In the merge/purge, all prospect lists are placed in the same priority just below your house file names. Names appearing on any two or more lists, called multis, are randomly allocated among the specific lists to assure that these more responsive names similarly influence the response of each list. This creates a level playing field that allows for valid comparisons of response among all the lists.
This approach still works for catalogers. But it may be overly simplistic for many consumer mailers today given the changing environment.
Of co-ops and response rates
I recently examined the list performance of a 15-year-old consumer catalog at the behest of a potential investor. When the business started out, numerous lists were profitable — giving a response well above breakeven and allowing for list roll-out following tests. The business quickly grew, reaching its peak in sales after six years. As expected, the number of lists performing above breakeven slowly began to contract as they were stripped of responsive names by repeated mailing. But the contraction continued even after prospecting was reduced and sales growth leveled.
At the point of my review, the catalog was mailing only four continuation lists above a breakeven and justified mailing several others below breakeven based on a lifetime value. The sales of the business had declined 20% since its peak year; profits were dismal.
Since Abacus, the first cooperative catalog database, was created in 1991, numerous others have arisen. Today four catalog databases, Abacus, Z-24, Prefer Network, and I-Behavior, have reached or are approaching critical mass and are routinely used by catalogers as a source of prospects. And others, such as NextAction, are under development.
It’s safe to say that most consumer mailers that rent their names regularly contribute transaction information to co-op databases. Abacus alone has more than 1,600 catalog members. Therefore, it is likely that a substantial portion of the prospect names rented for a given mailing are also in the “score file” modeled by Abacus, the best of which end up being delivered in the model output to the cataloger.
In fact, the better the quality of the Abacus model, the more likely it will create multis with the cataloger’s best lists, and the multis themselves will include the names most likely to respond to the mailing. Multi creation is even further compounded if you use multiple cooperative databases, all of which to some extent are modeling names resulting from the same purchase transactions on the same catalogs.
So what are the implications of this new prospecting reality? New approaches in segmentation and merge/purge structure.
Slicing and dicing
First, it is no longer acceptable to evaluate lists and apply an all-or-nothing strategy. Rather, you must use the merge/purge and secondary optimization to separate the performing names in each list from the nonperforming names. Every list has great names and poor names; some lists simply have more of one or the other.
Second, it is no longer acceptable to evaluate multis as if they are one great homogeneous group of names. They must be sliced and diced in a variety of ways to determine how they perform based on their specific multiple sources.
Third, cooperative models cannot be viewed as simply another list in the merge. You can use their interaction with outside lists to isolate high-performing names from weaker names. There are also cost implications, since names from models cost less than rented names and more than exchanged names.
I routinely place the models from each co-op database and rental/exchange list into separate groups in the merge and isolate the crossover names between the groups as discrete types of multis (see chart, below left). With rare exceptions, the names that cross over between outside lists and two model sources perform better than any other prospect segments mailed. This stands to reason, since two models looked at the same names that were contained in the outside lists that I rented and identified the same subset of responsive names. Very often, 20% of the prospects in a merge fall into this high-response group. And since you paid for the names three times, you may mail them in three drops!
Almost as good are the crossover names between just Abacus and outside lists or just Z-24 and outside lists. Again, at least one model has evaluated at least a portion of the names you rented and identified those with the most potential. Also, crossover between the model outputs has identified great names even though they were not part of the outside lists — the models have the benefit of looking at many more names than you have rented for a specific mailing. It stands to reason that if two models look for the perfect prospect and agree on “John Doe,” they are much more likely to be right than if only one of them finds “John Doe.”
The more multis, the merrier
The chart directly below shows the response rates to various types of multis in a client’s recent mailing. While these results are typical of many of our clients, there have been significant exceptions. In some cases, multis have actually performed poorly. These exceptions usually occur with catalogs whose merchandise appeals to a very narrow, committed audience where propensity to purchase by mail is secondary to their interest in the product.
After you isolate the multis, you’re left with unique names — those from only one source, be it a co-op or an outside list. You need to trace and evaluate the performance of these names carefully to determine the value of a each source.
For instance, you may choose to mail only multis in the future and to discard all unique names. You may determine that you need to further model or optimize the unique names to separate the performing ones from the nonperformers.
Depending on your ability to use up the multis with future mailings, you may decide to no longer rent even high-performing lists if the performance was driven by the multi portion of the list and you can obtain the same names from a less expensive model. Or you may choose to exchange lists aggressively expressly to create multis against high-performing models, reducing your mailing costs.
Merge optimization in action
With the client we’ve been discussing, we brought 746,100 outside list names into the merge for a mailing (see chart above).
Once the merge was completed, 112,724 of these names hit the house file and were mailed under a house file key code. Another 280,611 names became multis with models, leaving 382,173 outside list names. Of those 382,173 names, 108,766 names were deemed mailable above breakeven, and the remaining 273,407 names were sent to Abacus for optimization. By optimizing these unique names at Abacus, 106,986 were identified as profitably mailable. Finally, 166,421 names were discarded.
All told, outside lists contributed more than 400,000 multis. Those that matched the house file performed at $5.37. The multis created by merging outside lists with models performed at $3.11 per book and, depending on how many hits they created, could be mailed two or three times. Names that appeared on more than one rented list performed at $2.87. Uniques names without optimization performed at $2.12 per book, and optimized lists performed at $1.80 per book — both within breakeven amounts.
In summary, you have much to gain by carefully crafting your merge/purge, family groups, and segmentation to isolate the unique names and multis from every model and list and to create different kinds of multis based on family group interaction. Through testing, you can isolate significant groups of nonperforming unique names to discard.
If you are a list purist, you can maintain multi segmentation under every individual list so that you can combine the segments and evaluate the entirety of a specific list. Conversely, you may choose to capture different types of multis under the model segments so that you can evaluate the performance of models in their totality.
John Lenser is president of Lenser, a list firm and catalog consultancy in San Rafael, CA.