How Big Data Can Fight Against Financial Fraud?

Financial institutions and insurance companies have always been subject to fraud, but the nature of that risk has changed as these institutions provide services through new channels: the internet and mobile access, specifically. Hackers have shown their ingenuity in exploiting new technologies, so even as banking and insurance move into new platforms, and the risk of new attack vectors increases.

How can these institutions stay on top of threats and fraud attack? By using the same advanced technologies that open the door to threats, organizations have a means of confronting and identifying those threats, and being able to better identify all the old frauds and tricks as well. This gives financial institutions and insurance companies a chance to interrupt the activity before any damage is done. Big data – a staple of the financial industry today – can provide financial crime and fraud analysts with a massive amount of information which, if used properly, can help mitigate and prevent financial cybercrime.

## The Problem of Financial Fraud

According to Thomson Reuters report, financial crime cost businesses an estimated aggregated lost turnover of $1.45 trillion (£1.08 trillion), or around 3.5 percent of their global turnover and 40.3 million people are victims of modern day slavery.

Most financial crime is happened to earn easy money through frauds or illegal crimes such as below.

PWC economic crime and fraud survey stated that financial fraud is rapidly growing and 49% companies are reported financial loss and facing internal, external, regulatory and reputational fraud risks.

Some insurance groups estimate that 34 percent of insurance reimbursements are for fraudulent claims, though the number is not really known, and might be much lower – in which case companies need to understand whether they are spending proportionately or disproportionately on fraud prevention.

Companies are using gradually prevailing technology and big data analytics tools to combat against economical frauds. Big data analytics can make use of systems already in place to identify patterns and behaviors, and to agglomerate information that allows fraud to be identified, not only accurately, but efficiently, and use that intelligence to staff appropriately.

Big data analytics allows the development of predictive models using historical data, and correlate those models to real-time events to identify fraud while it is happening. By being able to crunch the data on medical claims, reimbursements, disability payments, past claim records, criminal records, call center notes, social media data, and voice recordings, associations and patterns can be described that will identify suspect claims more accurately, more efficiently, and ideally earlier in the process.

At the very least, suspicious claims can be flagged for investigation based on a dataset previously unavailable – and if available, not usable. Big data analytics has changed that.

## Using Big Data to Detect Fraud

In order to prevent fraud, we need a better understanding and ability to predict where it might occur. Data analytics has given us a new tool in this goal: big data analytics, with its coexistent model of parallel computing. This is usually done in the cloud these days, though some organizations may have their own intensive computing data centers. Either way, parallel computing enables organizations to consolidate, map, and normalize large amounts of data so it can be effectively analyzed.

The amount of data produced by business is staggering. Transactional data, the core of financial computing, can produce mountains of data that can lead to valuable insights. In addition, companies are capturing petabytes worth of information on their operations, vendors, and customers. Financial services firms collect a great deal, but traditionally only use less of this information. With big data analytics, it is possible to use, correlate, and get much more meaningful results from the aggregate.

That used information was originally processed in order to identify sales patterns and promote marketing efforts. Companies now know what products sell where, and even at what times. Using this same information with behavioral analytics and real-time data feedback, it becomes possible to identify aberrations of these normal behaviors, which can signal an attack or security breach.

With big data, we can add in specific customer information. So if a particular customer normally purchases less than $100 worth of a product every month on the 5th, and one month an order for $1000 comes from that customer on the 20th, that can alert the company to single-card fraud or even a system breach. Additional information can be used to identify abnormalities, such as the usual platform or location of a specific user. A change of device or country, or the use of multiple devices in one day from someone who always uses her home computer, could be another warning signal. This is known as continuous or behavioral authentication, and is an effective tool in identifying fraud.

Now, of course, a huge amount of data can be additionally mined from social media data, which might flag discrepancies that signal fraud. Imagine a claim being made for the broken leg of the guy who had to post a picture himself surfing two days after it was said he broke his leg. Information like this is certainly available; the problem usually is making sense of all the unstructured information, finding patterns and consistencies in unlike data.

Although customer transactions were all structured, basically conforming to the same data definitions and residing in matching databases, information acquired from social media lacks that conformity, and either has to be parsed as unstructured data or transformed into a structure. Dealing with this kind of data is among the things big data analytics applications specialize in.

Big data applications allow banks access to and the ability to process huge amounts of unstructured customer data from different sources besides social media, such as logs and call center conversations, and that data can also be of potential help in flagging unusual activity. Software can track and cross-reference such factors as how frequently a user logs in at a certain time of day, what type of machine they log in from – even how quickly they type their username and password. Variations beyond the norm can be flagged for investigation.

