We’d gotten a solid reputation for growing companies. We also worked with major mailers handling one of their most delicate and important tasks. IT people like procedures and routines. They do not relish working with unstructured opportunistic marketing requests. In several companies we were called just to take some of the workload temporarily. In this case, we signed on to manage circulation while a new order processing system was installed.
We received their data after New Years and managed a database, model and pull by February 10th. After their Ph.D. Statistician checked us out for a week and decided our methods were fundamentally sound, we were then tested against their internal RFM system. After beating that, we were tested against an algorithm they had purchased from a nationally known modeling firm. Their Ph.D. shared that we had done ‘substantially better’ than the two methods. He then created an alternative method which we also tested against. Each test was close to a half million pieces in total. This client had quite a few million customer records so the final pulls were taking a day and a half. We decided that this was way too long and tweeked it down to about 4 hours. This enhancement also sped up validation. I used to be able to take a coffee break during validation processing… but now it is too quick for that.
Their marketing people got involved when the conversion took quite a bit longer than expected. They were interested in taking advantage of their printer’s versioning capability with their variety of products. We got to the point of doing four models each month… dove tailing the pulls together to insure unduplicated names… very very complex and interesting. We also created dozens of new variables… new varieties of Recency, several varieties of ‘sale’ indicators… a few new geo-demographics and finally, a group of ‘werewolf’ variables (relating purchase patterns to full moons).
Overall, the mailing performance was solid, but we were never quite satisfied with our ability to corral customers into buying within the proper categories. It almost seemed that no matter what the category, customers bought the alternatives. Finally, the client reverted back to larger general books and began the testing processes again.
We suspect that the retail parent didn’t really believe in catalog marketing… so most of the tests were hold outs (where some customers were deliberately not mailed). In every case, the catalog more than made up for its cost with extra profit. At the same time, four different modeling alternatives from other modeling companies were tested. Three from large consulting/database companies and a forth developed by the retailer. We were not told about these tests until after the results came in.
There were four companies and eight tests, our unique names generated at least 12% more profit but in the ‘best’ case, 321% more (Even our worst cells out performed their best names). The average was probably in the 60% range.
“It was great working with John and his team. Their system seemed to produce consistently better results than both ours and some of the other modeling companies we tried. Best of all, they made working with data fun and understandable. John was always coming up with new variables… like ‘percentage of mobile homes’ or a dozen kinds of ‘sale indicators’. If we had problems, they were willing to burn the midnight oil to keep us on deadline. Even our IT people seemed to appreciate how it simplified their workload.”