Securities Brokerage CHAID Model
Note: if you are not highly familiar with CHAID segmentation modeling techniques, we suggest that you first visit our Introduction to CHAID page, and then return here to continue reading the brokerage case study.
Introduction This case study documents a predictive market segmentation model designed to identify and profile high-value brokerage customer segments as targets for special marketing communications efforts. The dependent variable for this ordinal CHAID model is brokerage account commission dollars during the past 12 months. Predictors include proprietary client data (various account status and trading behavior variables) as well as syndicated demographic and lifestyle variables. We begin by splitting the client's entire customer file into a modeling sample and a validation sample. (Once the model is built using the modeling sample, we apply it to the validation sample to see how well it works on a sample other than the one on which it was built).
CHAID Segmentation Model CHAID (Chi-square Automatic Interaction Detection) is a predictive, tree-based segmentation technique that uses a dependent or criterion variable as the seed for the segmentation. The tree is built by splitting an initial predictor variable into categories or ranges of values such that each category or range of values represents a statistically significant difference on the dependent variable. Additional predictors may be added under each category or value range of the first predictor, adding more branches to the segmentation model. A stopping algorithm determines when the model is complete, and the final “twigs” of the segmentation tree represent the final segments. Such a tree can be developed in automatic tree-generation mode, or it can be handcrafted by manually selecting predictors to use at various levels of the tree. Often a tree is first generated automatically; then, based on the specific business objectives of the analysis, as well as knowledge of the product/service category being analyzed, the tree is manually revised. In our example, the resulting CHAID model has 55 segments. [NOTE: while a typical CHAID segmentation model might optimally contain perhaps 12-15 segments, occasionally it is useful to develop a more extensive, more granular segmentation, which can sometimes help us identify and understand important “niche” segments that represent either a significant problem or an unusually lucrative opportunity. A more extensive segmentation model also allows us to take finer cuts at a marketing database. This can yield targeting precision that is functionally equivalent to that achieved by regression modeling, while also giving us a market segmentation that regression techniques are unable to provide.] For reasons of confidentiality, we will not display the segmentation tree diagram here. However, the results are summarized in the following comb chart, showing the segment indexes (indexes of average dollar value), and in the associated Gains table, farther down.
This summary comb chart is a quick way to confirm that the CHAID segmentation model discriminates degrees of customer value quite well. The top segment has an index of just over 900, reflecting an average annual commission value of just over $1,000. In contrast, the bottom segment has an index of just two, reflecting an average annual commission of less than three dollars. Next is the gains table, which provides quantitative detail useful for financial and marketing planning. In the gains table, we have highlighted the top 20% of the file in blue, the remaining above-average segments in green, the bottom 20% of the file in bright red the remaining below-average segments in dark red. Among other things, we can see that the top 20% of the file is worth an average of about $334 per account, which is nearly three times the average account value for the entire sample.
CHAID Gains table: Average Annual Brokerage Commission Dollars
We can use the data in the gains table to perform various financial calculations. For example, by multiplying the size of one or more segments by the average segment dollar value, we get a total value. Using this information, we can better plan our communications/promotion budget. Similar calculations performed on the under-performing segments provide information about potential cost savings achieved by reducing marketing efforts directed at these under-performers. We can perform additional diagnostics on the most lucrative segments to identify those customers who are below average for their segment. This means that they have the same or similar characteristics as their better-performing cohorts on the model's predictive variables, but are not providing as high a level of commission dollars. By using the demographic, lifestyle and trading behavior variables to define them, we can develop marketing and advertising communications strategies tailored to them, the goal being to convince them to trade at levels similar to their segment’s more lucrative counterparts. There are many other decisions which the gains table and the segmentation rules can help us make. For example, we might wish to conduct some market research among customers in under-performing segments, or among under-performing customers in the better segments. We can use the variables that comprise the segment definitions to help us identify possible issues and question areas to include in the survey.
Before we try to apply a CHAID model, we perform a validation against a holdout sample, to confirm that it is a
good model. In this example, our database was split randomly into equal-sized modeling and validation samples. We
can reconstruct the CHAID segmentation model on the validation sample, and examine the results. For example, when
we perform correlation analyses on the segments from the two samples, we obtain a correlation of approximately 0.98
for the segment sizes and approximately .97 for
the segments' average dollar value, indicating a very high degree of correspondence. Back to Marketing Analytics
page | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
. Marketing Analytics
. Market Research
. Operations Research
. Decision Analysis
. Project Management
|
Copyright © 1996-2010, SmartDrill 15 Royal Crest Drive,
Suite 4 Marlborough, MA 01752-2416 U.S.A. Phone (508) 485-2704
All rights reserved. No portion of this website may be copied or
used for any purpose without the prior written consent of SmartDrill.