1 Marketing Analytics Foundations
Marketing analytics uses data to improve decisions about customers, products, channels, and budgets. A strong analysis begins with a decision and a target population, not with a dashboard or algorithm.
1.1 Start with the decision
A useful problem statement has four parts:
- Decision: What action could change?
- Outcome: What business result matters?
- Population and horizon: For whom, where, and when?
- Comparison: Relative to what alternative?
For example, “Should we move budget from paid search to connected TV next quarter to increase incremental contribution margin among new customers?” is more useful than “Which channel has the highest return on ad spend?” The first question identifies an action, outcome, population, period, and counterfactual.
1.2 Four analytical jobs
Marketing problems often mix four different jobs:
| Job | Question | Typical methods |
|---|---|---|
| Descriptive | What happened? | Aggregation, cohorts, dashboards |
| Predictive | What is likely to happen? | Regression, classification, forecasting |
| Causal | What changed because of an action? | Randomized tests, quasi-experiments |
| Prescriptive | What should we do? | Constraints, response curves, optimization |
A predictive model can rank customers without proving that contacting the highest-ranked customers will cause additional purchases. A descriptive trend can identify a problem without identifying its cause. Match the claim to the design.
1.3 Customer journey and funnel
A simple funnel may include awareness, consideration, conversion, retention, and advocacy. It is a planning abstraction, not a literal path followed by every customer. People can skip stages, move backward, use several devices, or purchase offline.
Use funnel metrics to locate friction:
- reach and qualified reach for awareness;
- engaged visits or product exploration for consideration;
- conversion rate and contribution margin for acquisition;
- repeat purchase, retention, and churn for existing customers; and
- referral or advocacy measures when they are tied to a real program.
Do not optimize a stage in isolation. Increasing low-quality clicks can raise traffic while reducing conversion rate and profit.
1.4 Unit economics
Revenue is not the same as value. A campaign can produce positive revenue and still destroy profit after media cost, discounts, fulfillment, returns, and service costs.
Common quantities include:
\[ \text{CAC} = \frac{\text{acquisition cost}}{\text{new customers}} \]
\[ \text{ROAS} = \frac{\text{attributed revenue}}{\text{advertising spend}} \]
\[ \text{Incremental ROAS} = \frac{\text{incremental revenue caused by advertising}}{\text{advertising spend}} \]
ROAS depends on an attribution rule; incremental ROAS requires a credible counterfactual. Neither should be compared across teams without checking the numerator, denominator, time window, and treatment of returns and discounts.
1.5 A practical workflow
- Translate the business request into a decision and estimand.
- Draw a simple causal or process diagram before selecting fields.
- Define the metric, unit of analysis, eligible population, and time window.
- Audit instrumentation, missingness, duplicates, and identity resolution.
- Choose the least complex method capable of supporting the claim.
- Quantify uncertainty and perform sensitivity checks.
- Recommend an action with expected benefit, risk, and monitoring criteria.
1.6 Interview checkpoints
- Why can a high-ROAS channel still be a poor place for the next budget dollar?
- Give one example each of a descriptive, predictive, and causal marketing question.
- What information is missing from the statement “conversion increased 10%”?
- How would you prevent a team from optimizing clicks at the expense of profit?