Recommendation engines suffer from a “cold-start” challenge when they must recommend content to new users for whom there is no existing data on tastes and preferences. When recommending podcasts, platforms like Spotify and Apple can consider applying the playlist data of users who listened to songs but have not yet listened to podcasts. Nazari et al. confirm that this strategy works by building cross-domain recommendation engines (i.e. using models to recommend podcasts to users based on their musical taste). Specifically, they evaluate collaborative filtering (CF) models that leverage some combination of demographics, music metadata, and music taste embeddings against baseline popularity-based recommendation models, and find improvements of up to 50% in consumption.