Mission:
Find best way to find specific customer that are interested in a particular
product.Challenge:
Client had over-worked client list of millions of persons.
Solution:
Modeled sales trials on a small sample then "scored" remaining clients to
estimate purchase likelihood. Customer saved
32% on
marketing costs while increasing sales by 24%
Take Action
 |
|
|
Objective
A large insurance company needed to find a way to determine which
customers, of millions, were interested in the likelihood of
prospective customers to purchase a specific insurance product
so that they did not need to market (call) the whole massive list, but yet
capture the bulk of the sales opportunity.
The Methods
The customer marketed to a sample group. Multiple types of data
were gathered so as to determine which information is useful, to what degree
and decide on a model to use for scoring a forth-coming file. This scored
file will be marketed to evaluate the final effectiveness of the model(s).
300,000 records and 1335 data columns, including whether they bought, was sent BioComp for analysis. An optimized model was constructed using BioComp's non-linear
adaptive modeling technologies,
iUnderstand, and in a
matter of hours a sample (held back) was scored and sent to the customer in
a blind test (by the way... we were told that SAS was also working on
the same problem but it took SAS "engineers" months to write and test
model-scripts trying to find good logistic models, where with
iUnderstand, it merely
takes minutes and there are NO scripts to write).
Results
It was found that out of the 1335 variables
provided, only 49 were needed to provide good scoring on the data provided.
The "lift curve" of the scoring model, predicted vs actual buyers, is shown
on the right, when the results are deciled. The customer determined
that they saved
32% on marketing costs while increasing sales by 24% |
|
