FlowIQ — Smart Inventory & Supply Chain AI

A regional grocery chain with 38 stores was over-ordering perishables (creating waste) while simultaneously stocking out of high-velocity packaged goods on weekends. Their planning team worked from spreadsheets updated once a week. FlowIQ moved them to daily ML-driven replenishment orders with automatic supplier API integration — halving waste and eliminating empty-shelf incidents.

• 1

  Perishable items have demand patterns driven by day-of-week, weather, local events, and promotions — linear models failed completely.

• 2

  38 stores × 4,200 SKUs = 159,600 individual forecast series to run daily within a 2-hour overnight window.

• 3

  Supplier lead times were inconsistently documented and frequently changed — the system needed to learn supplier reliability dynamically.

• 4

  Store managers distrusted 'black box' AI orders; adoption required transparent, editable recommendations — not mandates.

We built a hierarchical forecasting system. At the top level, a global XGBoost model captures chain-wide trend signals. At the store × SKU level, we use an ensemble of Meta Prophet (for seasonality and holiday effects) and a lightweight LSTM (for promotion-driven spikes). External signals — weather API, local event calendars, and commodity price indices — are fused as exogenous features. The optimisation layer converts forecasts into replenishment orders using safety-stock formulas that account for supplier reliability scores (learned from 18 months of delivery variance data). The UI shows managers a ± confidence band and lets them override with a reason code — overrides feed back into model retraining.