You run a vacation home exchange/booking site, which you run in 3 instances in America, Europe and Asia because base you found that people are happier when the site is snappy without those pesky 50ms delays.
Now suppose it's the middle of the night in one of those 3 regions, so no big deal, but the rollout of new version brings down the database that's holding the bookings for two hours before it's fixed. Yeah, it was just the simplest solution to have just a one database with the bookings, because you wouldn't have to worry about all that pesky CAP stuff.
But people then start to ask if it would be possible to book from regions that are not experiencing the outage while double-bookings would still not happen. So you give it a consideration and then you figure out a brilliant, easy solution. You just