Does this require a separate data source (e.g. a message queue) to ingest and log the events before reacting to them? Or could you retroactively use the mongodb oplog? I imagine there is way too much data for the oplog to be feasible in this case. And if you didn't design the system from day 1 around an event model, wouldn't it be difficult to retroactively introduce this new architecture?
Event sourcing isn't without it's difficulties but if the domain is pretty well understood then it can be a very powerful pattern.
Stripe has an awesome product and seemingly an equally great engineering culture. I wonder if they have debated using CQRS/ES.