Lilly Tech Systems
AI-Powered Audience Building
The foundation of effective retargeting is knowing who to target. Machine learning transforms raw behavioral data into intelligent audience segments ranked by conversion probability, enabling precise budget allocation.
Behavioral Segmentation with ML
AI clusters visitors based on behavioral patterns that predict conversion intent:
| Signal | Weight | Segment Implication |
|---|---|---|
| Cart abandonment | Very High | Highest-intent segment: specific product interest with purchase intent |
| Product page views | High | Active consideration: knows what they want, needs a push to convert |
| Category browsing | Medium | Research phase: broad interest but no specific product commitment |
| Homepage only | Low | Awareness stage: minimal intent, needs nurturing before hard sells |
| Repeat visits | High | Strong consideration: multiple touches indicate genuine interest |
Lookalike Audience Modeling
- Seed audience selection: Use your highest-value converters (not just all converters) as the seed for lookalike models
- Feature engineering: ML models identify the behavioral and demographic traits that best predict conversion from your seed audience
- Similarity scoring: Algorithms like k-nearest neighbors or embedding-based similarity rank prospects by resemblance to seed
- Expansion control: Adjust the lookalike similarity threshold to balance reach vs. precision based on campaign goals
- Iterative refinement: Feed conversion data back into the model to continuously improve lookalike accuracy
Predictive Scoring Models
Conversion Propensity
Gradient boosted models score each visitor's probability of converting within a defined window. Bid higher on high-propensity users.
Lifetime Value Prediction
Beyond single conversions: predict customer lifetime value to allocate retargeting budget toward the most valuable potential customers.
Recency Decay Models
Conversion probability decreases with time since last visit. AI models the decay curve to determine when retargeting becomes unprofitable.
Churn Risk Scoring
For existing customers, predict churn risk and deploy retention-focused retargeting before they leave, not after.
Building Your Audience Strategy
- Instrument tracking: Capture granular behavioral events (views, clicks, scrolls, time on page) beyond just page URLs
- Define conversion events: Clearly define what counts as a conversion for model training — purchases, sign-ups, or key actions
- Train propensity models: Use historical data to train conversion prediction models with proper train/test splits
- Create tiered audiences: Segment visitors into tiers (hot, warm, cold) with different bidding and creative strategies for each
- Exclude converters: Automatically suppress users who have already converted to avoid wasting budget and annoying customers