---
name: Marketing Analytics Science
description: The quantitative core of modern marketing — customer lifetime value, multi-touch attribution, churn prediction, and dynamic pricing — modelled honestly so budget, retention, and price decisions rest on defensible numbers and disclosed assumptions, not vendor-dashboard vanity.
audience: growth lead · marketing analyst · founder · product manager
---

# Marketing Analytics Science

## What this is
A method for the four quantitative pillars of growth: predicting **CLV**, **attributing** conversions across touchpoints, forecasting **churn**, and structuring **dynamic pricing** — each modelled with stated assumptions and validated against reality.

## What this is NOT
- **Not last-click gospel.** Attribution models are explicit about their assumptions; every model is a lens, and the skill names the lens rather than presenting one number as truth.
- **Not price discrimination or dark-pattern pricing.** Dynamic pricing here means demand-and-value-based structure with fairness and disclosure constraints — not surveillance pricing, drip fees, or exploiting the vulnerable (FTC).
- **Not a substitute for a data scientist or an economist** on causal pricing experiments or high-stakes revenue models; it frames and estimates, and routes the rest.

## When to use
Estimating CLV to set acquisition budgets; comparing attribution models before reallocating spend; ranking accounts by churn risk for retention; structuring a value-based or promotional pricing test.

## Operating principle
Every number carries its model. A CLV, an attributed conversion, or a churn score is only as good as the assumptions under it — so those assumptions travel with the number, and the model is backtested against what actually happened.

## Capabilities
- **CLV & attribution** — Goal: what a customer is worth and what earned them. Method: cohort or probabilistic CLV (retention × margin × horizon, discounted), attribution across first-touch / last-touch / linear / data-driven with the differences shown side by side, incrementality flagged vs correlation. Output: a CLV model + an attribution comparison, each labelled. Quality bar: CLV is backtested against realised cohorts; no single attribution model is presented as the truth.
- **Churn prediction** — Goal: who's leaving, in time to act. Method: propensity model on behavioural signals, ranked risk with precision/recall at the intervention threshold, reasons surfaced, sliced by segment. Output: a ranked risk list + an accuracy card. Quality bar: performance stated per segment; the model *ranks* outreach — the retention offer is a human decision.
- **Dynamic & value-based pricing** — Goal: price to value, fairly. Method: demand/elasticity estimation, willingness-to-pay by segment, scenario modelling of volume-vs-margin, fairness and all-in-disclosure constraints built in. Output: pricing scenarios with projected volume/revenue + a fairness note. Quality bar: projections labelled estimate; no hidden fees, no exploitative or personalised-penalty pricing; a real price change is validated by experiment.

## A worked example
"Where should the budget go?" → Data-driven attribution shows paid-social is over-credited by last-click by ~30%; CLV by channel (backtested on realised cohorts) says organic brings lower-CAC, higher-retention users. Budget shifts — but the deck shows all four attribution lenses and labels every projected figure as a modelled estimate, and the price test that follows is run as a real experiment, not toggled in a dashboard.

## Guardrails & escalation
Causal pricing decisions → a controlled experiment + economist review. Personalised pricing that could disadvantage protected groups → declined; route to legal (FTC/consumer-protection). Auto-executing spend or price → human sign-off. Pairs with the Statistical Modeling and ML for Business skills for the underlying models.

## References
CLV modelling (cohort / BG-NBD family); multi-touch and data-driven attribution; churn/propensity modelling; price-elasticity and value-based pricing; FTC guidance on pricing disclosure and unfair practices. Verify with controlled experiments; label estimates.