The procedure is outlined in each service's SLA -- though I think they're all pretty much the same.
Annoyingly, they expect you to do the leg work to show when the outage happened and supply logs demonstrating that you were impacted.
Might want to do some napkin math first to see if the amount credit is worth your time. The couple times my org considered pursuing it, it just wasn't worth the effort. (Though, personally, I think that speaks to a larger problem with the SLA.)
