Predictive analytics · Multi-branch retail
Situation
Jubba Retail Group operates dozens of high-footfall stores in urban markets. Planners were balancing two painful poles: over-ordering (cash tied in slow movers) and stock-outs on hero SKUs in peak windows (Eid, school terms, and supplier-led promos). Their ERP’s static min/max rules were not using store-level micro-trends or multi-week promotional tails.
What we did
NeuralFaaruuq delivered weekly, store-by-SKU forecasts with explicit calendar features (holidays, local events) and a blended stack of strong baselines plus ML lifts on volatile categories. The buying team approves system-suggested order quantities; exceptions are flagged for categories with data sparsity or new SKUs. Integration exports into the existing purchase order workflow to avoid a rip-and-replace of ERP.
- Training back-tests per region to set confidence bands the merchants trust
- “What-if” views for promos: attach uplift curves from past campaigns or supplier sheets
- On-shelf availability KPIs wired into a simple weekly review deck
Outcomes (post go-live, annualized where noted)
- ~22% fewer in-stock miss incidents on targeted categories vs. the prior year control cohort
- Meaningful reduction in excess safety stock without hurting peak-week sell-through (exact inventory % under commercial NDA)
Tech stack (high level)
Parquet/warehouse extracts, feature pipelines in a managed job runner, model registry with rollback, and BI surfaces the category managers already use.
