What's the data problem for drones?
Flying well in easy conditions is the simple part. The data that matters covers wind, tight spaces and close work next to structures, where a mistake with real hardware means a crash, a damaged aircraft or a safety incident.
Autonomy teams also need judgment: was an approach stable, was the drone too close to the tower, should it have aborted and tried again? Good pilots make those calls quickly, and that's hard to write down as rules.
Simulators let trainers fly those situations many times with no aircraft at risk. The limit is that sim aerodynamics and sensors aren't your airframe's, so flight sim demos are strongest for decisions, paths and timing, used alongside real flight logs.