Geometric should-cost: estimating manufacturing cost from CAD

A supplier sends a quote for a machined housing. Is it fair? A design engineer adds a feature to a part. What did it just do to the cost? A procurement team is negotiating a hundred line items. Which ones are overpriced? For most of manufacturing history, answering these questions meant either waiting for quotes to come back or relying on an estimator’s experience. Should-cost analysis — building up what a part ought to cost from the physics of how it is made — is the discipline that answers them analytically. And increasingly, the input that drives it is the one artifact that already encodes everything about the part: its geometry.

What should-cost means

A should-cost estimate is a bottom-up model of a part’s manufacturing cost, built from the actual process required to make it rather than from historical prices or a supplier’s quote. For a machined part, that means accounting for the raw material, the machine time given the geometry and tolerances, the number of setups, the tooling, the secondary operations, and the labour — each derived from the manufacturing process itself.

The distinction that matters is between this approach and parametric estimating. Parametric tools derive a cost from statistical regressions on historical data — useful, but a black box you have to accept on faith. A true should-cost model builds the cost from manufacturing science: how long a cycle actually takes given the machine’s capability, how much material is consumed given the part’s shape, what each tolerance demands in secondary operations. The result is an estimate you can trace line by line and explain — not just that costs differ, but why.

Why geometry is the natural input

Here is the key insight: the cost drivers of a manufactured part are overwhelmingly geometric. Complex shapes mean more machining time, more tool changes, harder fixturing. Tight tolerances force precision equipment and secondary operations. Thin walls, deep pockets, internal features, and difficult-to-reach surfaces all add cost — and all of them are encoded in the 3D model. The geometry is not a proxy for the cost drivers; it largely is the cost drivers.

This is why should-cost is converging on geometry as its primary input. Established DFM and should-cost tools already import STEP, STL, and IGES files to pull cost-driver features directly from the model. The logical endpoint is a system that takes the geometry and returns a defensible cost and the cost-driver breakdown automatically — turning the CAD file itself into a real-time cost signal.

Where should-cost pays off

Geometric should-cost analysis creates value at several points in the product and procurement lifecycle:

• Design for cost. When an engineer can see the cost consequence of a design choice as they make it — before the part is released — cost stops being something discovered after quoting and becomes a design constraint. Most of a part’s cost is locked in during design definition, so this is where the leverage is highest.
• Quote validation. A should-cost estimate gives procurement an independent benchmark to test a supplier quote against. It does not need to match the quote exactly — it estimates the manufacturing burden, not the supplier’s margin or market conditions — but it reveals which quotes are out of line and arms the negotiation.
• Sourcing and process selection. The same geometry can be costed across alternative processes and materials, helping teams choose the most economical path while options are still cheap to change.
• Competitive teardown. Costing a competitor’s part from its geometry reveals how they hit their price point and where your own design carries avoidable cost.

Why it is hard, and getting easier

Accurate should-cost from geometry is genuinely difficult. Cycle-time predictions from generic CAM systems are notoriously inaccurate because they ignore the dynamics of the actual machine. Cost depends on process knowledge that varies by shop, by machine, by region. Historically this meant should-cost required significant manual setup and expert input even when a CAD model was available — the model gave you dimensions and feature counts, but a human still had to supply much of the process logic.

Two developments are changing that. First, geometric feature recognition has improved to the point where many cost-driver features can be extracted from a model automatically rather than selected by hand. Second, the same geometric-understanding technology that powers shape search — turning a model into a rich machine-readable description of its form — provides exactly the structured geometric input a should-cost model needs. A system that already understands geometry deeply enough to find duplicates is well positioned to understand it deeply enough to estimate cost.

The connection to part reuse

Geometric should-cost and geometric search are two applications of the same underlying capability: making a library of parts understandable by their shape. Once you can represent and reason about geometry at scale, finding similar parts and estimating their cost are adjacent problems. The same index that answers “do we already have this part?” can help answer “what should this part cost?” — and a part you can both find and cost is one you can make far better decisions about.

This is also why should-cost is a natural extension of a part-intelligence platform rather than a separate universe. The hard part — building a system that genuinely understands manufacturing geometry — is shared. Reuse is the first thing you do with that understanding; cost is one of the next.

Where this is heading

The trajectory is clear: from cost discovered after the fact, to cost estimated manually from a model, to cost computed automatically and continuously from geometry as the design evolves. The endpoint is a world where every CAD file carries a live, defensible cost estimate — where an engineer sees the price of a decision the moment they make it, and procurement holds an independent benchmark for every quote. The geometry has always contained the cost. The work that remains is teaching software to read it.