These case studies show how I apply marketing science to revenue problems: the business problem, the method, and what the work delivers. Each links to the full article, with data, code and charts.
Methods: XGBoost and gradient boosting, SHAP explainability, survival analysis, RFM and predictive CLV, association rules, Markov chains, collaborative filtering, deep learning ensembles, AI agents. Tools: Python, R, SQL, Salesforce, AWS and GCP.
B2B Pipeline Intelligence (three-part series)
Problem. Sales leadership sees pipeline problems too late to act on them, and every open deal gets the same attention.
Method. Funnel conversion, deal-velocity and RFM diagnostics; an XGBoost propensity model trained on Salesforce pipeline data to predict Closed Won; leakage audits and GTM score thresholds to move the model into production.
What it delivers. Pipeline health visible up to three quarters ahead, deals ranked by likelihood to close, and scores wired into the forecast.
Prospect Propensity: The Cold Start Problem
Problem. A single blended propensity model looks accurate overall but quietly fails on prospects with no purchase history.
Method. Split customers and prospects, compared model performance for each population, and made the case for a separate prospect model.
What it delivers. Exposed a prospect gap nine times larger than the blended model showed, and a clear case for a dedicated acquisition model.
Customer Retention: Who Will Churn, and When
Problem. Retention programs know which customers are at risk but not how much time is left to intervene.
Method. Gradient boosting (XGBoost) for churn risk combined with survival analysis for time-to-churn, demonstrated on the IBM Telco dataset.
What it delivers. A ranked list of at-risk customers with an expected churn window, so retention spend goes where and when it matters.
Cross-Sell Recommender Systems (three-part series)
Problem. Cross-sell offers are generic, and customers are not shown the next product they are most likely to buy.
Method. Market basket analysis (association rules), Markov chains for the customer journey, and collaborative filtering for look-alike behavior.
What it delivers. Next-likely-purchase recommendations for B2B and B2C cross-sell that move high-potential customers toward the Champions segment.
Segmentation and Customer Lifetime Value
Problem. Account lists are prioritized by size or gut feel rather than by value and likelihood to buy.
Method. RFM segmentation, predictive customer lifetime value, and contact-level personas beneath the account model.
What it delivers. A portfolio view of accounts by value and potential, and targeting that reaches the right contacts inside each account.
Forecasting and Channel Effectiveness
Problem. Revenue forecasts ignore field knowledge, and channel plans treat digital and direct outreach separately.
Method. Hybrid forecasts that combine field sales judgment with statistical and deep learning ensembles; multi-channel lift analysis of digital nurture plus direct outreach.
What it delivers. More accurate B2B revenue forecasts, and evidence that coordinated channels triple conversion lift.
AI, Agents and Data Quality
Problem. AI applications are only as trustworthy as their data, and teams are unsure where human judgment still matters.
Method. Human-in-the-loop data quality checks, a head-to-head test of an AI agent against my own model interpretation, an AI agent built on my methodology, and generative engine optimization.
What it delivers. Practical guidance on where to automate, where to keep people in the loop, and how to make content visible to AI assistants.
Interested in applying this work to your organization? See Consulting & Fractional Leadership, my resume, or the full Article Archive.