Improve Name Selection & Profits in Your Housefile (1263 words)
Regression modeling is one way to analyze and act on a more expansive view of transactional-predictive behavioral data. Performance gains of 10 percent to 15 percent are common when using data obtained through this technique compared with traditional RFM methods.
Those are great numbers, but even with substantial data covering your whole customer population, a model is limited. It defines predictive behavior with only the most significant predictive characteristics included in the equation. This means the model essentially is a best-fit equation defining expected performance. This can leave groups of customers under-represented in a housefile-scoring effort, resulting in low scores for groups that include potentially high-performing names.
So, is regression modeling for your housefile outdated and ineffective? Not at all. Model scores, when coupled with other predictive characteristics, can provide a more discriminating segmentation scheme coming into and out of a segmentation formula based on, say, offer, seasonality, recency and/or frequency. The key is to use all available tools in developing the most discriminating segmentation methodology you can.
It's not an either/or question; the best solutions consider all available analytical tools and customer intelligence when it comes to leveraging segmentation methods. For example, some catalogers using regression modeling as a housefile-segmentation tool have incorporated additional behavioral characteristics to define their segmentation to improve performance. These same catalogers have achieved additional gains beyond modeling of up to 16 percent.
Tip #4: Create segmentation from list interactions, not just "hits" files.
Comparing a housefile to rental files to help identify external buying activity can be beneficial. Housefile names that have little activity (e.g., inquiry names, giftees) or activity considered too old (e.g., 60-month singles), but that match active rental file names, can be profitable. Typically, catalogers segment these names into two groups: those with and without current outside activity.