Optimizing Your Audience with Machine Learning for Tailored Direct Mail Solutions
Effortless Cloud Operations: Reliable BAU Support, Maximum Uptime, and Simplified AWS Management
Direct mail might sound retro, but it’s still a powerful channel—if you know who’s going to respond. Our client mailed thousands of letters across different industries and creative themes. Response rates were all over the place, and postage costs were burning a hole in their budget. They needed a smart way to target households and businesses most likely to convert, and one‑size‑fits‑all rules weren’t cutting it.
Challenges
- Predictive targeting: The marketing team needed a scalable way to predict which households or businesses would actually buy when they received a mailer.
- Diverse verticals: Each campaign targeted a different industry—retail, healthcare, B2B services—so model behaviour had to adapt.
- Manual inefficiency: Manual segmentation and blanket rules sent too many pieces to people who weren’t interested, wasting time and postage.
The Solution
Knackforge turned decades of direct mail history into a labeled dataset (sale vs. no sale) and trained classification models using XGBoost and Random Forest.
- Data was split by vertical, with separate models tuned for each industry, so nuances like seasonal behavior or demographic quirks were captured.
- Feature engineering derived demographic, firmographic, and purchase‑history proxies, and SHAP summaries explained which features drove conversions.
- Predictions were piped into an automated filter that trimmed lists down to only high‑probability prospects while preserving coverage across priority segments.
- An automated ETL and scoring pipeline on AWS S3 allowed marketing to request optimized lists anytime, complete with explanations of why certain households were included.
The Impact
- Higher ROI: Compact, high‑probability prospect lists slashed postage and production waste while boosting conversion rates.
- Explainable targeting: SHAP‑based explanations gave marketers confidence to trust the models and tailor creative around the traits that actually drive responses.
- Repeatable playbook: The per‑vertical approach became a repeatable framework for future campaigns and new product launches, helping the client retain existing advertisers and win new ones.
Technologies Used:
- Python
- scikit learn (XGBoost, Random Forest)
- pandas
- SHAP
- AWS S3
- automated ETL & scoring pipelines