The traditional research approach has many commendable characteristics, most notably, the ability to build concepts inside mathematical models and present those concepts to virtual customers without the risk of a market pilot. This is one of the seminal achievements in marketing science in the last 30 years. Yet, even these advanced approaches have limitations. Specifically, unless very carefully controlled, simulation models often don't correlate very highly with stated and, by extension, actual behavior. A new product model is considered to work pretty well if it correlates to calibration concepts above an "r-squared" of 0.30. But, empirically, this low level of correlation is barely enough to believe the directionality of the model, let alone to use it for insights into what features to put into a new product. Additionally, current generation new product simulation models have other critical limitations:

• New product simulators are manual; they don't exhaustively find the combinations of features, servicing, pricing, and delivery that will maximize demand.

• New product simulators usually produce only an academic measure termed "share of preference." What is really needed is a sales model that projects unit sales or revenue opportunity.

• Most markets have different preference profiles or segments of preference. Traditional research does not accurately determine those buyer value segments, nor does it build those insights into the new product design simulator.

• The optimal amount of resources to spend on advertising vs. product design is not clear.