An important variable in retail future planning is forecasting client flow in stores. This research aims at introducing two Long Short-Term Memory network architectures for time series forecasting of client flow in retail stores. These models are allied with three main data preprocessing approaches: a data imputation method that standardizes store schedules; a harmonic regression method that captures and removes the seasonal and trend components of the time series and a sliding window sampling method to construct the network’s training phase. Results were not extensively optimized but the framework leaves an open door for further improvements.