Click for a Printer-friendly Version - Adobe PDF
Integrate Skill Sets with Data Mining Techniques
By Jim Wheaton
Principal, Wheaton Group
Original Version of an article that appeared in the
February 10, 2003 issue of "DM News"
Sophisticated target marketing is a process that requires significant human
input coupled with the intelligent integration of multiple data mining
techniques. The human input is most effective when individuals with
disparate skill sets work as a team to develop innovative targeting
Generally, such a team consists of direct marketers, creative professionals,
and data miners. The data miners are quantitatively oriented
professionals who must be well versed in a wide array of quantitative
techniques such as predictive models, clusters, demographic profiles, and focus
group and survey research. They must also understand how to combine these
techniques into a robust foundation for sophisticated targeting.
Few catalogers - or, for that matter, direct marketers in general - have
successfully blended multiple skill sets and data mining techniques into a
well-considered program of targeting. This article provides a blueprint
for doing so.
A statistics-based predictive model is a mathematical equation that
rank-orders individuals in terms of most to least attractive future
predicted behavior. Generally, this rank ordering is divided
into equal sized groups of similarly performing individuals (e.g.
A predictive model generates heterogeneous rather than homogeneous segments;
that is, segments containing individuals with no guaranteed characteristics in
common beyond their future predicted behavior. For example, customers
within a given segment might be a combination of the young and old, as well as
males and females. Also, they might display many different patterns of
historical merchandise purchase behavior.
With a predictive model, all database fields with the potential to isolate the
"goods" from the "bads" are systematically evaluated. The model itself
can be easily implemented into a production environment. All customers
above a predetermined predicted performance are promoted, and the balance are
not. Therefore, models are an advanced way to help determine whom to
Sophisticated targeting, however, also requires insight into what to
promote. This is where the additional techniques of clusters, demographic
profiles, and focus group and survey research come into play.
Unlike predictive models, clusters provide
segment homogeneity. By definition, segment homogeneity exists
whenever a group of individuals has at least one thing in common.
Examples are life-stage, merchandise category needs, or permutations
Segment homogeneity is a prerequisite for one-to-few marketing.
One-to-few marketing - unlike its much-hyped cousin, one-to-one marketing - is
almost always cost-effective.
An Example of Clustering
Consider a form of clustering
called "product affinity analysis," in which groups of customers
are defined by their merchandise purchase patterns within and across
orders. Assume that six product affinity clusters are created,
and that a given customer has purchased just once - a single item
within Cluster #1:
Ad hoc efforts can be made to sell other items within Cluster #1, as follows:
Web recommendation agents, at the time of purchase, when the medium of purchase
is the e-commerce site.
E-mail micro-targeting, subsequent to the purchase.
Ink-jet messaging, subsequent to the purchase, with a catalog cover "call out"
involving one or more of the Cluster #1 pages.
Interactive call center efforts, either at the time of the purchase or during
Layout fine-tuning, subsequent to the purchase, for both print media and the
e-commerce site. This allows the positioning of merchandise to be
adjusted to reflect typical purchase patterns.
Formal specialized predictive models can be implemented, as follows:
- "Affinity group" models, to rank-order Cluster #1 buyers in terms of their future predicted purchase volumes across Cluster #1 merchandise.
- "Cross sell" models, to rank-order non-Cluster #1 buyers based on their likelihood of eventually purchasing Cluster #1 merchandise. This is most often done with high-value clusters, to drive "prospecting" efforts within the customer base.
Affinity group and cross sell models can drive focused initiatives such as
merchandise-specific special offers, including email. They can also
spearhead selective binding involving supplemental signatures of Cluster #1
Focus Group and Survey Research
models, focus group and survey research provide attitudinal insight.
Unfortunately, many - and perhaps most - catalogers do not systematically
employ such research.
A specialty cataloger replaced RFM Cells
with a statistics-based predictive model. As a result, wasteful
circulation was eliminated, and the significant promotional savings
were reinvested in sophisticated targeting programs as follows:
On average, males were one-half as responsive as females. Using
clustering techniques, a subset of very responsive males was identified:
those who had purchased female-oriented jewelry.
Unfortunately, by analyzing the database itself, there was no way to determine
who in the household was driving the actual activity. It could, for
example, have been daughters using their father's credit cards. Or,
perhaps men purchasing gifts for the significant women in their lives.
Subsequently, this customer subset of responsive males was overlaid with
demographic information such as age, income, marital status, and presence of
children. The results indicated that these jewelry-buying households were
families with children, living in single-family suburban homes, with
professional, technical and managerial occupations.
Knowing that the target audience was married suburbanites rather than single
city-dwellers was helpful in tailoring the catalog copy and layout.
Nevertheless, it provided no insight into the individual within the household
who was driving the jewelry purchases. To gain a definitive answer, focus
group and survey research was commissioned.
The research indicated that the majority of these individuals were gift-giving
husbands. They were what the research company dubbed "unimaginative male
gift givers." These were men who - despite their solid professional
success - dreaded purchasing birthday, anniversary and holiday gifts for their
spouses. They were at a loss for what kinds of presents their wives might
In order to fully leverage these findings, a task force was convened.
Comprised of representatives from marketing, creative, and analytics, the task
force's mandate was to develop a loyalty program to appeal to these
"unimaginative male gift givers."
On the prospecting side, the cataloger's circulation department began working
with its list broker to identify male-oriented lists for which to target
prospect offers. These offers included a description of the loyalty
program as well as a form for signing up.
Over time, the cataloger was able to extend aggressively into a new and very
different target market. Ultimately, its top and bottom lines were
enhanced significantly because of the combination of multiple skill sets and
data mining techniques.
Jim Wheaton is a Principal at Wheaton Group, and can be reached at 919-969-8859
or firstname.lastname@example.org. The firm specializes in direct marketing
consulting and data mining, data quality assessment and assurance, and the
delivery of cost-effective data warehouses and marts. Jim is also a
Co-Founder of Data University www.datauniversity.org.