The ubiquitous David Cambria said this during a SuperConference panel and I got to thinking what he means. Partly he means that if you look back at numbers, that is more limiting than when you project numbers forward. But projection (modeling) does rely on historical numbers. The distinction blurs to the degree your past metrics were necessary to enable you to calculate the formulas and identify the future trends that the model does (See my post of Jan. 19, 2011: clarify patterns and formulas from data with the spreadsheet function of trend lines.). The assumptions the modeler builds in makes all the difference in the accuracy of a model, so to that extent you are swapping one form of uncertainty (future data) for another (premises of the model).
Some models used by law departments include Monte Carlo simulations (See my post of May 15, 2005: Monte Carlo simulations as computational models; and June 26, 2009: simulations and models.) and litigation risk analysis (See my post of June 17, 2009: decision tree software with 6 references.).