Data Analytics · Read + Write

Forecast → replenishment

Automate Rep · High Reason · High

The problem

Buyers manually forecast and place replenishment orders weekly across thousands of SKUs. Stock-outs and over-stock both cost money.

The AI approach

Demand forecast per SKU per location → reorder-point calc with lead-time and safety-stock → auto-create POs for confident SKUs, route the rest to buyer.

The outcome

~−30% stock-outs

Stock-out rate down ~30%; over-stock down ~15%. Buyers focus on new launches and exceptions, not the long tail.

Try itInput → Process → Output

Input — history + stock + lead-time
SKU CL-117 · pipe-4"plant A · 80 wks history
avg/wk 412stock 580 (5d)lead 14d
SKU CL-204 · elbowplant B · stocky · sparse
avg/wk 14stock 38 (3w)lead 21d
Process — AI pipeline
1Forecast demand per SKU × locationReadPredictive
2Compute reorder pointRulesSymbolic
3Decide auto-PO vs review queueRulesSymbolic
4Generate PO drafts for buyerWriteGenerative
Output — PO recommendations
Click Run demo to forecast demand and propose reorder POs.
2 PO recommendations 1 auto · 1 review
SKU CL-117plant A · pipe-4"
+1,200 units
฿42,000
Auto-PO
SKU CL-204plant B · elbow · sparse demand
+50 units
฿1,900
Buyer review
CL-117: demand +14% MoM; stock cover <7 days; auto-PO of 1,200 covers 3 weeks. CL-204: sparse intermittent demand — Croston forecast unsure, route to buyer for judgement call.

Three AI types in this use case

SymbolicReorder formula (ROP = avg demand × lead time + safety stock); buyer approval thresholds; supplier MOQ rules.
PredictiveHierarchical time-series forecast (Prophet / DeepAR / Croston for intermittent); seasonality + promo lift.
GenerativeLLM writes the rationale on each suggested PO ('demand up 14% MoM, stock < 7-day cover').

The stack

  • Forecast · Prophet / DeepAR / Croston
  • Reorder solver · per-SKU ROP
  • ERP / supplier API · auto-PO
  • Buyer dashboard · review queue

When this works

  • ≥ 12 months of history per SKU × location
  • Lead times are stable and known
  • Safety-stock policies are explicit

When it fails

  • Long-tail SKUs with sparse sales — forecast is a guess
  • Promotion data missing → forecast over- or under-orders
  • Lead-time variability not modelled
  • Supplier price changes break the reorder math