As the volume of data was large (pertaining to over 800,000 subscribers) it was recognised that there would be errors in some data which would make the system reject that data. Thus the data migration was designed to happen in several iterations, with the faults identified in each iteration being corrected in subsequent iterations.