Fraud and financial crime from internal sources can also be identified with these kinds of analytic methods. Using behavioral analytics, organizations can have a view into their own structures, and can identify if certain positions or locations are associated with the highest average losses, and which employees have the most opportunity and perhaps motivation to commit fraud.

Even small banks can benefit from big data analytics, particularly by hiring vendors who specialize in analyzing customer data specifically in the interests of preventing fraud.

## Implementation of Data Analytics and Integration

Implementing something large, like a big data analytics framework, can be an intimidating proposal. As with most things, if an organization is just starting to integrate and analyze data from multiple sources, it is recommended they start small – and not to start alone.

• One way to start is to re-examine the data the organization already has, rather than deciding to “get more data”. Applying more sophisticated analytics techniques to existing data can turn up surprising results to help define and refine patterns for groups and individuals. Just asking more questions about the data you already have is an excellent place to start.

• Working with a vendor, or at least doing research and opening conversations with colleagues and experts in the field, is another good way to start. In situations of fraud and financial crime, the more data and models are shared, the more useful they are.

• MPP (Massively Parallel Processing) databases consolidate the management of data. They can facilitate faster and more efficient access to large data sets by using parallel queries and tasks, enabling the meaningful analysis of large amounts of data more quickly than with traditional databases.

• Map/Reduce is a technology used in big data analytics which makes use of distributed processing to work with large data sets across clusters of computers, delegating tasks to worker nodes, collecting the results from the multiple processors, and reducing them into one. Simplification enables more meaningful comparisons, giving investigators better tools for identifying correlations between unstructured data and transactional structured data.

• As your organization’s analytics get more sophisticated, you’ll want more control over the software, specifically the queries you can run against the software. It’s good to try out different ways to analyze and view the data, and to keep in mind that an algorithm doesn’t work better because it gives you a desired outcome; looking at your data in a new light means not having a desired outcome, and being open to seeing new relationships as you try different algorithms. A rule can be derived from the data, for instance, generating a list of employee emails who have responded to phishing emails which can, in turn, generate a list of suspicious IP addresses that were probably used by the fraud artist. Having this information can thus help prevent other employees from being targeted.

• Once your organization has defined models that give you good insight into behavior that can help identify fraud, you must keep experimenting. Fraudsters are known for their persistence and ingenuity, and organizations must be equal in keeping pace with the bad guys. Some fraud detection software vendors understand the importance of innovation, and have made tools available to the customer so they can create or adapt their own rules for analytics. The software vendor often benefits by having a wider library for users to share – it can’t be stressed enough that sharing data can lead to identifying new operating methods of fraudsters.

• Remember to take security into account. Big data necessitates the use of multiple processors and as the amount of data increases the more it makes sense to move to a cloud to handle the distributed computing. Additional equipment even in your own data center increases the risk of mechanical failure and offers more attack points, and migrating to the cloud is no different – except that your cloud vendor can share responsibility for security with you, and has expertise in that field that probably exceeds that of a local IT department. Most cloud computing service providers choose an SSL certificate for the prospect of customers’ data security. SSL security plays an important role to establish secure environment for in-transit data over the cloud. Still, the larger your computing foundation is, the greater your risk of data security failures, data loss, and mechanical failure are.

Even many Banks are using data analytics to detect what are called cross-channel anomalies – that is, comparing data from different sources, or “channels” in order to find discrepancies that could suggest fraud. Matching up data from different channels to identify fraud is a case of very complex problem solving, and it’s the kind of problem-solving that big data analytics is well-suited to handle. The problem is more complex than aggregating fraud alerts from various sources; key information may lie in data that isn’t associated with alert-generating information at all.

Finding discrepancies between differently-sourced data is a massive challenge, and not all analytics software is up to the task. The right software can allow your experts to put together a picture from separate pieces of data in a way that makes sense and from which conclusions can be drawn that will help the business in a number of ways, including fraud detection and control.

Big data requires new thinking that encourages departing from institutional wisdom developed during the days of architecting data warehouses locally, where it was advantageous to summarize data whenever possible, removing data that didn’t seem to have immediate bearing on the business. But in the days of big data, that approach is risky, because important clues hide in the unstructured, multisource data, clues which could lead to the solution for preventing or mitigating a future fraud or financial crime.

## Conclusion

Big data analytics is the most rational, best-tuned solution to identify rules and patterns in behavior that can help identify fraud or financial crime, because it is the best tool to identify correlations between individual or group behavior recognized from multiple sources and cross-checking that behavior with transactional data.

By doing so, banking institutions will be relying on nontraditional data combined with traditional data to predict attacks, based on trends that are identified in their own data and that of the industry.