We present a new method for determining the Galactic gravitational potential based on forward modeling of tidal stellar streams. We use this method to test the performance of smooth and static analytic potentials in representing realistic dark matter halos, which have substructure and are continually evolving by accretion. Our FAST-FORWARD method uses a Markov Chain Monte Carlo algorithm to compare, in 6D phase space, an "observed" stream to models created in trial analytic potentials. We analyze a large sample of streams evolved in the Via Lactea II (VL2) simulation, which represents a realistic Galactic halo potential. The recovered potential parameters are in agreement with the best fit to the global, present-day VL2 potential. However, merely assuming an analytic potential limits the dark matter halo mass measurement to an accuracy of 5 to 20%, depending on the choice of analytic parametrization. Collectively, mass estimates using streams from our sample reach this fundamental limit, but individually they can be highly biased. Individual streams can both under-and overestimate the mass, and the bias is progressively worse for those with smaller perigalacticons, motivating the search for tidal streams at galactocentric distances larger than 70 kpc. We estimate that the assumption of a static and smooth dark matter potential in modeling of the GD-1 and Pal5-like streams introduces an error of up to 50% in the Milky Way mass estimates.