In order to implement tokenization, there will need to be two databases: the database containing the raw, original data, and the token database containing tokens that map to original data. Having two-factor authentication is nice, but certainly not required. Encryption keys are not necessary for tokenization. Two- person integrity does not have anything to do with tokenization.

Tokenization involves the replacement of sensitive data fields with key or token values, which can ultimately be mapped back to the original, sensitive data values. Masking refers to the overall approach to covering sensitive data, and anonymization is a type of masking, where indirect identifiers are removed from a data set to prevent the mapping back of data to an individual.
Encryption refers to the overall process of protecting data via key pairs and protecting confidentiality.

The Cloud Security Alliance publishes vendor-neutral cloud security best practices (like the Cloud Controls Matrix). AWS and Microsoft publish their own (AWS/Azure) guidelines, which are vendor-specific. (ISC)?provides certification info but not a full design pattern repository.

Whereas a contract spells out general terms and costs for services, the SLA is where the real meat of the business relationship and concrete requirements come into play. The SLA spells out in clear terms the minimum requirements for uptime, availability, processes, customer service and support, security controls and requirements, auditing and reporting, and potentially many other areas that define the business relationship and the success of it.