The rapid growth of e-commerce in emerging economies presents new opportunities for Micro, Small, and Medium Enterprises (MSMEs). However, the main problem lies in the difficulty of accurately predicting online sales across regions with heterogeneous socioeconomic and infrastructural conditions, which often leads to inefficient resource allocation and missed market potential. This study aims to develop a location-aware predictive framework that integrates spatial intelligence into machine learning models for forecasting MSME online sales in Indonesia. The proposed model adopts a two-stage approach that combines XGBoost regression with spatial lag features, allowing the model to capture both local demand drivers and inter-regional dependencies. The datasets include historical e-commerce transactions, demographic indicators, infrastructure accessibility, and socioeconomic profiles aggregated at the regional level. To ensure robustness, spatial-temporal cross-validation is applied, and model performance is evaluated using RMSE, MAE, and MAPE. The results show that the location-aware model outperforms baseline approaches, reducing forecasting errors by up to 18% and identifying high-potential sales regions more effectively. Explainability analysis further highlights population density, regional income, and proximity to logistics hubs as key predictors. Future work will focus on extending the framework with deep learning and graphbased models to capture dynamic spatio-temporal interactions, as well as integrating real-time data streams for adaptive sales forecasting.