From Correlation to Causation

In short: Causal decisioning selects marketing actions by their measured incremental effect on an outcome, rather than by correlations that describe which customers already look valuable. Predictive analytics forecasts the status quo; causal decisioning changes it. This guide explains how, where today's "AI decisioning" falls short, and what a causal engine has to look like to move revenue.

For twenty-five years, across GE, Target, Lowe's, and Nike, I watched some of the best-resourced marketing organizations in the world make the same expensive mistake. They mistook correlation for a lever: they found the customers who looked valuable, spent more on them, and called it strategy. It rarely worked. Often it destroyed value.

Capable AI hasn't made this worse. It has, for the first time, made the right approach operational at scale: deciding by what an action will cause, not what a pattern merely describes. That is the difference between forecasting growth and compounding it.

A quick orientation that runs through the whole series: the Decision OS decides what to do; the Marketing Harness is where those decisions run, learn, and compound. Causal decisioning is the logic; the Harness is the layer that executes it across every channel.

The series

1. Predictive vs. Causal Decisioning, the correlation trap, told through LTV segmentation.
2. Causal Inference for Customer Retention, tighten effects, not forecasts.
3. Why AI Decisioning Systems Fail, cold start, boomerang effect, and local maxima.
4. Multi-Armed Bandits vs. A/B Testing, why smart companies still choose the worse option.
5. What Is Warehouse-Native Decisioning?, decisions where the data lives.
6. Decisioning-Centric vs. Model-Centric AI, the real capital race.
7. The Marketing Harness, where causal decisions actually get executed.

Key terms

• Causal decisioning: choosing actions by their measured incremental effect on an outcome, not by correlation.
• Predictive vs. causal: prediction describes what happens if you do nothing; causation tells you how to change it.
• Boomerang effect: a decisioning system hurting the KPI early because it starts without causal priors.
• Multi-armed bandit: dynamically shifting traffic toward better-performing options as results arrive.
• Warehouse-native decisioning: running decision logic in SQL inside the warehouse, not a separate model service.
• Decisioning-centric architecture: starting from the decision problem; the model informs priors, and RL optimizes allocation.
• Marketing Harness: the orchestration layer where causal decisions execute, learn, and compound across channels.

Who this is for

CMO and CDO/CDAO teams at D2C enterprises deciding how to turn first-party data into decisions that move ROAS, CAC, LTV, and retention, not another dashboard.