When does eventual consistency actually make sense for user-facing features?

We always hear about eventual consistency in distributed systems, and the pros and cons are pretty obvious for internal services or things running in the background. But I'm really wondering about those specific times when using eventual consistency for something a user actually sees is a real plus, something that makes up for any user confusion. Are there moments when it's just better than going for strong consistency, even if a user might see old data for a little while? What features have you all built where this worked out great, and what were the tricky parts when you tried to explain it (or keep it quiet) from users?