Abstract
This paper examines how predictive customer analytics is changing marketing. It focuses on three
connected themes: anticipating customer behaviour, personalising campaigns, and supporting
marketing decisions. The central problem is that a model can predict who is likely to buy or leave
without indicating who will change in response to a particular message, offer, or channel. In other
words, prediction is not the same as cause.
The review brings together research on churn prediction, purchase forecasting, customer lifetime
value, recommendation systems, triggered communication, advertising measurement, marketingmix models, concept drift, privacy, and AI governance. It compares what the evidence supports
with the claims often made about marketing AI. The analysis also considers whether results
remain reliable when customer behaviour, markets, platforms, or consent conditions change.
The findings show that predictive tools can improve ranking, targeting, and planning in particular
settings. However, personalization can also create fatigue, crowding out, choice overload,
substitution, and privacy concerns. Attribution and predictive fit do not prove causal return on
investment. The evidence for fully autonomous strategic marketing decisions is weaker than the
evidence for bounded decision support.
The paper therefore recommends an experiment-calibrated, privacy-aware, drift-resilient causal
policy layer. This approach uses predictive scores, but tests actions against a control or baseline,
measures incremental economic and customer outcomes, monitors subgroup and long-term
effects, and keeps human accountability. The main conclusion is that marketing AI should
improve decisions without removing responsibility.
