AI Inventory & Warehouse Management Agent

Description

Three warehouses, one WMS, and a stock accuracy figure that hovers around 94% — which sounds fine until you realise it means several hundred wrong lines at any moment, and our pickers have learned not to trust the system. We want an agent that watches transaction patterns and predicts which SKU-location combinations are likely to be wrong, so cycle counting goes where the errors are instead of running alphabetically. Then it should explain the prediction: "counted correct on the 3rd, 14 picks since, two short-picks reported" is actionable, a risk score is not. Also in scope: reorder suggestions for the 300 SKUs with seasonal demand, where our current fixed reorder points are visibly wrong twice a year.

• Model discrepancy likelihood per SKU-location from WMS transaction history
• Direct the cycle-count schedule from that model and measure the hit rate against the current approach
• Produce written explanations for each flagged location rather than an opaque score
• Build seasonal reorder suggestions for the 300 identified SKUs, as suggestions a planner approves
• Report stock accuracy movement per warehouse over the engagement

• Discrepancy prediction integrated with the cycle-count schedule
• Explanation output per flagged location
• Seasonal reorder suggestion queue with planner approval
• Accuracy improvement report per warehouse