Using statistical and machine learning methods to represent audience groups, model their likely behaviors, and predict their responses to messaging or offers. It is the step between having data and knowing what to say to whom.
Also known as propensity modeling, audience prediction modeling
Audience modeling turns raw data into usable representations of how an audience behaves and what they are likely to do next. A lookalike model finds people who resemble your best customers based on behavioral patterns. A propensity model predicts which customers are likely to convert, churn, or upgrade within a given time window. A response model estimates how different audience segments will react to different messages.
These models are trained on historical outcomes: who converted, who did not, who stayed, who left. The model learns the patterns that distinguish one group from the other and applies that learning to new data. The output is a score or a segment label that drives targeting decisions in paid media, email, and sales outreach.
Audience modeling is not the same as audience description. A demographic profile describes who people are. An audience model predicts what they will do. That predictive dimension is what makes it operationally useful for campaign planning rather than just interesting for strategy decks.
Agencies that can model audience behavior, not just describe it, produce better-performing campaigns and more credible strategy recommendations. The ability to show predicted behavior rather than reported attributes is a material differentiator in competitive pitches.
Lookalike modeling extends reach intelligently. Rather than targeting only existing customers, lookalike models find new prospects who share behavioral characteristics with the best existing customers. The quality of the seed audience determines the quality of the lookalike, which means improving first-party data is strategic work, not a technical project.
Propensity modeling focuses effort. When a model can identify the 10% of the database most likely to convert in the next 30 days, marketing spend concentrates on them rather than spreading across everyone. That is a significant efficiency improvement with the same total budget.
Models decay and need maintenance. Audience behavior changes. A model trained on pre-2020 purchasing patterns performs poorly on post-2023 behavior. Agencies deploying audience models for clients need a retraining schedule, not just a launch plan. The model that was accurate last year may be confidently wrong this year.
An agency building a reactivation campaign for a subscription client runs a churn propensity model on the dormant customer base. The model identifies two groups: customers with high probability of reactivating with a discount offer, and customers with low reactivation probability regardless of offer type. The agency recommends concentrating the campaign budget on the first group and saving the second for a different approach. The client tests both. The propensity-targeted group reactivates at three times the rate of the control group selected by tenure alone.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.