How should availability zones and regions be designed for production in Cloud Architecture?

Map every dependency to its failure domain, use multi-zone services first, and add multi-region only when business recovery requirements justify it.