Up to the moment a mould is cut, almost every decision in a hardware programme is reversible. Afterwards, very few are. A tool represents weeks of lead time and a large fixed cost, and modifying one is constrained by an uncomfortable rule: you can remove steel easily and add it back only with difficulty.
Has the design stopped changing?
Not "is it finished" — designs are never finished — but has it been stable through a full round of physical testing without a change that affects geometry? If the last prototype round produced dimensional changes, the next one probably will too.
Do you know your real production volume?
Tool specification follows volume. Cavitation, steel grade, hardening and cooling design are all chosen against an expected total. A tool built for fifty thousand units is a different object from one built for five thousand, and building the wrong one is expensive in either direction — either you overpay up front or you pay again when demand arrives.
Have you tested the material you will actually use?
Machined or printed prototypes rarely behave like moulded parts. Stiffness, surface finish, shrinkage and impact response differ, sometimes substantially. If a functional conclusion was reached on a prototype in a different material or process, it is provisional.
Is the assembly sequence proven by a person?
A design can be perfectly manufacturable at part level and still be a poor product to assemble. Someone should build it by hand, in order, with the intended fasteners and adhesives, and time it. Problems found here are drawing revisions. The same problems found after tooling are tool modifications.
Is the quality standard written down?
Acceptance criteria agreed before production is a specification. Agreed after a disputed shipment, it is a negotiation. Write down what a defect is, which classes are acceptable at what rate, and how each is measured — before there is a commercial reason to argue about it.