Every time someone designs a piece of hardware, she also needs to design a manufacturing line to produce that hardware. This is a non-trivial problem. The early lines of hardware products are always highly un-repeatable, because the designers are still experimenting with different mould injections, circuit designs, CAD software, not to mention the software itself. In this early stages of engineering, where they are doing skews of <100,000 units of their product, they have to pay a manufacturer to design a new line for every change that they make. And in the early stages, those changes are significant, therefore expensive. NRE, or non-recurring engineering, is a central part of every hardware product’s lifecycle. In talking about “things that don’t scale”, YC founder Paul Graham talks about NRE very briefly, referencing the “importance of a screw" It’s worth noting that the idea isn’t really new. 

Mark Leslie, founder of Veritas Software talks about the sales learning curve. At the beginning of an organization, or a new product cycle, the product-market takes a while to be defined, but even once it is, the sales cycle takes a while to really gel. While you may know what problem your product solves, and your customers may even understand it relatively well, or demonstrate a huge market for it, it takes a while to learn how you price them, what story you tell to sell to them, and the channels you need to employ to best access those customers. This explains why big splashy launches make so little sense. No engineer would in their right mind do a run of 1,000,000 units before significant smaller skews of NRE. All business processes have a learning curve, and at the beginning of the learning curve in any process, it is very clunky and very manual. I need to remind myself of that as I work on both product and sales for guild, and grateful to Paul Graham for the reminder.