 |
Click for a Printer-friendly Version - Adobe PDF
Multiple Analytical Tools for Optimal Target Marketing
By Jim Wheaton
Principal, Wheaton Group
Original version of an article that appeared in the
February 1999 issue of "Catalog Age"
The key to cutting-edge target marketing lies with the integration of two data
mining initiatives:
-
The determination of whom to promote, via statistics-based predictive
models. This is a binary - "yes/no" - process involving the estimation of
future purchase activity.
-
Insight into what to promote, using four quantitative techniques: tree
analysis, demographic profiling, focus groups, and survey research.Few
catalogers have attained this sophisticated level of target marketing.
Instead, they are mired in the traditional world of RFM Cells.
This article will provide a framework for integrating multiple analytical tools
into a sophisticated program of differential catalog marketing. It will
focus on a mid-sized cataloger with merchandise geared mostly to women;
actually, a composite of several current and past clients.
RFM Cells - Original Customer Classification System
According
to Catalog Age's 1998 Benchmark Report on Lists and Databases, over
one-half of catalogers still rely on RFM Cells to make promotional
decisions. Until recently, the same was true of our mid-sized
cataloger. Accordingly, just three characteristics were used
to classify customers:
-
Recency, or the number of months since the most recent
paid order. Each customer was placed in one of four possible monthly
categories: 0-6, 7-12, 13-24 or 25+ months since the last paid order.
-
Frequency, or the total number of historical orders.
Customers were classified by 1, 2 or 3+ previous paid orders.
-
Monetary, or historical Average Order Size. Our
cataloger created three equally-sized groups - "Low," "Mid," and "High."
The result was 36 (4 x 3 x 3) Recency/Frequency/Monetary permutations, or RFM
Cells.
RFM Cells - Determining Whom to Promote
Let's examine
how our cataloger used these 36 RFM Cells to determine whom to promote.
First, some assumptions:
-
The cataloger is willing to mail to breakeven, and no further. Breakeven
was - and is - $1.25 per piece mailed.
-
Generally, a re-mail is done several weeks after the initial drop, with a
response rate that is about one-half of the first.
To determine which customers to promote, the cataloger examined historical
results in order to develop by-Cell estimates of future purchase volume.
Then, it created a hierarchy of cells, from highest to lowest estimated
performance. Those cells that were expected to do at least $1.25 per-
piece-mailed were contacted. Likewise, those at $2.50 or more on the
initial drop were mailed a second time.
RFM Cells - A Crude Technique for Selection
Consider
the Frequency input to our cataloger's 36-Cell segmentation strategy.
Customers who have purchased 10 times, for example, have been categorized
identically to those who purchased three times. In reality,
they have been much more loyal and should have been designated as
such. However, significantly increasing the number of categories
to accommodate these differences reduces each Cell's sample size
to such an extent that it becomes difficult to determine the RFM
hierarchy.
This problem is compounded exponentially if we include into the RFM Cells other
key database fields such as:
-
Merchandise categories.
-
Satisfaction indicators such as returns, exchanges and allowances.
-
Service measures such as backorders and out-of-stocks.
-
Payment indicators such as cash versus credit, and phone versus mail.
-
Overlay demographics such as age, income, marital status and presence of
children.
Clearly, an RFM strategy that took full advantage of our cataloger's
comprehensive customer database would require thousands - and perhaps tens of
thousands - of Cells. This would be an unwieldy strategy to implement,
and one with many opportunities for error. And, Cell sizes would be far
too small to create a stable RFM hierarchy.
Statistics-Based Predictive Models - A Selection Tool for
Today's Sophisticated Databases
Statistics-based predictive
models do not have these limitations. All database fields
with the potential to separate future buyers from non-buyers can
be input to the modeling process. There are none of the sample-size
issues that are inherent in RFM Cells. And, the result of
the model - a rank-ordering of customers by predicted future purchase
volume - results in a straightforward implementation. All
customers above a predetermined predicted performance are promoted,
and the balance are not.
