I have used Cassandra in the past and would consider using it again but I agree its strength is not for 'analytics'. I'd say its strength is needing to scale reads and writes out on large large quantities of relatively simple data that you do not want to lose and you need low latency access to, and for when data needed to be easily partitioned across machines and data centres. I used it for data needed in realtime bidding systems for ad exchanges. It seemed well suited for that.
Then again, it's been 5 years since I worked with it, there have been improvements. Back then it had a _lot_ of warts, and I spent a lot of time fighting JVM GC pauses.