Techniques are described for scheduling data access jobs based on a job dependency analysis. Preliminary data access jobs subsequently accessed by a primary data access job are identified. Execution durations of the primary data access job and the preliminary data access jobs are determined based on historical data. A total execution duration of subsets of serially dependent preliminary data access jobs and predicted execution duration of primary data access job are determined. A time-sensitive subset is identified and start times for the data access job are scheduled. The results of the preliminary data access jobs are provided to the primary data access job prior to the start time of the primary data access job so the primary data access job completes by a target completion duration.