|
 |  |
 |
 Overview Success Stories
|  |
Comparing Statistical Techniques When a major Dutch bank compared selection techniques for marketing purposes, Chordiant Decision Management turned out to be the absolute winner by all tested criteria. The ChallengeChordiant Decision Management was compared to various statistical techniques, (discriminate analysis, linear and logistic regression) to neural networks and to human marketing experience for which a selection was made by the banks product managers. With the exception of the neural network test (which was carried out by a specialised external bureau), all of the results reported below were produced by and at the bank. How It Was Done The assignment for each technique was to predict clients likely to show interest in a particular type of loan. All techniques were given the same data; a sample of 16,207 cases on which to develop models or selection rules. The sample contained 3,436 examples of customers who had a known interest in the offer and 12,771 who had not. The contestant models were tested on a validation sample of 14,040 existing customers. Of these, 6,376 had a known interest (45.4%). The bank's own data was enhanced with geodemographic information obtained through an external bureau. The different techniques were applied with minimal effort. As Chordiant Decision Management does not require any preliminary data preparation, the Decision Engine module was run in fully automatic mode. No user interaction was required whatsoever and no modifications were made to the data. The final Chordiant model was the result of a single run without any further optimization. The different techniques were compared on the basis of a number of statistical criteria, all of them in one way or another measuring the efficiency in which 'goods' (interested in loan offer) were separated from the 'bads' (not interested). 
The OutcomeAnalysis of the results showed that Chordiant Decision Management performed significantly better than all the other techniques on all of the criteria. To illustrate what this means in practice, the cumulative proportion of good cases at select rates ranging from 5% to 95% was calculated. For mailing purposes this would correspond to the response rate at various mailing rates, see figure 1. The figure also shows the percentages of customers that would have to be mailed in order to obtain about 5000 responders (out of 6376). The Commercial Implications Chordiant Decision Management enjoyed a significant margin of advantage over other techniques. For the same level of response the other techniques would have required mailing from 10% to 30% more offers. In interpreting these numbers it should be appreciated that small percentages typically are equivalent to a very substantial increase in profit. |  | |
 |
| © 2000-2008, Chordiant Software, Inc. All rights reserved. Legal/Disclaimer | Privacy Policy | Site Map | |  |
| Solutions | Retail Banking | Card Services | Lending | Insurance | Telecommunications | News & Events | About Chordiant | |
