Predictive Stock Level Optimization

@rosscarter1011

Hello, Retail inventory optimisation is better treated as a forecasting and decision problem than simply fitting a time-series model. I would first establish SKU-level demand history, stockouts, promotions, lead times, and seasonality, because observed sales can understate true demand when products were unavailable. The modelling stage would compare appropriate forecasting approaches using time-based backtesting, then convert demand forecasts into reorder points and quantities using lead-time demand, variability, and the required service level. I would report forecast performance separately from inventory-policy performance, so improvements are measurable against the existing rule. The final package would include reproducible Python code, configuration for new data, and planner-friendly visual outputs. Have a nice day

@robertb297

Hello, Predictive retail inventory optimization fails when intermittent slow movers are forced through the same time-series model as fast movers. Fast moving items benefit from gradient boosted demand forecasting or LightGBM with lag features, while intermittent or lumpy SKUs require probabilistic Croston or negative binomial models to set correct safety stocks without ballooning carrying costs. I'd start by cleaning the raw transaction history, handling stockout censoring so suppressed demand during zero-inventory periods does not bias the forecasts downward, and categorizing SKUs by sales velocity. Next, I will build the hybrid pipeline combining lead-time demand distributions with dynamic reorder point calculations, validated through rolling back-tests. The final deliverable will be a single-command Python pipeline and a refreshed summary dashboard for your planners. Do your raw extracts track supplier lead time variance per SKU, or is lead time currently treated as a fixed constant? Share a small sample of your inventory tables and I will outline the data pipeline structure. Appreciate your time

@ahmaddar2970

We can turn your inventory history into a living replenishment engine focused on service level and cash efficiency. I would start by profiling the raw tables to identify missing values, duplicate SKUs, stock movement anomalies, and lead-time inconsistencies, then reshape the data into a clean SKU-date panel that can refresh automatically. For modeling, I would use a hybrid approach: demand forecasting at the SKU or segment level, combined with probabilistic safety-stock and reorder-point calculations. That allows the model to capture seasonality, promotions, supplier variability, and the asymmetry between fast movers and slow movers. I would validate the system with rolling back-tests, compare it against your current rule-of-thumb method, and quantify the trade-off between forecast accuracy, service level, and inventory holding cost. Deliverables would include production-ready code, a concise README, and a simple dashboard or visual report layer so planners can refresh outputs with one command. The end result would be a practical replenishment decision engine that recommends what to reorder, when to reorder, and how much to hold by SKU.