AI Reordering and Inventory Intelligence for an Auto Parts Distributor

An auto parts distributor ordered from foreign suppliers on a cycle that took about a month end to end. The delay was not shipping. It was working out what to order, from whom and in what quantity across a catalogue of tens of thousands of parts, with supplier catalogues arriving in different formats and different part numbering. Slow moving stock was equally invisible: without a view of which parts had quietly stopped selling, working capital kept accumulating on the shelves unnoticed.

We built a reorder engine that reads the sales and stock export and produces a priced order proposal per supplier. It forecasts demand from each part's own weekly history rather than a flat average, then applies safety stock against a target service level, pack rounding, minimum order quantities, a stockout trim, a spike guard and a seasonality clamp. The buyer edits lead time and review period on the page and the whole proposal recomputes live. On top of it sits an analyst built on Claude Opus 5: it writes a year in review for each account, drafts the offer message to send them, and answers questions over the live catalogue through read only tools. Every tool the model can reach reads and none of them write, each caps the rows it returns, and each states where a figure was inferred rather than read directly.