Designing an AI tool taught us to design for uncertainty and scale. We learned that AI recommendations need enough relevant results to feel useful, while accounting for variables like seasonality, destination, timing, and activity type. Not every destination or activity can produce the same number of valid recommendations.