In stochastic simulation the input models used to drive the simulation are often estimated by collecting data from the real-world system. This can be an expensive and time consuming process so it would therefore be useful to have some guidance on how much data to collect for each input model. Estimating the input models via data introduces a source of variance in the simulation response known as input uncertainty. In this paper we propose a two stage algorithm that guides the initial data collection procedure for a simulation experiment that has a fixed data collection budget, with the objective of minimising input uncertainty in the simulation response. Results show that the algorithm is able to allocate data in a close to optimal manner and compared to two alternative data collection approaches returns a reduced level of input uncertainty.