From https://postgres.heroku.com/#protect : "Every change to your data is written to write-ahead logs, which are shipped to multi-datacenter, high-durability storage. In the unlikely event of unrecoverable hardware failure, these logs can be automatically 'replayed' to recover the database to within seconds of its last known state."

I think the question is whether the customer will be able to perform a point in time recovery, and what the size and resolution of the backlog is.