An electronic device is programmed to start a specific operation at 07:30 every morning to ensure it is synchronized with other systems by 09:00. Through extensive testing, it is observed that the device encounters a timing failure once in every 20 operations when starting at 07:30. Alternatively, when the device starts at 07:20, it experiences a timing failure once in every 50 operations. Assuming the timing behavior follows a normal distribution, determine the latest starting time for the device to prevent more than one timing failure in every 100 operations.