Should-cost,
while it can still change.

AI should-cost analysis means estimating what each part in a bill of materials ought to cost from its geometry, material, process and volume, using your own past quotes as the reference, so a design team sees cost as it designs rather than when supplier quotes return.

Classical should-cost models are built bottom-up: cycle time, machine rate, material mass, scrap, overhead. They are accurate and slow to set up, so most teams keep one for a handful of critical parts. The quote history most companies already hold, years of RFQs answered by real suppliers, is a second source that rarely gets used because it sits in emails and spreadsheets with inconsistent part descriptions.

We combine the two. The bottom-up model handles the physics of each process; the quote history calibrates it to the suppliers and volumes you actually buy at; and the model reads the CAD and BOM so nobody types a part in twice. Every estimate shows the drivers behind it, because a number nobody can explain will not survive its first supplier negotiation.

We hold back a set of recent parts that were quoted and bought, estimate them blind, and compare. The useful measure is not a single average error but where the estimate is wrong and why: one process consistently low, one supplier region consistently high. Those patterns tell you whether the model can be trusted for a given commodity, and where it should be marked as uncertain.

The BOM comes from your PLM or ERP (Windchill, Teamcenter, Arena, SAP, Odoo), and the quote history from wherever it lives today, often a mailbox. Commodity indices for resin and metal prices are read in where you subscribe to them, and the estimate records which index date it used. Because a side action or an extra set-up changes cost, the same features found by DFM checks feed this model.