In an embodiment, for each customer a pre-sale predictive model generates a value prediction signal indicative of potential value of a sales transaction for that customer. The predictive model can provide various types of value prediction signal including, for example: (a) buy-only signal, representative of the likelihood that the customer will accept the offer to purchase the product; (b) lapse-only signal representative of the likelihood that the customer will lapse in payments for the purchased product; (c) buy-don't-lapse signal representative of the likelihood that the customer will accept the offer to purchase the financial product and will not lapse in payments for the purchased product; as well as predictive models providing combinations of these signals.
Predictive models 410 effect a degree of feature selection. In various embodiments, predictive models 410 identify high importance features that have the most pronounced impact on predicted value. Different types of model may identify different features as most important. For example, a model based upon a buy-only signal may identify different leading features than a model based upon a lapse-only signal.