I don't think they can do it. This is a major headache in any data wharehouseing operation. And no one has every tried to collect data from such wide ranging sources. It's not just a technical problem that can be overcome with some clever engineering, even though that's a massive hurdle. It's also administrative, everybody involved in supplying the data would have to give their total cooperation.
It's staggering to contemplate. How many business are in this country? 20 million or so. That's 20 million data feeds. Every POS system in the country would have to be re-programmed to provide a standard data extract, and that data would have to be transmitted, loaded, checked and then merged. The way I look at it, the IRS can't even implement tax systems where almost all the data arrives with a common key. How are the feds going to handle data coming from 20 million point sources with no common key, and deliver finished data in anything remotely resembling timeliness? I've matched data from large sets for a living, and 80 percent is a good match rate - fine for marketing, but completely inadequate for an investigative tool. The gaps in this process are absolutely HUGE, and any terrorist with a brain will have no trouble avoiding detection.