Wow! It has been quite a while since I've posted. We've been busy reinventing cement (http://recocement.com) and beating modeling tests (two of the top modeling companies in the past 3 years) plus several others. A Ph.D. statistician told me this past week that we have a 'modified bayesian system'. He also said, "that happens to be what the consensus now admits achieves the most successful results... lucky you stumbled onto it." Well, I like to think it was more than a stumble... we just kept trying to make it better and better and now its the best. People ask me if it is better than RFM... please :). We had one client that retested and re-tuned their RFM system for 18 months, incorporating what we were teaching them. Nevertheless, when we would meet at the DMA in New York, they would buy dinner. "Why are you buying dinner?" "I'm the vendor!"
"Well, every time we use your system, we find about 20% more names than we would find without it... and they make money."
One of the tests which we contested recently, we generated twice the unique names of the competition. Now often, if you have more names, they aren't as good, but in this case, even the worst segments of ours beat the best of theirs. What this means is that our system is finding the good names in the bad spots and ignoring the bad names in the high segments - that incidentally is the best that modeling can do for you.
John <><
