The 10 SQL queries against 1,311 rows represent a lightweight data retrieval phase; compute concentration almost certainly lies in LLM synthesis rather than SQL profiling. This workload is well-suited for scheduled runs given the predictable query volume and row count, though on-demand execution remains viable if latency tolerance exceeds a few minutes. The primary lever for cost reduction is optimizing prompt design and output constraints in the synthesis phase—reducing redundant API calls or token consumption per call will yield far greater savings than query optimization at this scale.