The main purpose of data center cooling infrastructure is to provide a good thermal environment for the protection and reliability of IT equipment, especially the servers.
Some data centers use supply or return air temperature to control CRACs. Others base their control policy on distributed physical sensors on the ceiling of hot and cold aisles. However, these approaches can't reflect the ...nonuniform temperature distribution at server inlets—caused by any recirculation or bypass of air between hot and cold aisles—with sufficient granularity.
Nonuniformity of inlet temperatures often produces "false-positives" that will increase overall supply airflow (higher fan energy). Nonuniformity could also result in cooler supply temperatures (lower chiller efficiency). And if you use nonuniform control mechanisms, you will not be able to target cooling where it is needed, which means you will probably overcool servers that don't need the cooling.
Ideally, the IT equipment itself should be able to communicate its discrete cooling needs to the facility's control system. In most real world situations, however, the connection between server thermal status and cooling system is indirect and inaccurate.