Unexpected retrieval costs top the list: a single ad-hoc query across terabytes of cold data can trigger per-GB egress fees that dwarf the monthly storage savings. Poor metadata management is the second risk, because cold data without a functioning catalog becomes practically inaccessible to analysts who do not know which partitions to target. Format portability is the third concern, and storing cold data in open formats like Parquet from the start eliminates future migration headaches.