In short, predictive models are more stable than RFM Cells. They're
easier to implement. And, they do a substantially better job of
determining future purchase behavior.
The Superiority of Statistics-Based Predictive Models -
An Example
Table 1 shows the performance of a predictive
model that replaced our mid-sized cataloger's traditional RFM segmentation
strategy. Recall that our cataloger promotes to $1.25 per
piece mailed, and experiences a 50% fall off with re-mails.
All customers have been:
-
Scored by the model.
-
Rank-ordered from highest to lowest predicted performance.
-
Grouped into ten equal performance segments ("Deciles").
Table 1:
Customer Predictive Model
Customers in Decile 1 generate $8.14 per piece mailed, while Decile 10 brings
in just $0.44. Combined, all of the Deciles average $2.49. The Lift
column shows the performance for each Decile compared with this $2.49 average,
while the last column records the cumulative ("running") lift. For
example, Deciles 1 through 5 have a cumulative lift of 158. This means
that, if the cataloger limited a promotion to just this group, the results
would be 58% better than the $2.49 average for the entire file.
Table 2 illustrates the incremental power of the predictive model versus the
traditional RFM Cells:
Table 2
Customer Predictive Model
-
Deciles 8 through 10 all are below our $1.25 breakeven, and should be
eliminated from future promotions. Our cataloger's original RFM Cells, on the
other hand, were only able to identify about 10% of the customer base as being
unprofitable. With the statistics-based predictive model, the money that
is freed up by eliminating additional unprofitable circulation can be allocated
to profitable re-mails, as well as to innovative targeted marketing programs
(more on this shortly).
-
Deciles 1 through 3 can be re-mailed. Even with a 50% performance
drop-off, dollars per piece mailed will still be over $1.25. RFM Cells
were only able to identify about 15% of the database as being re-mailable.
-
Decile 1 performs so well that it can be re-mailed twice. That's because
the resulting performance - even after a 75% drop-off - remains comfortably
above the $1.25 breakeven. The RFM Cells were unable to identify any
customers who profitably could be re-mailed twice.
Additional Analytical Tools - Determining What to Promote
We've
seen how statistics-based predictive models are superior to RFM
Cells for determining whom to promote. However, there's much
more to state-of-the-art database marketing than optimally identifying
the mailable portion of customers. We also need to determine
what to promote, via customized catalogs. And, to do so requires
the use of additional analytical tools. This is because statistics-based
predictive models, for all of their power, create what is known
as heterogeneous segments; that is, segments containing individuals
with multiple characteristics.
Decile 1 of our mid-size cataloger's predictive model, for example, contains
men and women, as well as the old and the young. In addition, it includes
customers who have purchased many different combinations of merchandise.
In fact, these customers have just one guaranteed similarity: they are
expected to order an impressive amount of merchandise in the future.
Tree Analysis - Identifying Multiple Types of Customers
Tree
analysis, on the other hand, creates homogeneous segments; that
is, each segment contains individuals with identical merchandise-purchase
and/or demographic characteristics. The following are two
hypothetical tree analysis segments:
-
40-50 year old female jewelry buyers, with four or more purchases, averaging
$500+, and at least one purchase within the past six months.
-
30-35 year old male electronics buyers, with three or more purchases, and at
least one within the past twelve months.The beauty of tree segments is that
they provide us with marketing insight. Feed these descriptions to a good
creative staff and the result will be a brainstorming session on how to tailor
the catalog to the demographic characteristics and product interests of
multiple customer groups.
There are many ways to tailor a catalog. Perhaps the simplest are ink-jet
messaging and "blow-ins." More involved techniques are over-wraps,
customized covers, and "glue-ins." The ones that require the most
up-front investment are selective binding and specialty books.
Tree Analysis - An Example
In order to identify
multiple types of customers, our cataloger ran a tree analysis off
database fields limited to merchandise categories and demographics.
These are the most likely to offer clues on how to match the content
of a catalog to a given customer's needs. Other variables
- recency, frequency, average order size, and the like - are helpful
for predicting future purchase behavior. However, they provide
little insight into customer lifestyles or interests.
