In aerospace manufacturing, sourcing engineers receive supplier quotes and are expected to evaluate them. But evaluation without an independent cost estimate is not evaluation; it is acceptance. The sourcing professional has no leverage, no floor to negotiate from, and no way to explain to leadership why a quote is too high or surprisingly low.

Commercially available should-cost tools exist (aPriori, Costimator, SEER-MFG) but they require engineering data, trained analysts, and enterprise licensing. They are not accessible to a sourcing team making fast portfolio decisions across dozens of parts.

Feature-based costing tools like aPriori read CAD files and count holes, pockets, and tolerances. That precision is valuable, but it requires tooling that most sourcing teams don't have at the point of negotiation. The vast majority of sourcing decisions are made from a drawing title, a material, and a complexity characterization, not a fully instrumented CAM file.

A parametric model trades geometric precision for speed and accessibility. Given material, part weight, complexity tier, tolerance tier, volume, and region, the model produces a defensible price band: not a single number, but a structured range that tells the sourcing engineer where the floor is, where the ceiling is, and which cost driver dominates the estimate.

The model doesn't replace aPriori. It fills the gap between "I have no idea" and "I have a routing." That gap is where most sourcing conversations happen.

The engine is built in Python with a strict separation of concerns: the cost engine produces all numbers deterministically, a separate AI layer generates plain-language narrative from those numbers, and an Excel exporter surfaces everything in a format a sourcing team can actually use in a meeting. Every number traces back to an assumption. No black boxes.

The engine is fully modular. Data files (material prices, machine rates, buy-to-fly ratios, outside process costs) live in CSV and JSON, editable without touching code. All assumptions are surfaced in the Excel output so every number is auditable.

V2 added STEP geometry ingestion via cadquery, auto-populating bounding box, volume, and buy-to-fly ratio from uploaded part files, and a Streamlit UI wrapper that puts the full engine in a browser without requiring Python.

The model produces three scenarios simultaneously: Low (optimistic assumptions: best-case material pricing, lowest machine rates, tightest margin), Mid (baseline assumptions, the primary reference point), and High (conservative assumptions: worst-case on all four variable inputs). The band is not a confidence interval in the statistical sense. It is a structured range that reflects real variance in the four inputs that move most: material price, machine rate, scrap rate, and supplier margin.

In leadership reviews, the instinct is to anchor on the mid number. The right framing is different: the low is the floor to negotiate toward, the high is the ceiling that justifies walking away, and the mid is the starting position for a supplier conversation. The band is the tool. The mid is a reference point, not a target.

For every estimate, the engine perturbs four key variables by +10% and measures the resulting change in the mid unit price. The top three drivers are returned in rank order. This answers the most important pre-negotiation question: where is the cost actually coming from?

For machined parts, machine rate and machining hours dominate almost universally. Material price becomes a significant driver only when buy-to-fly is high or material is inherently expensive (Ti, Inconel). Knowing this in advance changes the negotiation strategy.

Every estimate carries a confidence rating (High, Medium, or Low) assigned by a demerit system that evaluates input quality and inherent model uncertainty. The score is deterministic and fully traceable. It is not a gut check; it is a structured signal about where the estimate sits on the spectrum from "parametric baseline" to "specific enough to negotiate from."

Override semantics are explicit: an override tagged as Verified (sourced from a routing, CAM output, or supplier process sheet) carries no confidence penalty; it improves specificity. An override tagged as Unverified (a user estimate) reduces confidence by one demerit per field. The distinction prevents the confidence system from penalizing good data.

The following outputs were generated from the live engine using the synthetic dataset. All part numbers, programs, and values are illustrative. Region adjusts machine rates, overhead rates, and margin assumptions. Geometry and hours are held constant across regions.

Key observation: machining cost accounts for 49% of unit price on a Ti-6Al-4V Tier 4 part. Machine rate and machining hours are nearly equal sensitivity drivers (+$172 vs +$172 on a +10% move). Material price, despite titanium's cost, contributes only +$29. The negotiation lever is cycle time and supplier efficiency, not material markup. This is the insight a should-cost model surfaces before the first quote is received.

A should-cost model that doesn't acknowledge its own boundaries is more dangerous than no model at all. The following limitations are deliberate scope decisions, not oversights. Each has a corresponding V2 path if the use case demands it.

The Should-Cost Model is the technical centerpiece of a five-tool sourcing operations suite. Each tool is standalone, solves a specific operational problem, and produces outputs defensible in a leadership setting. The suite is designed for the sourcing professional who needs to present structured decisions to leadership, not for the analyst building dashboards.