They fail when tooling is commissioned for a solution that was already patented. When a plant is sized for demand nobody actually measured. When the real unit cost appears with the first production run and there is no margin left to redesign. And when the product works so well that installed capacity cannot keep up, and growth turns into a cash-flow problem.
All four are sequencing errors, not engineering errors. That is why we work in closed phases and in this order: freedom to operate first, then demand, then the operating account, and only then scale-up. Each phase ends in a deliverable and an explicit decision to continue, redesign or stop. Stopping in time is also a result.