The Elixir folks are more x-rubyists than Erlangers. They are trying to change Erlang so it will be less whole hog as you say. At the same time they've brought the rails mentality of heavy framework/macro use which has its own problems.
I think the investment can pay off if you're falling into an area that Erlang really excels at, or if you're going to fork an existing Erlang project that already does something you want. Erlang is really good for anything that is telecommunications-y like chat systems, message queues, and custom protocols. It has A+ syntax for parsing binary protocols, which can speed up development.
Erlang is also a good choice if you really need something that needs to be consistent rather than efficient. While requests aren't going to be handled at lightning speed, as you add users latency should stay in a fairly predictable spot. Erlang's message queues tend work best on high memory bare metal nodes, and as long as the queues are taken care of, the system will scale.
But I wouldn't generally recommend it. It's hard to predict if you'll need to scale, and if you do, you'll have to swim in a much smaller community where most of the knowledge is done face to face because everyone knows each other from Erlang Factory. The industry has finally gotten around to solving certain horizontal scaling issues, and as we've already discussed containers can make crashing acceptable at the host process level).
I do recommend reading Joe Armstrongs Erlang book. Erlang is a great way to expand your horizons, and I hear that in Europe they use Erlang to teach classes on concurrency.