A special machine is the only industrial product that is designed, built, tested and shipped exactly once. That single sentence explains most of what buyers find surprising about the price, the lead time and the change orders — and the arithmetic behind it can be written down before anybody signs.
A purchasing manager once asked us, perfectly reasonably, why a machine that contains about 60,000 euros of bought-in components costs three times that. It is the right question and it has an arithmetic answer rather than a commercial one. A special machine is the only thing an industrial company buys that was designed once, built once, tested once and then shipped. Everything else in the plant — the drives, the robots, the sensors, the panels — had its development paid for by thousands of units before the one in front of you was made. Batch size one pays for its own development, alone, and that cost does not sit in the bill of materials where a buyer can see it.
This note sets out where the hours actually go, why a change that takes fifteen minutes in a layout can take three days after commissioning, what the learning curve gives a machine family and refuses a one-off, and which parts of the design method exist specifically because there will be no second unit to learn from. The figures are from the kind of machine we build most often, and every one of them can be checked.
The hours, counted honestly
Take a six-station assembly machine with a rotary indexing table, two pneumatic press stations, a screwdriving station, a camera inspection, a reject gate and a part feeder for each component. On a project of that size the engineering comes out at something like 400 hours of mechanical design, 180 hours of electrical design and panel schematics, 320 hours of control software and HMI, and 120 hours of project management and documentation: 1,020 hours before anybody touches steel. Manufacture, assembly and wiring take around 600 hours, and commissioning with the customer parts another 200. That is 800 hours of build against 1,020 hours of engineering, so engineering is 56 per cent of the labour on the machine.
Two things follow from that ratio. The first is that a special machine is an engineering product that happens to be delivered as hardware, and comparing two quotations on the bill of materials alone compares the smaller half. The second is that the biggest single lever on the price is not procurement but scope: an hour of specification work removed from the requirements removes several hours of design, and an hour added to the requirements late adds several more.
The learning curve the first machine never gets
Manufacturing has known since the 1930s that labour per unit falls by a roughly constant percentage every time cumulative output doubles. On an 85 per cent curve, the second unit takes 85 per cent of the labour of the first, the fourth 85 per cent of the second, and so on. The exponent that describes it is the base-two logarithm of 0.85, about minus 0.2345, so the tenth unit takes 10 to that power, about 58 per cent of the first, and the fifth takes about 69 per cent.
A one-off machine sits at the very start of that curve and stays there. Every hour that would have been saved on units two through ten is spent on unit one, because unit one is the only unit. The drawing that would be reused is drawn from scratch, the fixture that would be adjusted is designed, the software function block that would be copied is written and debugged against real hardware for the first time. When a customer says that the second machine of the same type should be cheaper, they are right, and the curve says by roughly how much: around 15 per cent less labour on the second, around 30 per cent less by the fifth. We quote repeat machines on that basis and say so.
The corollary matters more. If 55 per cent of those 1,020 engineering hours sit in parts that genuinely repeat — the frame family, the panel architecture, the safety circuit, the software framework, the HMI structure, the documentation set — then on the fifth machine of a family those 561 hours fall by the same curve to about 385, while the 459 unique hours stay. The total goes from 1,020 to about 844 hours, a reduction of 17 per cent on the whole machine, achieved without making any single machine less special. That is the entire economic case for modularity in one-off machinery, and it is why we keep a frame standard and a control architecture rather than starting each project from a blank sheet. It is also why the first machine of a new family is genuinely more expensive than the fourth, and why saying so is more useful than discounting it.
Why a late change costs a hundred times an early one
Design-for-manufacture literature has used a rough 1 to 10 to 100 to 1000 scale for decades: a change caught in the concept phase costs one unit of effort, in detail design ten, in manufacture a hundred, after delivery a thousand. It is a heuristic rather than a law, and on special machinery it is conservative. A worked example makes it concrete.
