretations, hypotheses, and unknowns.

## 5. Shared guardrails

- Do not call observational association causal proof.
- Do not use a meaning equation or hypothesis tree as an accounting formula.
- Do not treat a proxy as the underlying construct.
- Do not choose a KPI merely because it is visible.
- Do not approve an intervention without a return definition, review period, and stopping rule.
- Do not use exact-looking thresholds without a basis.
- Do not suppress adverse long-term or customer signals to preserve a short-term KPI.