It's hard enough when you have delimiters in quoted fields, but dealing with quoted newlines starts to become unreasonable, especially for line-based tools.
CSV files, as you say, are absolutely wonderful to create. Problems come up when you try to parse files other people write. Not everyone follows RFC 4180.
Plus you've got encodings. If you're accepting CSVs from users, they'll generally come from Excel, which will produce different encoding in different circumstances.
Compare this to connecting to a customer's database directly. You can refresh your (cached) chart data on your own schedule and the customer doesn't need to be involved at all.
Oh and if anyone is thinking, "Yeah but I can schedule a cron job to extract the CSV file and publish it every X minutes", yes you could do that but again that's work for the customer. If the pitch is "sign up, plug in to your database, and boom charts!" there really should be a "schedule automated cron job" component.
All that aside, it is useful to be able to manually add data like this. Particularly for static data where the manual work is infrequent.
http://naa.gov.au/naaresources/govhack-2013/PassengersArriva...
CSV is one of the most open things possible.
As a programmer I do a lot of one-time makeshift data reports for other people, and I always use CSV (or precisely, tab separated) because that's what every program happily emits and consumes. If it does not, it's trivial to transform thanks to UNIX sort, awk and uniq.