The ABCs of database marketing Jul 1, 1998 12:00 PM
, Sherry Chiger
JobZone
Search and post jobs for the Multichannel Merchant. Including jobs for brand & agency marketers, e-commerce, catalog marketers, ops & fulfillment, direct marketing and more.
A full 51% of the overall respondents to the 1998 Catalog Age Benchmark
Report on Lists and Databases (see p. 63) don't perform any sort of
database modeling. And among respondents with annual sales of less than $10
million, the figure climbs to 68%.
One reason so few catalogers take advantage of their databases could be
that they're confused about their options. All around them, database
experts are proselytizing about the benefits of predictive modeling and the
wonders of RFMP without defining the terms. So to get you started on
exploring the possibilities of database marketing, Catalog Age is providing
a primer of terms.
STATISTICAL RESPONSE MODELING
Also known as predictive modeling, this method of analysis enables you to
predict a response by looking at a number of factors together. "Using the
variables simultaneously allows you to pick up the true value of a factor
in context," says Dan Steinberg, president of database firm Salford
Systems. For instance, if you rate your customers solely by age, you could
find that your older customers are your best customers. But if you add
another variable, such as household income, you may discover that most of
your older customers are also your wealthiest-but that these wealthy older
buyers actually spend less than younger buyers who earn the same amount of
money.
PROFILING
If statistical modeling is predictive, profiling is descriptive-it's an
umbrella phrase for any of several methods of analyzing your database so
that you can describe the characteristics of buyer segments. For instance,
profiling could show you that your buyers with the greatest lifetime value
are semi-industrial suppliers with 50-75 employees located in midsize
towns, or that your most responsive customers are single women ages 25-34
with an average income of $25,000 who buy primarily around the holidays.
RFM An acronym for recency, frequency, monetary value, RFM is "a low-cost,
effective way to get into modeling without a lot of expertise and
statistical background," says Eric Ruf, vice president of database
marketing consultancy Ruf Strategic Solutions. With RFM, you score
customers in terms of how recently they bought from you, how frequently
they've bought from you in a given time period, and how much they spent
with you in that same period. A simplified example:
RECENCY
customers who bought in the past month = 3 points
customers who bought in the past six months = 2 points
customers who haven't bought in at least six months = 1 point
FREQUENCY
customers who made at least three purchases in the past year = 3 points
customers who made one or two purchases in the past year = 2 points
customers who didn't make a purchase in the past year = 1 point
MONETARY VALUE
customers who spent more than $300 in the past year = 3 points
customers who spent $100-$299 in the past year = 2 points
customers who spent less than $100 in the past year = 1 point
Your best customers would score a total of 9 points; your least productive
customers would score just 3 points. Then when you planned your next
mailing, you could decide if you wanted to mail to just your 20,000
highest-scoring customers. Or for certain specials, you might decide to
focus more on bigger spenders, even if they didn't score so well on
frequency.
Take RFM a step further, and you have RFMP: recency, frequency, monetary
value, product. If you sell several categories of product or mail several
titles, adding the product variable to your RFM scores can prevent you
from, say, mailing an offer of PC software to Macintosh users.
REGRESSION ANALYSIS
Like RFM, this modeling method uses scoring to predict results, but it
takes into account more historical and demographic variables and relies on
statistical calculations to weigh the variables. Certain standard
industrial classifications (SICs), for instance, might be assigned a higher
value than others; household income might be factored by a number so that
wealthier buyers are ranked as proportionately more valuable than
less-wealthy customers.
TREE ANALYSIS
Using flow charts known as decision trees, this method breaks down your
database by variables. Let's say you want to determine the characteristics
of your best buyers. First, you might sort your house file by age groups;
then you might want to sort those age groups by whether they own their
homes; then you could separate those who have children from those who
don't. A very simplified version of the resulting decision tree could look
like this:
HOUSE FILE
You could then compare response among each of these categories, or
branches, to determine which group-homeowners under the age of 40 with
children, perhaps, or nonhomeowners at least 40 years old without
children-has the highest lifetime value (a descriptive function), or which
group is most likely to respond best to your next offer (a predictive
function). CHAID (Chi-Square Automatic Interaction Detector) and CART
(Classification and Regression Trees) are two of the more common types of
tree analysis.
CLUSTERING
This method of profiling is "very good for taking the mass of information
you may have about customers," Ruf says, "and grouping market segments
together that you can then develop strategies around." You can cluster
customers by age group, SIC, income or sales level, geography, and other
variables to create a demographic or psychographic profile of your most
sizable segments of customers. "Ideally you want to market to each customer
individually, but the economics aren't there," Ruf says. "Clustering allows
you to segment your market so that you can treat each segment as a distinct
audience."
