How do you guys approach decision making when it comes to ML-powered solutions?

The approach I have followed so far in my career follows decision-making based on collecting evidence as quickly as possible based on evaluating the performance on the ML-level (anecdotal to iterative creation of test sets), rapid prototyping on the user interface and UX level (as most real-world production systems require a human-machine interaction approach), and building functional PoCs as soon as possible.

Based on as much insight as can be collected in a period of weeks to sometimes several months we then try to make a GO/No-GO decision if the solution is feasible and has a good return on investment.

I am extremely curious how others approach this decision-making task.