Mainly, he said, rolling back on code or updating is just putting the binary on the server and run this. This means a downtime in the service. Which might be pretty harsh for a real time data logging and messaging service. With Erlang, and it's hot code reloading feature, this would be a non-issue.
Also when he talked about the system rarely crashing, could have also been better addressed with Erlang, as it is part of the main philosophy to deal with this.
I think Erlang would have been an even better choice for this company.