Should Costing in 2026: the methodology, the tools, and what actually works.

Accuracy depends on two things: the quality of input data and the consistency of methodology — not on the calculation engine itself. For mature manufacturing technologies (turning, milling, stamping, injection molding) with current master data, a well-built should-cost model typically lands within 5-10% of supplier cost. For complex assemblies, new technologies, or low-volume programs, the band is wider and requires iteration. But accuracy is rarely the right metric on its own. What matters in a negotiation or a steering committee is defensibility — whether every assumption can be traced to a source and every cost driver can be explained. A model that is 10% off but fully defensible wins arguments a 5% off model with hand-waved assumptions cannot.

In Excel with a generic template, a single part typically takes anywhere from a few hours (simple turned component) to several days (complex assembly with multiple sub-processes). With dedicated product cost calculation software, populated master data, and pre-configured process templates, the same work compresses to hours. The deeper question is whether your speed matches the rhythm of the business: NPI cycles need cost feedback in days, not weeks; procurement negotiations have closing windows measured in days. When should-cost capacity cannot keep up, requests get dropped, rough estimates replace real calculations, and the value of the program collapses. Speed is not a feature — it is the variable that determines how much of your spend actually gets analyzed.

Should-cost methodology works best for discrete manufacturing — processes where a part can be decomposed into materials, process steps, and overhead, and each parametrically modeled. Strong fit: casting (die, sand, investment), forging and forming, machining (turning, milling), sheet metal forming and stamping, injection molding (plastic, rubber, micro), PCBA. Tset supports 220+ pre-configured templates across 35+ such technologies. Weaker fit: continuous process industry (chemicals, food, refining), commodity raw materials priced on indices, additive manufacturing (still maturing in cost methodology), textile and apparel manufacturing. The fit question matters at portfolio level — procurement spend dominated by weak-fit processes will not see the ROI of a structured should-cost program.

For a single analyst working on a dozen models, Excel is the right tool. The breakdown is consistent and predictable: it happens when multiple analysts share workbooks and arrive at different numbers for the same part; when master data drifts between sheets and no one is sure which version is current; when models become a single point of failure tied to one person's knowledge; and when other departments need access to the methodology but cannot operate the workbook themselves. At that point, the issue is not Excel's features. It is the structural fact that knowledge lives in people and files instead of in a system. Moving to dedicated product cost calculation software is not a feature upgrade — it is a different mode of working that lets cost engineering scale without growing.

Parts of it, increasingly. AI is genuinely useful for extracting bill-of-materials data from 3D models and CAD files, suggesting comparable parts from a benchmark database, classifying parts by manufacturing process, and flagging anomalies in supplier quotes. What AI cannot yet replace is the judgment layer: choosing the right process for a given part, defining defensible assumptions about overhead, and explaining the cost story to a non-expert in a steering committee. The most useful pattern in 2026 is AI as accelerator, not replacement — automation of the mechanical work, with the cost engineer staying in the loop for judgment and defensibility. Models that lose the human in the loop also lose the ability to defend the number when it gets challenged.