Should-cost software, and where it actually fits
A procurement manager opens a meeting with a number your team did not produce. It is broken down by operation, it carries a cycle time for each, and it is 11 percent below what you quoted. Somebody in the room asks where it came from and the answer is should-cost software.
That moment is the reason this category exists, and understanding what generated that number is worth more to a supplier than owning the tool. The question this post answers is where should-cost software fits, and the honest answer depends entirely on which side of the table you sit. The buyer’s guide covers where this category fits against the others.
What should-cost software actually models
It reasons from geometry and first principles toward what a part ought to cost, without reference to any particular supplier. The engine reads a model, recognises features, assigns operations to each, pulls cycle estimates from a process model and multiplies through rates for a given region and machine class.
Done properly, and aPriori is the serious example, this is real engineering. The process models carry decades of manufacturing knowledge, the feature recognition on a clean model is largely solved, and the output is a should-cost that can be defended line by line.
It scales to parts nobody has made yet, which is genuinely useful. A designer can see the cost consequence of a tolerance while the geometry is still soft, which prevents expensive parts from being designed at all. That is worth paying for and it happens long before a supplier is chosen.
The category also covers lighter tools that model a narrower process. A fabrication estimator reading a flat pattern and returning a cut and bend cost is doing the same thing with a smaller model.
What matters for the argument later is how the number is assembled. The engine picks a machine class, applies a rate for that class in that region, estimates a cycle from the geometry, adds setup spread over an assumed lot size, and adds a margin the buyer chooses. Every one of those 5 inputs is an assumption somebody set.
That is not a criticism. Any estimate is a stack of assumptions and at least these are written down. It does mean the output is only as good as the assumption set, and the assumption set was configured by whoever bought the licence.
Why a buyer gets more from it than a supplier
Look at the problems the tool solves and notice whose problems they are. Consistency across 300 suppliers. Costing a part before anyone has made it. Steering a design away from expensive features. A defensible number to open a negotiation with.
Every one of those belongs to a buyer or a designer. A large manufacturer sourcing globally needs one yardstick applied the same way to quotes from 4 countries, and a model gives them exactly that.
The tool was built facing the other way.
A supplier has the opposite situation. One floor, known machines, known people, and parts that mostly resemble parts already run. The question is not what a generic efficient plant would charge. It is what this work costs here, which is a question about your record rather than about physics.
That asymmetry is the whole post. The buyer needs a model because they have no access to your actuals. You have your actuals, which means reaching for the model is choosing to argue with less information than you already hold.
It shows up in how the 2 sides use the same output. A buyer runs the model across 40 parts and looks at the spread, because they are hunting for the supplier who is 20 percent off the pack. A supplier runs it on one part and gets a single number with no pack to compare against, which answers a question nobody asked.
It also shows up in what each side does when the model is wrong. A buyer who gets a low number opens a negotiation and finds out. A supplier who quotes a low number wins the work and finds out over 6 years, which is the expensive way to learn that a process model did not know about your fixturing.
What the model cannot see on your floor
A should-cost engine prices a theoretical plant, and you do not run a theoretical plant. The cycle times assume a machine in catalog condition. Your rotary transfer holds a cycle the catalog would not print. Your older grinder needs a dresser pass the model has never heard of.
The rates are regional averages. Your burdened rate on the tight cell is specific, earned and usually different. The model assumes clean CAD, where the package that arrived is a scanned print and a revision letter.
Above all the model has no memory. It does not know the plater always comes in low on this finish, that this customer’s cleanliness spec doubles the wash, or that the last part in this family scrapped 4 percent in its first month. That knowledge is not in the geometry.
Take the honed bore on the worked part this site prices everywhere. A feature-recognition engine sees a 0.3125 bore with a tight tolerance and assigns a textbook hone cycle. The record knows a closely related job ran at $0.141 against the $0.135 an estimator wanted to carry, a difference worth $7,200 a year at this volume. No vendor update will ever contain that fact.
Geometry has no memory.
There is a fairness point owed here. History has failure modes too. A record can encode an old inefficiency as though it were physics, pricing tomorrow’s work on a cycle the floor has since improved. A match can be superficially close and structurally wrong, the same shape in a nastier material.
The discipline is to treat the record as evidence to be weighed rather than a verdict to be copied, which still beats a theory, because evidence at least argues back when you question it.
Where a supplier should still run one
There is one use worth paying for even when you price from history, and it is defensive. Your large customers already run these engines against your quotes. Running the same logic against your own number before it leaves tells you which lines sit furthest from the theoretical figure.
Those lines are exactly where procurement will open. Walking into the meeting knowing your hone line sits above the model, and knowing precisely why from your own actuals, converts a haggle into an explanation. The model makes a poor estimator and a useful sparring partner.
The second case is a genuinely novel part. A new process, a first venture into a family you have never run, a greenfield line with no record behind it. There the model is what exists, and a theoretical number beats no number.
The third is the annual price-down conversation. When a customer asks for 3 percent, a supplier who can show what the job costs now against what it cost at award is negotiating from evidence. Knowing where the model disagrees with that evidence tells you which line they will attack.
Buying one without arguing on the buyer’s terms
The trap is letting the model set your price rather than test it. A supplier who quotes from should-cost has adopted the buyer’s assumptions, the buyer’s rates and the buyer’s idea of an efficient plant, then has to defend a number built from someone else’s view of their own floor.
The longer case for pricing from job history works through why that trade is worse than it looks. The short version is that evidence argues back and theory does not.
Scope the purchase accordingly. Buy it as a review step that runs after your number exists, give it to whoever handles the commercial conversation rather than to the estimators, and never let its output populate a quote directly.
Test the number, do not set it.
The licensing usually reflects who the tool is for. These are enterprise products priced for a procurement department at a large manufacturer, with an implementation to match, and a contract manufacturer buying one is paying for a deployment scaled to somebody else’s problem.
If the goal is only to know roughly where the model will land on your parts, there is a cheaper route. Ask the customer to share the should-cost breakdown behind their target price. Plenty will, because they believe the number, and reading 3 of those teaches you more about their assumption set than a licence would.
If the budget only stretches to one system, and you have a usable record of what jobs cost, the record is the better investment. If you have no record at all, fix that first, because it is the input every other option on the comparison depends on.