Part 1: Inflow - acquiring new customers
Our focus is direct marketing, so we will not look at advertising campaigns but instead focus on understanding marketing campaigns (e.g. direct mail). This is the foundation for almost everything else in the course. We look at measuring and improving campaign effectiveness including:
The importance of test and control groups. Universal control group.
Techniques: Lift curves, AUC
Return on investment. Optimizing marketing spend.
Part 2: Base Management: managing existing customers
Considering the cost of acquiring new customers for many businesses there are probably few assets more valuable than their existing customer base, though few think of it in this way. Topics include:
1. Cross-selling and up-selling: _Offering the right product or service to the customer at the right time._ - Techniques: RFM models. Multinomial regression. - b. Value of lifetime purchases.
2. Customer segmentation: _Understanding the types of customers that you have._ - Classification models using first simple decision trees, and then - random forests and other, newer techniques.
Part 3: Retention: Keeping your good customers
Understanding which customers are likely to leave and what you can do about it is key to profitability in many industries, especially where there are repeat purchases or subscriptions. We look at propensity to churn models, including - Logistic regression: glm (package stats) and newer techniques (especially gbm as a general tool) - Tuning models (caret) and introduction to ensemble models.
Part 4: Outflow: Understanding who are leaving and why
Customers will leave you – that is a fact of life. What is important is to understand who are leaving and why. Is it low value customers who are leaving or is it your best customers? Are they leaving to competitors or because they no longer need your products and services?
Topics include: - Customer lifetime value models: Combining value of purchases with propensity to churn and the cost of servicing and retaining the customer. - Analysing survey data. (Generally useful, but we will do a brief introduction here in the context of exit surveys.) |