Accurate sales forecasting is essential for retail Micro, Small, and Medium Enterprises (MSMEs) to optimize operations and inventory planning in the digital economy. This study compares the forecasting accuracy between Artificial Intelligence (AI)-based methods (Random Forest, Decision Tree) and traditional techniques (Moving Average, Exponential Smoothing) using 3,600 transaction records from five retail MSMEs over three months. A quantitative experimental approach was employed to evaluate model performance under real-world conditions, including market fluctuations and seasonal anomalies. Evaluation metrics include Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and cross-validation techniques. The findings indicate that the Random Forest model achieves superior accuracy (MAPE = 8.5%) compared to traditional methods (MAPE = 15.2%). Explainable AI (XAI) using SHAP and LIME further enhances transparency and managerial trust. Although traditional methods offer faster computation and ease of interpretation, AI-based models show resilience against unpredictable sales patterns. This research recommends hybrid adoption strategies that balance predictive power with interpretability for MSMEs with limited technical capacity. The results contribute to the discourse on digital transformation and intelligent forecasting in the MSME sectors.