Friday, November 17, 2006

Reasons Why we Model

What business problems need models?

The use of modeling generally falls into one of three categories.

1. Prediction -- the most common interpretation is about predicting the future, either in aggregate or at the individual customer level. Examples include sales forecasting, acquisition and propensity scores.

2. Classification -- modeling is also useful for categorizing people (or items) into buckets. The goal is to minimize the differences within a given group while maximizing the differences across groups. Good segmentation schemes rely on classification so that messages can be better targeted.

3. Estimation -- plugging holes, e.g. missing data, or calculating a new metric, e.g. Life Time Value, represents the third area of modeling. This differs from prediction in the sense that there is no time element involved.

Prediction and estimation often use a similar set of techniques, just different data and assumptions; classification problems are best solved with a different set. Thus, any good modeler will want to understand the business problem and objective first before reaching into his satchel to pull out a tool.

For another view, see Katie Cole's piece in MultiChannel Merchant.

No comments: