Abstract
Retail small and medium enterprises (SMEs) face persistent challenges in balancing inventory availability against excessive holding costs. Forecasting errors directly translate into stockouts, lost sales, and inefficient capital utilization. This paper presents an end‑to‑end, SME‑oriented framework that integrates demand forecasting with inventory control and cost‑based performance evaluation. Using five years of daily sales data from the Kaggle Store Item Demand dataset, we compare simple heuristic baselines, classical exponential smoothing, and machine learning (ML) forecasting models. Forecast accuracy is evaluated using time‑series cross‑validation, and forecasts are converted into safety stock and reorder point decisions under a continuous review policy. A daily inventory simulation incorporating holding and stockout costs is used to quantify business impact. Results demonstrate that ML‑based forecasting consistently improves service levels and reduces total inventory‑related costs compared to traditional moving‑average heuristics, highlighting the practical value of data‑driven planning for SMEs.