Watson became something of a celebrity after beating the champion human contestants on “Jeopardy”, an American quiz show. Its skill is to be able to process millions of documents quickly by reading and “understanding” ordinary written language. Computers have no trouble with searching data neatly sorted in databases. Watson's claim to fame is that it can do the same with “unstructured data” such as those found in e-mails, news reports, books and websites. IBM hopes that Watson may, in time, do some of the work that human analysts do now, such as reading the financial pages of newspapers, looking at thousands of company results and forecasts and producing a list of companies that might be takeover targets soon.









Citigroup has hired Watson to help it decide what new products and services (such as loans or credit cards) to offer its customers. The bank doesn't say so, but Watson's first job may well be to try to cut down on fraud and look for signs of customers becoming less creditworthy. If so, Watson will be following other computers designed to deal with “big data”. Across a slew of new firms in Silicon Valley and in big banks across the world, a range of new ideas is being tried to crunch data. Some have the potential to change banking from the bottom up.
In most financial institutions the immediate use of big data is in containing fraud and complying with rules on money-laundering and sanctions. Even seemingly simple tasks, such as checking the names of clients against those on a sanctions blacklist, become immensely complicated in the real world, where banks may have thousands of customers with the same names as those on the blacklist. Each becomes a false positive that may embarrass the bank and ruin a client relationship. So banks have had to turn to computers that can amass data from a variety of different sources, including the customer's nationality and address, the names of family members, and whether they have travelled to or received money from countries on sanctions lists.
When moving on to more complex tasks, such as identifying the tiny percentage of fraudulent transactions among the millions of legitimate ones, the demands become ever greater. The problem is getting bigger because as banking has moved onto computers and mobile phones, and payments have shifted from cash to cards or electronic transfers, the opportunities for fraud have proliferated.
The danger of fraud is particularly acute in areas such as card payments and some of the more innovative kinds of money transfers that are offering cheaper or more convenient services than those already available. PayPal, which dominates online payments, barely survived its first year in business after it came under sustained attack from fraudsters, and several of its early rivals were cleaned out and had to close down.
PayPal came up with Igor, a computer system named after a Russian thief and hacker who had opened fake accounts and taunted the firm's security team in e-mails. Igor would look for patterns, such as a concentration of payments close to the top limit and their destinations, and then compare those payments with all the others in the system. What started at PayPal soon spread to the rest of banking and beyond it.
A better kind of crystal ball
The firm that has perhaps gone furthest in finding useful connections in disparate databases is Palantir Technologies, which takes its name from the magical all-seeing crystal balls of J.R.R. Tolkien's mythology. It was founded by a group of PayPal alumni and backed by Peter Thiel, one of PayPal's co-founders. Its speciality is building systems that pull together information from different places and try to find connections. Some of its earliest adopters have been spy agencies. In America the CIA and the FBI use it to connect individually innocuous activities such as taking flying lessons and receiving money from abroad to spot potential terrorists. Its other main market is in banking, where big firms such as JPMorgan and Citi use it for a range of activities from structuring equity derivatives to reducing loan losses.
A stablemate of sorts to Palantir is Xoom, a firm that specialises in cross-border remittances. It is backed by some of Palantir's investors and has swapped a senior employee with it, but more importantly it shares Palantir's belief that given enough data even the toughest risks can be managed. Xoom accepts payments from bank accounts or debit cards in America, then hands over cash in countries such as the Philippines or India. It does not have much time to find out if it has been swindled on a payment before it has to produce the cash. So it has devised a sophisticated computer system that analyses a range of data, the nature of most of which it will not disclose.



Some of these checks may seem obvious, but some are not easy to do when processing millions of transactions and moving billions of dollars. Moreover, few of these pieces of information on their own are powerful enough signals for Xoom to decline or agree to make a payment. Yet when the computer looks at all of the payments in its system, it is remarkably good at weaving together the bits of information to spot fraud.
It also learns as it goes. When it recently noticed a string of payments funded by Discover credit cards and originating in New Jersey, its algorithms raised a red flag even though each payment looked legitimate. “It saw a pattern when there shouldn't have been a pattern,” says John Kunze, Xoom's chief executive. The pattern it found turned out to have been an effort by a criminal gang to defraud the firm.
The other big users of fraud-fighting computers are credit-card associations such as Visa and MasterCard. Their systems, as well as those of big card issuers such as Capital One, look at vast numbers of transactions for unusual patterns or connections. This has allowed them to graduate from simple rules-based fraud detection (such as whether a credit card has been swiped in locations a long way apart in a short space of time) to more complex sorts.
None of these systems is cheap, but they are usually a lot cheaper than falling victim to fraud. Xoom puts its losses through fraud at 0.35% of the sums transferred. The average for credit-card firms is about 0.1%, and the best achieve rates of about half of that, says Mike Gordon of FICO, the company that invented credit-scoring and now also supplies fraud-detection software. Losses on cashed cheques in America run to about 1% a year. For companies selling goods online, loss rates are considerably higher. CyberSource, an electronic-payment and risk-services company, says that online retailers in Britain reckoned on losses of 1.8% of revenue last year.
The high cost of fighting card fraud has changed the balance of competition in banking, weakening smaller banks that lack the scale to build the necessary systems. Many closed or sold their own credit-card businesses and instead signed their customers up to cards issued by large specialists such as MBNA or Capital One. Many smaller banks now think this was a mistake, depriving them not only of an important source of revenue but also of the opportunity to form the deeper and more lasting relationship with their customers that comes from selling them several financial products. Most important of all, perhaps, it has deprived them of a rich source of data on their customers' spending patterns.
That may soon change, for two reasons. The first is that card associations such as Visa and MasterCard are getting better at spotting fraudulent transactions as they pass through the network, relieving the burden on smaller banks, says FICO's Mr Gordon. The main strength of these network-level systems is that they are able to look at far more transactions than any single bank could, which helps them to spot fraud patterns on an international scale.
Second, the systems used to crunch data are becoming commoditised and their price is coming down. Thomson Reuters reckons that last year venture firms invested a total of $2.47 billion in companies that want to crunch big data. Much of this investment was in database and storage outfits that are not specific to banks, yet the tools being developed elsewhere are quickly spreading. Whereas a decade ago the big banks would get their systems custom-made at huge cost, smaller banks can now buy similar ones off the shelf at a small fraction of the price.
Bankinter, the tech-savvy small Spanish bank, last year started using a system to analyse complex loan portfolios on computers run by Amazon, an online retailer. Cloud computing enables it to hire massive number-crunching capacity whenever it needs it. These two factors are making it easier for smaller banks the world over to keep their credit-card businesses to themselves and lean against the powerful forces for more and more consolidation in banking.
From The Economist 2012
Link http://www.economist.com/node/21554743