Re: Market Segmentation Analysis Applied To The Online World

From: Jim Novo <jimnovo_at_sprintmail.com>
Date: Tue 7 Dec 1999 15:11:51 -0500

BRYCE ISHAM SAID:
> Market segmentation analysis involves classifying
> customers into different groups (often called
> clusters). Each cluster has distinctly different
> demographic and lifestyle characteristics. Cluster
> codes can be assigned to any customer file with
> addresses or at least zip codes.
>
> The result is that marketers can determine who their
> customers are, what they are like, which customer
> groups are their best customers, what media mix and
> message to use to reach more people like their best
> customers, and where the people similar to these
> customers are located. In short -- actionable
> information for targeted marketing.

Since Bryce works for Claritas, one of the providers of
this "clustering" information (PRIZM), I'd expect him
to be hopeful his products would be of some help to
online marketers.

Clustering was a hot idea in the 70's and early 80's
because many firms had limited customer data. Just
plug is a zip code, and Wham! instant customer profile.
Pretty neat. Good stuff for analyzing potential sites
for a retail store, profiling sales territories, etc.
For these applications, it works very well.


It is important to understand these clusters are much
more accurate in the aggregate than on an individual
basis. If you have a very large customer base, or are
profiling all the households in a particular zip code,
clusters can give you a good idea of what you have
demographically.

But as soon as you are talking about 1 or 2 customers
in a whole zip or tract, the chances of that single
person deviating substantially from the profile
increase dramatically. Go to
http://yawyl.claritas.com/ and try it yourself. Enter
your zip code and check out the profiles, thinking
about who lives in your neighborhood. There's a good
chance the clusters will portray much of the
neighborhood, or the *average* household, but I'm sure
you can think of households (probably yours) outside
the profile.

I suppose for ad supported sites, you might get some
interesting info to feed the agencies and the ad sales
force. If you have a large customer base, you might
gain some insight into average demos for your file.
But when you try to bring it down to the individual
level, be careful about using those profiles to
"personalize" anything.

Some of the original cluster approach has been modified
by using address level matching for the addition of
"harder" financial data, car ownership, etc. at the
household level, but this info is available without
clustering as long as you have a physical address.
Presumably, that's what DoubleClick is doing with the
Abacus database.

You're much better off looking at real behavior
(purchases, web logs, response to e-mails, etc.) to
glean actionable profiles, if you are a direct
marketing oriented site. All the best personalization
I know of is driven off real behavior, not
self-reported or appended information.

"Market Segmentation Analysis", as Bryce called it,
does not have to rely on demographics or the physical
definition of a "market." Most e-commerce sites would
find segmentation by recency and frequency of purchase
a far more effective tool.

But if you're an ad supported publisher who wants to
have a feel for who is using the site in the aggregate,
and you have physical addresses, by all means, cluster
away!


Jim Novo
Marketing Mercenary
jimnovo_at_sprintmail.com




Received on Tue Dec 07 1999 - 14:11:51 CST