I can't imagine the current state of AI supporting these types of interactions (or humans wanting to deal with AI on the other side), but would be happy to be proven wrong. I suggest focusing on this single use case of yours and applying it to your idea: "Find 30 leads for customer development for exploring a new market segment." Whether an MVP is successful or not will be a good indicator of feasibility.