The results our cataloger's tree analysis are illustrated in Table 3. We
see that female customers generate sales-per-piece-mailed of $2.88, over
twice that of the males' $1.42. However, the analysis reveals a small
pocket of male customers who, at $2.95 per piece mailed, are actually more
responsive than the average female. These are buyers of jewelry
merchandise.
Our cataloger's marketing staff, intrigued by this finding, set out to uncover
additional clues about the purchase dynamics of this profitable customer
segment. Accordingly, an analysis was performed to determine the types of
jewelry merchandise being purchased. In almost all cases, it was
merchandise targeted for females. Two plausible hypotheses were
developed:
-
The actual purchasers are women who are using credit cards that are in the
names of their husbands.
-
They are men purchasing gifts for the significant women in their lives.
Profiling - Developing Portraits of Your Customers
Our
cataloger's next investigative step was to send its customer database
to an overlay company for a demographic and psychographic enhancement.
As a result, scores of fields were added to each customer's record,
including age, income, marital status, and presence of children.
The results suggested that these jewelry buying households - regardless of
whether the purchase is being made by a card-toting wife or a gift-giving
husband - are leading "Ozzie & Harriet" lifestyles. These are
families with children, living in single family suburban homes, with
professional, technical and managerial occupations.
Knowing that these are married suburbanites rather than single city-dwellers
will be helpful in tailoring future catalog copy. Nevertheless, our
cataloger still didn't know if the purchaser is the husband or the wife.
To gain a definitive answer, it commissioned a series of focus groups as well
as some comprehensive survey research.
Focus Group and Survey Research - Developing Psychological
Insight into Jewelry Customers
The focus group and survey
research determined that the majority of these individuals are gift-giving
husbands. They are what the research company dubbed "unimaginative
male gift givers." These are men who - despite their solid
professional success - dread buying birthday, anniversary and holiday
gifts for their spouses. They very simply are at a loss for
what kinds of presents their wives might find appealing.
In order to fully leverage these findings, our cataloger convened a task
force. Comprised of representatives from Marketing, Creative, and
Analytical, the task force's mandate was to develop a loyalty program to appeal
to these "unimaginative male gift givers." Based upon the insight gained
from the research, the loyalty program was built around the following key
features:
-
Automatic reminders for upcoming birthdays, anniversaries, and other personal
milestones. This is a registry based on self-reported information, so
that our cataloger can remind its participating male customers of upcoming
special events that will require a gift. Based on customer preferences,
these reminders can be sent via phone or e-mail.
-
A consulting service for gift selection. The cataloger staffed its
in-bound call center with specially trained "gift consultants." This
service was tied to the existing database, to ensure that duplicate gifts are
not recommended.
-
All gifts are wrapped and then mailed to either the male program participant or
- if he's out of town - his wife. Likewise, a card is included, either
blank or fully addressed.On the prospecting side, the cataloger's circulation
department began working with its list broker to identify new male-oriented
lists to target prospecting catalogs. These catalogs include a
description of the loyalty program as well as a form for signing up.
Summary
Our cataloger had for many years used traditional
RFM Cells to determine whom to promote. However, it recognized
that statistics-based predictive models do a substantially better
job of determining future purchase behavior - as well as being much
easier to implement. As a result, it was able to optimize
its promotional strategy, in terms of both initial and follow up
mailings.
The cataloger also realized it is critical to develop deep insight into what to
promote, and to establish specialized, targeted marketing initiatives such as
loyalty programs. Accordingly, it employed four quantitative techniques -
tree analysis, demographic profiling, focus groups, and survey research - to
identify, cultivate, and grow a small but profitable group of male customers.
Over time, our cataloger - by optimizing both whom and what to promote - was
able to drive significant increases in its revenues as well as profits.
Jim Wheaton is a Principal at Wheaton Group, and can be reached at 919-969-8859
or jim.wheaton@wheatongroup.com. 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.
Top >>
|
|
