AI Framework for Multi-Task Forecasting in Small and Medium Enterprises

This paper presents a unified artificial intelligence platform designed for forecasting sales, customer traffic, and operational loads in small and medium-sized enterprises (SMEs). SMEs often face fragmented data environments and limited computing resources, which restrict the adoption of advanced predictive analytics. The proposed system integrates a unified feature and data layer with a multi-task temporal forecasting architecture that enables cross-task representation learning. The platform incorporates model compression techniques such as pruning, quantization, and knowledge distillation to allow deployment on resource-constrained infrastructure commonly used by SMEs. Additionally, drift detection and incremental learning mechanisms are introduced to maintain robustness under distributional changes. Empirical evaluation is conducted using two years of operational data collected from retail, e-commerce, restaurant, and SaaS firms in Los Angeles County, USA. Compared with classical statistical models including ARIMA and Prophet, as well as machine learning baselines such as gradient boosting and single-task deep learning models, the proposed multi-task framework achieves a 15–25% reduction in prediction error across sales, traffic, and operational load forecasting tasks. Case studies demonstrate practical operational benefits including reduced stockout rates, improved inventory turnover, optimized cloud resource allocation, and enhanced labor utilization. The results indicate that unified, resource-efficient AI platforms can significantly enhance predictive capabilities and operational efficiency for SMEs operating under limited infrastructure conditions