Blog · · 5 min read

What will my prototype cost? Why the first real bill of materials doubled the estimate

A robot concept put its parts at about $500. The first full bill of materials said $1,055. How an AI builds a BOM from the CAD model, separates real prices from estimates, and what to check before you trust the total.

A robot concept put its parts at about $470, and after one revision at $506. The first CAD model with a complete bill of materials came to $1,055.08. Same robot, same goal, more than twice the price.

This is not a story about a bad estimate. It is the normal path of a hardware project, and it is the reason a cost number only means something when you know where it comes from.

The robot is Jump & Run Robot, a public project on makemachine.ai: a small humanoid of about 30 cm that people can build themselves. Every version is on the project page, including each price the AI wrote down along the way.

Two kinds of cost numbers

Early in a project, cost is an estimate built from the big decisions. The first concept explored robots of 0.8 to 1 m, at $6,000 to $13,000 in parts. The next version shrank the robot to 30 cm and chose six fast brushless joints, six small bus servos, a small single-board computer and a lithium-ion pack: about $470 in parts. One new part, a smart brushless joint at about $36, still had to be developed.

That is a useful number. It tells you whether an idea sits at $500 or at $10,000, and you learn it before anyone draws a part. Decisions like the size of a product are often made on numbers like these.

A bill of materials (BOM) is something else: a list of every part in the model, with what it is, how many, whether it is made or bought, its mass and its price. The first CAD version of the robot had one. It came to $1,055.08 in parts against a target of about $500, and the model weighed 2,024.4 g where the concept had planned 1,450 g.

Why the total grows when the model gets real

A concept prices what it can see. A CAD model has to contain everything that holds the product together, and each of those things has a price and a mass:

  • fasteners, bearings, shafts, springs and spacers,
  • connectors, cables and the parts that guide them,
  • brackets and covers that exist only because two parts have to meet,
  • printed or machined parts with real volumes of material,
  • bought parts that turn out bigger, heavier or more expensive than the ideal one the concept assumed.

None of this shows up in a sketch. All of it shows up in a BOM. The useful reaction is not to distrust the concept, but to read its number as the start of a range and the BOM as the first figure you can check line by line.

How the AI builds the BOM

On makemachine.ai the CAD module writes the 3D model as code (see CAD as code), and the bill of materials comes from the same model: every part of the assembly becomes a line. For each line, the AI records:

  • Made or bought. A printed leg needs a material and a process. A motor needs a manufacturer, a part number and an offer.
  • Quantity. Six identical joints are one line with a count of six.
  • Mass. For made parts it comes from geometry and material, for bought parts from the datasheet.
  • Price and source. A bought part gets a link to a public offer or datasheet and the unit price at the quantity you need. Where no offer can be found, the price is marked as an estimate.

Two numbers come out of this: the total, and how much of it rests on real prices. The second one tells you how far to trust the first.

A check that catches missing parts

A BOM can look complete and still miss things. One simple test: add up the masses on the list and compare the sum with the mass of the model. In version 1.6 of the robot's CAD model, the BOM totals 3,133.59 g and the model weighs 3.13359 kg, a deviation of 0.0 %.

The same check shows what is not finished yet: 506.7 g of that total are placeholders, masses without real geometry. That second number matters as much as the first. Placeholders are honest, but every gram of them stands for a part whose shape, and often whose price, is still open. How far a model has come is a question of its maturity stage (see the four maturity stages of a CAD model).

Quantity changes the answer

Prices depend on how many you buy. Distributors list price breaks, some parts come with minimum order quantities, and a custom part that is expensive as a single piece gets cheaper in a batch.

When a later concept changed the robot's legs, it compared the parts cost at two quantities: about 6 % less than before for one robot, and about 1 % more in a batch of ten. Small numbers, but a real lesson. A design can be the cheaper one for a prototype and the more expensive one for a small series. Compare at the quantity you will actually build.

What AI cannot do for your BOM yet

  • Quotes for custom parts. A real price for a machined or moulded part needs a quote from a manufacturer, based on the actual file, material and tolerances. An estimate only holds the place for that quote.
  • Prices that last. Online prices change, stock runs out, and shipping and import duties come on top. A BOM is a snapshot with a date.
  • Buying. On makemachine.ai nothing is ordered automatically. A person decides what to buy and when.
  • Judging substitutes. A cheaper motor can fit the budget and ruin the jump. The AI can propose an alternative and show the trade-off; whether it is good enough is an engineering decision.

Five questions before you trust a total

  1. How much of the total rests on real offers, and how much on estimates?
  2. Do the masses on the list add up to the mass of the model?
  3. How many placeholders are left, and what do they stand for?
  4. For which quantity were the prices taken?
  5. How old are the prices?

If a BOM answers all five, its total is a number you can plan with. If it cannot, it is still a concept estimate, just with more lines.

Follow the numbers

On the Jump & Run Robot page you can follow how mass and cost changed from the first concept to the latest CAD version. The numbers will keep moving as the placeholders turn into real parts. That is the point of writing them down for every version.

Watch hardware get built by AI. Follow public projects on makemachine.ai, read every result in full and get an email when there is something new.

Get started See the live projects

More from the blog

Open-hardware projects built with AI