Business Context
MarketWise, a digital marketing agency with 10,000 clients, aims to optimize its campaigns by understanding customer behavior through data analysis. The marketing team wants to segment customers effectively and evaluate the impact of campaigns using Bayesian methods, which can provide probabilistic insights and handle uncertainty well.
Dataset
| Feature Group | Count | Examples |
|---|
| Customer Demographics | 5 | age, gender, income, location, membership_duration |
| Campaign Interaction | 10 | clicks, impressions, conversions, engagement_rate, last_campaign_date |
| Purchase History | 8 | total_spent, avg_order_value, last_purchase_date, purchase_frequency |
- Size: 200K customer records, 23 features
- Target: Categorical — customer segment based on behavior (e.g., 'high_value', 'low_value', 'churned')
- Class balance: Relatively balanced, with 40% high_value, 40% low_value, and 20% churned
- Missing data: 10% missing in engagement_rate and last_purchase_date fields
Requirements
- Implement a Bayesian model to classify customers into segments based on their interaction with marketing campaigns.
- Evaluate the effectiveness of recent campaigns using Bayesian A/B testing techniques.
- Provide insights into customer behavior and segment characteristics through posterior distributions.
- Address missing data appropriately and ensure the model is interpretable.
Constraints
- The model should be computationally efficient enough to run on a weekly basis for 200K records.
- Insights must be easily understandable by non-technical marketing staff.
