A founder uploads a STEP file to a manufacturing marketplace at 9:47 in the morning. The platform returns a quote at 9:51. The number is 14.20 euros per piece, 200 minimum, ships in 12 working days. The founder places the order. Eleven weeks later, after three rounds of sample disputes, two unannounced material substitutions, and a tooling redesign, the actual delivered cost is 23.80 per piece on a partial batch. The first quote was not a lie. It was also not a number that meant anything.
This is what happens when a quote is generated faster than the part can be reasoned about.
A 4-minute quote prices what is in the CAD file. The CAD file is rarely the right spec. The drawing the founder uploaded inherits its tolerances from a CAD template their contract designer used in 2023, which inherited its tolerances from a different project that needed them, which inherited them from a textbook example. Almost nothing in a typical mid-volume hardware drawing actually needs the tolerances called out. Almost every shop quotes those tolerances anyway, because relaxing them is the customer's job.
A 4-minute quote does not ask about volume properly. The platform has a number field for quantity, and the algorithm uses it for pricing. The question that matters is not "how much would 200 pieces cost on the cheapest available process". The question that matters is "what process makes economic sense at this volume". A part designed for injection molding and quoted on CNC at 200 pieces will return a number. The number will be wrong by a factor of three.
A 4-minute quote does not vet the supplier. It allocates the job to whichever shop in the marketplace's network bid lowest, which in any commodity-routing system means the shop with the most idle capacity, which usually means the shop with the worst track record on the relevant part class. The marketplace's customer-facing brand is reputable. The shop the part actually lands at is often a job shop in a Polish industrial estate that has never machined this material at this finish. The founder will discover this in week six.
A 4-minute quote does not propose changes. It silently substitutes when supply is tight. 6061-T6 is in stock at the routed shop; the founder specified 7075-T6; the part gets quoted on 6061. The quote does not flag the substitution. The first sample arrives in 6061 and looks fine. The fatigue test fails twelve weeks later under a load 7075 would have held. The founder is now four months into a tooling decision that needed to be made on day one.
The 4-minute quote treats the spec as a finished input. It is not. The spec is the work.
Most contract manufacturing failures are decided before a factory is contacted, and the marketplace's economic model — fast routing, optimised pricing, minimal human contact — exists to skip exactly the step where those failures could be caught. The platform is not malicious. It is doing what it was built to do, which is to convert a CAD file into a number quickly. That conversion has commercial value, and there are programs for which it is the right tool. They are not most programs.
A 5-day spec read costs 2,500 euros and produces something different. It produces a written read on whether the part is manufacturable, which process makes sense at the founder's actual volume, which tolerances can be relaxed, what material is the right one, and which two or three shops are worth quoting in the first place. The quote that follows is a quote on a spec that has been pressure-tested, not on a CAD file that has not.
The number on the second quote is not always lower than the marketplace's. Sometimes it is higher, because the spec read identified a process the marketplace would have under-priced. Sometimes it is the same. The point is not the number. The point is that the number survives contact with a real factory and a real production run.
The fastest quote is fiction. The first useful number is the one that survives a spec read.