|
ZIP-level models, when
properly deployed, are powerful tools for the selection of prospecting
names from lists with large universes. Used regularly by many veteran
software publishers as well as multiproduct high-tech catalogers,
ZIP-level models can provide the necessary performance boost to
enhance testing activities and convert marginally performing lists
into profitable continuations. In this article, I'll define ZIP-level
modeling and several implementation strategies that will help you
attain maximized usage of your models.
Selection of prospect
names based on ZIP codes is an actionable tool for direct marketers,
since ZIP codes are common to all list universes. Perhaps you've
had a profile report generated on your customer base from external
information, yielding consumer demographic elements such as age
and income, and business elements such as SIC code, sales volume,
and number of employees. The problem in the 'real-world' usage of
this data is that these elements are typically not selectable on
outside lists. However, by equating unique customer characteristics
to geography, we are left with an element common to all lists for
the selection of names. The ZIP-level model is just that.
The creation of a ZIP-level
model is a structured procedure that deploys a variety of statistical
analysis processes performed on your house file and external data
warehouses, such as US Census data, automotive data (etc.) summarized
by ZIP code. The ZIP codes of your 24-month buyers are applied to
these and other external data sources where, using various regression
analysis techniques, statistically significant data elements are
isolated and ranked. These rankings are determined by both the summarized
data elements relating to the ZIP code as well as the current customer
penetration into each ZIP code. For example, a publisher of financial/investment
software might find that household incomes of $75,000+ and ownership
of 2 or more luxury cars are unique and defining elements of their
customers. This information is the precursor to the next phase of
building the model, which is to evaluate the significant attributes
found in the ZIP stream of your customers and identify other ZIP
codes with those attributes. This process results in the ranking
of all U.S. ZIP codes by the significant attributes from best to
least-best.
It is typical that this
ranking is then divided into equal units of ZIP code clusters, referred
to as deciles, with Decile 1 representing the very best grouping
of ZIP codes that relate to your current customers. While the root
of the word 'decile' implies a factoring of "10", most ZIP models
are divided into 20 equal panels, with Decile 20 representing your
"least-best" zip codes. A ZIP select tape is then created from these
panels which includes the deciles to be tested. Let's now examine
how to apply and maximize the usage of the models.
We'll start with the
ABC Consumer Software Company, which has been renting the active
subscribers from a computer magazine list with a 300,000 name universe.
|
|
This list has been a
proven performer for ABC's prospect mailings, delivering profitable
results on a consistent basis.
However, performance
on the file has gone from profitable to marginal, perhaps due to
list fatigue or shifts in the subscription offer that changed the
profile of the hotline subscribers (i.e. from a paid offer to a
free 6 month trial).
The ZIP select tape
can be applied to this universe of active subscribers, selecting
names within the geographies that best-match your current customers.
The deciles included on the ZIP select tape allow stratification
of the model, facilitating the creation of test panels so you can
determine how deep into the model you can mail to maintain profitability.
Results from the test might reveal that the application of Deciles
1 through 4, when applied to the universe of active subscribers,
will provide enough of a performance lift to make that list a profitable
component of your direct mail program. Obviously, the application
of the model reduces the overall universe of rentable names within
that list, however, it rejuvenates a list that would be otherwise
unusable.
As a marketer, you must
be careful on how the model is used. A common mistake is mailing
too deep into the model (i.e., arbitrarily mailing into Decile 10
out of 20). Skewing too far away from the best deciles can yield
diluted results. Consider mailing into Decile 2, and, from that
point, testing and going deeper into the model as needed. Remember,
the overall goal of the model is the skim the very best names from
a list to maximize your prospecting activities.
The ZIP model could
also be used to determine compatibility with lists before they are
rented. As you know, testing a list can be costly, averaging $2,500
for 5,000 pieces (combined paper, postage, and list costs). To gain
a better insight on the list, ask your List Broker to obtain ZIP
frequency counts (from the appropriate List Managers) on the lists
you are considering. A List Manager should be able to provide this
information. If the list is maintained in an SQL environment, an
SQL SELECT command can be issued to perform the query and generate
the results to a text file (i.e. SELECT zip, count(*) from
order by zip group by zip to file output.txt). You might ask for
this output file via an electronic transfer to compare the data
to your model using SQL or a spreadsheet program. Obviously, the
higher incidents of matches in your primary deciles will indicate
a better performing list.
Your List Broker can
guide you on the development and use of ZIP-level models. When used
correctly, they are powerful tools that will help improve the profitably
of your direct mail campaigns. * * *
|
