Greetings. I was hoping someone could offer me some insight. I have a
transaction DB with 300K unique transactions and about 50K unique
customers. I'm trying to devise an index for ranking customers based on
three variables: recency of last visit, total purchases ($), and total
number of visists.

I've sorted by each variable and assigned a value of 1-5 at the natural
breaking points. The sum of the three values (between 3-15) indicate
the value of the customer. This is a very crude indexing method. Does
anyone have any more sophisicated suggestions?

Thanks,
Aaron

junkmail115 (AT) gmail (DOT) com wrote:
It sounds like you are looking for a value measure for the customers,
rather than a specific way of indexing them.

If this is the case, consider whether the three variables are additive
in the way you add them now. Maybe a customer with value 7 stemming from
total purchases = 5, last visit = 1 and total visits = 1 is worth more
than a customer with value 7 stemming from total purchases = 1, last
visit = 5 and total visits = 1. If this is the case, you have to
normalize the values to make them more realistic before adding them.

But maybe I didn't understand your question...?

Regards,

Michael.
--
Which is more dangerous? TV guided missiles or TV guided families?
I am less likely to answer usenet postings by anonymous authors.
Visit my home page at http://michael.zedeler.dk/

Michael,

Thanks for your response. Great point! I will definitely look into
either giving different weight to the three values or normalizing them
before adding them together.

Thanks,
Aaron

junkmail115 (AT) gmail (DOT) com wrote:
Its probably overkill, but if you want an accurate normalization map,
you can do it using statistics, having about one or two unknown model
parameters for each customer attribute and the gross profit of each
customer as (possibly only) variable. The advantage is that the
resulting valuation function will be very accurate wrt. predicting the
value of a customer, but it may be complicated to implement and maintain.

If you decide to proceed with something like that, start a thread in one
of the sci.math or alt.sci.math-groups (and cc me :-).

Regards,

Michael.
--
Which is more dangerous? TV guided missiles or TV guided families?
I am less likely to answer usenet postings by anonymous authors.
Visit my home page at http://michael.zedeler.dk/