How big data is being used in the fight against card fraud
Financial institutions have been working hard to implement technologies which better protect them and their customers. Innovations such as chip and PIN have made cards more secure, while the introduction of two-factor authentication has also greatly helped. However, none of these systems are perfect and fraud levels continue to rise in many countries.
Fortunately this is an area where big data can make a significant difference.
Fraud detection relies upon the use of both relational data and non-relational data – structured and unstructured data. Banks know where our home branch is and they know within a standard range of incidences how often we take out money or use our cards. They know where we work and the types of transactions we make most commonly. These known quantities form part of the fraud detection process.
The unpredictability of humans….
But the nature of human beings is that we do not always stick to the path most travelled. Consider a man buying an engagement ring. This is a perfect example of something which many people may only ever do once in their lives. Looked at in terms of that data point it appears to be an entirely random act. A man on the hunt for an engagement ring may never even have set foot in a jewellers shop in his life. He will almost certainly have never spent a month’s wages on a single item of jewellery. In the world of data analysis such screaming anomalies should be cause for concern.
So why are more people not challenged when making this kind of transaction? Because it is incumbent upon banks not to leave us frustrated or without cash in our hour of need. The man buying an engagement ring may have enough on his mind without having his card declined at the point of sale. For every transaction, a bank or a card issuer is dynamically analysing data and weighing up the risk of fraud.
The key lies of course in big data.There are many types of data, unrelated even to the individual which can quickly be referenced – in split seconds thanks to the power with which big data is processed – from data relevant to that individual, such as past travel habits, to data relevant to the nature of the suspect transaction. Patterns of where fraud has been conducted and past behaviours of the rightful card owner are a key element in detecting and predicting fraud, to the extent that diverse frauds can be traced quickly back to specific businesses where a rogue employee may be stealing card numbers or cloning cards. Banks now have that reservoir of data related to fraudulent activity and quickly assess the likelihood of transactions being suspect.Back to our hopeful man shopping for an engagement ring, he may go unchallenged because that reservoir of data shows that very few fraudsters try to conduct card fraud in jewellery shops. This may be because of high levels of CCTV coverage in such shops.
…is actually quite predictable
Believe it or not, the apparently unusual purchase of engagement rings is actually fairly predictable. Somebody may never have bought an engagement ring before, but his card issuer may see from his profile that he fits the bill of somebody due to.
A recent mortgage, an increased focus on saving, a second name appearing on a credit card, an expensive holiday in contrast to the budget breaks or golf holidays he has been on with his friends in the past, regular outlay on home furniture which suggests he is setting up home with somebody, a move from a sporadic expenditure on food to a more structured approach which suggests more nights in at home, less time spent in bars, more time in restaurants. These are all behaviours which become powerful data points in the overall modelling of customer profiles. He may think his purchase of an engagement ring is extraordinary and a huge step but his bank – or at least the data it holds – may have been able to tell it was coming for some time.
Ultimately, data has the ability to look at us dispassionately and non-judgmentally. We may think it outrageous that we celebrate a birthday once a year by spending over the odds on dinner at a fancy restaurant we’d never normally visit on any other day of the year. But if this is something we only do once a year then the data barely needs to be sweated to find an answer as to whether this is fraudulent behaviour. Once per year may seem occasional to us, or even out of character, because it is at odds to the way we behave on the other 364 days of the year, but the data sees a creature of habit. It knows it’s our birthday – not least because our date of birth is among the relational data held about us. That fact can easily be linked to any spike in spending on that date.
Even our one-off purchases are almost certainly not as one-off as we think. Most people exercise a degree of financial caution and display behaviours our banks will be able to understand from the data they collect and analyse. We may only buy one engagement ring in our lives but the money spent on that engagement ring will almost certainly be within a predictable parameter based on our salary, our regular outgoings and the value of our mean and median purchases and the statistical outliers we will have created with purchases such as a car or a house.
All of this means that the vast majority of our spending is reassuringly predictable. The insights into our behaviour that big data analysis brings about provides a clear benchmark against which fraudulent activity can be rapidly identified and shut down.
About the Author
Adebayo Sanni is the Country Director Nigeria at Oracle.