Media Perspective: Mining behavioural data could help tap into marketing gold.

There's a guy in Las Vegas who builds computer systems and trains them in Non-Obvious Relationship Awareness. The casinos are regular targets for fraudsters and tricksters, often colluding with employees, hoping to sneak themselves some undetected winnings under cover of the mass of gaming transactions that happen every minute.

So these systems have to act incredibly quickly: matching information from employee databases with known lists of bad guys, arrest records, customer information, credit reports and all sorts of publicly available data. They're looking for non-obvious relationships - like the fact that the guy who's winning lots of cash on table 14 shares a cellphone number with the dealer on the adjacent table. The kind of thing that should ring an alarm-bell somewhere, and might save you thousands of dollars.

Then there's a group at Microsoft working on what they call Surprise Modelling. They've devoured huge amounts of data about Seattle's notoriously horrible traffic and about everything that might cause delays: the weather, sports fixtures, public holidays. Then they set out to warn drivers about bottlenecks and jams. Which isn't really that exciting, because any experienced driver in the area already knows where the bottlenecks and jams will be, and no-one likes software that tells you what you already know.

So this system goes a little further and works out which of the jams it's predicting are ...