Suppose a part-present sensor needs to move 40 millimetres because the part turned out to sit differently in the nest than the drawing suggested. While the station is still a layout in CAD, the change is a mounting hole moved and a dimension updated: fifteen minutes. After the weldment has been cut and machined, it is a new bracket to design and make, two holes to drill in a finished frame, a modified assembly drawing, a changed wiring list and a re-test of that station: six hours or so, spread across three people. After the machine has been commissioned and accepted, the same 40 millimetres means a site visit, a controlled change to released software, an update to the risk assessment because the guard opening moved, re-validation of the station, a revised documentation set and a change record the customer quality department has to approve: three days of effort and a week of calendar. Nothing about the physical change grew. What grew is the number of documents, approvals and verified states that have to be brought back into agreement.
This is why we are stubborn about the requirements phase, and why we ask for real production parts rather than good samples before a feeder or a nest is designed. It is cheaper for everybody to argue about the specification for two extra weeks than to discover a part tolerance during commissioning.
The design method exists because there is no prototype
A series product is verified by building one and testing it. A special machine cannot be: the first article is the deliverable. So verification has to happen analytically, against the requirements, at every level of the design before the level below it is committed. That is what the V-model for mechatronic systems described in VDI 2206 formalises, and what the systematic design approach of VDI 2221 provides the front end for: a requirements list that separates what is demanded from what is wished for, then system design, then the mechanical, electrical and software domains in parallel, then integration, with a verification step facing each design step across the V.
In practice that means a set of calculations and simulations standing in for the prototype that will not exist. Tolerance stack-up is computed rather than discovered, because a nest that will not accept the part at the limits of its drawing is found on paper or on the machine. Cycle time is simulated station by station, including the index and the waiting, which is a separate discipline of its own. Robot reach and collision are checked in simulation before the cell layout is frozen, because moving a robot pedestal after the floor is drilled is expensive. Stress and deflection on the frame and on any moving mass are calculated, not assumed, and a machine FMEA is run on the station sequence to find the failure modes that a single build will never reveal by itself. None of this is optional on batch size one; it is the substitute for the units that will never be made.
Three risks, and who should carry each
Every special machine project carries the same three risks, and most disputes come from leaving one of them unassigned. Scope risk is the gap between what the specification says and what the plant needs, and it belongs to whoever controls the specification. Part risk is the variation in the components the machine has to handle, and it belongs to whoever owns the part drawings and the suppliers. Integration risk is whether the whole thing makes good parts at rate, and that one is ours.
Written down that way, the usual arguments become easier. A machine that jams because a stamping has a burr outside the drawing is a part problem with a machine symptom. A machine that cannot hold the rate because the specification never stated incoming part quality, air pressure or operator presence is a scope problem. A machine where two stations interfere at speed is ours. We put those three sentences in the contract, together with the output defined as good parts per hour over a continuous run and the assumptions it rests on, because a number without its conditions is not a guarantee.
The regulation changes in January 2027
Anybody buying or building machinery for the European market has a date to note. Regulation (EU) 2023/1230 on machinery entered into force in July 2023 and applies from 20 January 2027, replacing Directive 2006/42/EC. The essential health and safety requirements remain recognisable, and the familiar route through risk assessment to ISO 12100, functional safety to ISO 13849-1 and electrical equipment to IEC 60204-1 stays, but three practical points matter for special machinery.
Documentation may be supplied digitally, with the instructions for use available in the language of the member state and a paper copy on request within a stated period. The treatment of substantially modified machinery is explicit: a modification that introduces a new hazard or increases an existing risk can make the modifier the manufacturer, with a new conformity assessment and a new declaration. And machinery whose safety behaviour depends on software or machine learning is addressed directly, which affects anybody putting adaptive functions into a safety-related chain. For a retrofit project the second point is the one that changes decisions, and it is better established before the quotation than after the work.
What goes in a quotation for something built once
For a one-off machine we quote the engineering, the material, the manufacture, the commissioning and the documentation as separate lines, because they behave differently and the buyer should be able to see which one a change affects. Non-recurring engineering is named as such, with the note that a repeat machine of the same type carries roughly 15 per cent less labour and the fifth about 30 per cent less. Output is stated as good parts per hour under written conditions. The verification plan — what is calculated, what is simulated, what is proven at the factory acceptance test and what can only be proven on site — is part of the offer rather than an afterthought, because on batch size one the verification plan is the only substitute for a prototype, and it is the part of the price a buyer gets the most for.
— GANI Engineering engineering team