The COVID-19 pandemic has highlighted the need for policymakers and destination managers to have access to up-to-date and frequent data to inform decisions. Despite advances in big data, applications of tourism demand forecasting in the literature have been slow to adapt to a data-rich world and have predominantly relied on lagged survey data analysed at quarterly or annual intervals. This research proposes a novel framework designed for hourly-level tourism demand forecasting across a tourism destination, leveraging spatiotemporal mobile phone data collected through a telecommunications network. This dataset enables the measurement of hourly tourism demand for distinct visitor segments, at a suburb level, over a four-year period. A collaborative research agreement with the City of Gold Coast has made this analysis possible, highlighting the potential for smart cities to harness big data for urban planning and tourism management. This study compares the forecasting performance of local statistical time series models and global neural network techniques, aiming to bridge the gap between sophisticated modelling approaches and their practical application, which is often overlooked. We discuss some practical forecasting challenges and data science considerations pertinent to high-frequency tourism data usage including addressing complex seasonality, multi-step forecasting, and performance monitoring and model updating to combat concept drift. This study highlights the role of data science in bridging the gap between traditional forecasting and the demands of the big data era in smart cities, appealing to forecasters, tourism academics, and industry professionals keen on timely and actionable insights.