Clari5

Beating Scams with Smarts: Aiming for a Zero-Fraud Financial Ecosystem

Even as financial institutions implement more sophisticated fraud-mitigation techniques, they have not been keeping pace with criminals. A study by ISMG during fall 2018, to gauge fraud’s evolution and the impact of emerging technology, surveyed 150 financial institutions (primarily in the US), of which 37% had assets under management of $2bn or more.

79% said the number of fraud incidents has remained steady or increased over the past year, while 70% said financial losses from these incidents have also stayed steady or increased.Top forms of fraud were Payment card fraud at 56%, ACH/wire fraud at 49% and phishing (non-business email compromise) at 44%.

When asked about the biggest barrier to improving fraud prevention, 23% said their controls don’t speak with one another and that they don’t want to add any new anti-fraud controls that would impact customer experience. 33% believed technologies like artificial intelligence, machine learning and data analytics have high capability to detect / prevent fraud.

Despite complex processes and procedures in place, the reasons for continued fraud incidents are many. They range from lack of due diligence, inadequate auditing and lack of stringent checks and balances and anti-fraud technology from a bygone era. Meanwhile, sophisticated fraudsters nimbly change strategies to evade detection, even as the quantum of data generated daily by banks becomes more and more massive to sift through.

While banks are doing their best to further tighten their controls, processes and various audits (such as statutory audit, risk-based internal audit, concurrent audit, information systems audit and special audits), they also need to equally importantly consider some of the key technologies available to combat the menace.

Machine Learning

Being able to use ML to spot shifting patterns can form a vital part in improving detection rates while eliminating false positives.

While AI and ML are closely related, there are a few key differences. AI is the ability of a machine to perform actions without human intervention, while ML refers to a particular approach to AI that can take data and algorithms and apply it to new scenarios and patterns without being programmed directly.

AI can mimic actions it has either seen or been previously taught, without any new intervention, and is used to perform a range of specific tasks. Applied AI has been around for a while, for activities like auto-trading stocks based on a predefined set of rules, identifying/sorting images, or even playing chess.

ML is an extension of AI and is the next level in the evolution of the technology. The key characteristic of an ML algorithm is its ability to ingest large volumes of data and ‘learn’ for itself how to apply its knowledge to future scenarios.

This does not mean that ML is the only option for fraud detection use cases. For specific scenarios, where banks are looking for a narrowly-defined set of parameters, or reacting to a new fraud vector, using rules can be the answer for fraud prevention in real-time.

ML meanwhile is better-equipped to deal with spotting evolving patterns and reacting without instruction or human intervention. In fraud detection, AI can monitor the transaction patterns of a customer and send out an alert if it spots a deviant transaction.

With ML, the system can recognize more comprehensive changes in behaviors and bring in data from elsewhere to build its understanding of what a fraudulent transaction looks like without human influence.

Neural Networks

Neural network technology was born from the need to have an artificial system that could perform “intelligent” tasks similar to those shown by the human brain. The inherent nature of neural networks is the ability to learn and being able to capture and represent complex input/output relationships.

Neural networks resemble the human brain because it acquires knowledge through learning and its knowledge is stored within inter-neuron connection strengths (or synaptic weights).

Traditional linear models are inadequate when it comes to modeling data that contains non-linear characteristics. The real strength of a neural network lies in its ability to represent both linear and non-linear relationships and in their ability to learn these relationships directly from the data being modeled.

Fraud Analytics

Various rule-based anomaly detection methods are already being used by many banks, but they have their limitations. Fraud detection capabilities are vastly enhanced with analytics, giving a whole new dimension to fraud detection techniques.

    • Hidden pattern recognition – Fraud analytics helps identify scenarios, new trends and hidden patterns under which frauds occur. Traditional methods miss out on these aspects.

 

    • Data integration – Fraud analytics combs through data and combines data from multiple sources including public records and integrates it into a model.

 

    • Enhances existing efforts – enhances traditional rule-based methods instead of replacing them.

 

    • Harnessing unstructured data – Deriving value from unstructured data is an unexplored goldmine and fraud analytics helps achieve this. In most banks, structured datais stored in data warehouses. Unstructured data is where there’s a high chance for fraudulent activity to occur. Text analytics plays a key role in reviewing this data and preventing fraud.

 

    • Fraud analytics along with performance measurement helps to standardize, maintain control and enables continuous improvement.

 

Entity Link Analysis with Graph Database

Relational databases require datasets to be modeled with sets of tables and columns. By carrying out a series of complex joins and self-joins, rings in such scenarios can be uncovered.These queries complex to build, expensive to run and pose significant technical challenges on scaling. The full extent of this problem becomes apparent as we imagine the exponential explosion that occurs as the ring grows along with the total dataset.

Graphs are designed to convey relationships between data and can help uncover patterns that are difficult to detect using traditional representations like tables. Since they are designed to query intricately connected networks, the graph databases can be used to identify fraud rings in a fairly straightforward manner.

Social Network Analysis (SNA)

The scope of SNA is beyond just social media.The social network is a network of entities connected in a particular fashion. The entities include credit cards, companies, merchants and fraudsters. This can include IP address information, geospatial data, online transactions, and banking data, social media data, call behavior data and other forms of transactional data.

All such data is often stored in unstructured formats in telecom registries, social media, payment gateways or bank servers. There are methods to probe such large networks of relationships and establish suspicious patterns of behavior through graph database technology that has been specifically developed to work with big datasets.

Storing and retrieving interconnected information in a native ‘network graph’ format can deliver interactive network visualizations that identify links in transaction chains, discover hidden structures, locate clusters and patterns, and apply specialized algorithms to identify suspicious patterns.

Advanced analytics methods such as ML are already applied to detect fraudulent transactions. Along with such analytical methods, SNA with graph databases can significantly reduce the false positive ratio in fraud detection. 

Fuzzy Logic

Fuzzy logic is a method of analyzing financial and non-financial statement data. When applied to fraud detection, a fuzzy logic program clusters information into various fraud risk categories. These clusters identify variables that are used as input in a statistical model.

Expert reasoning is then applied to interpret responses to questions about financial and non-financial conditions that may indicate fraud. The responses provide information for variables that can be continuously developed over the life of the bank.

Continuous monitoring of unstructured data helps analyze sentiments, tones, and elements such as incentive, pressure, and rationalization. Fuzzy logic along with SNA can reveal threats of possible collusion.

In Summation

There’s no panacea yet to have a zero-fraud scenario but implementing stricter checks and measures with a layered approach plus architecting and activating an advanced real-time defense framework that harnesses an ideal blend of relevant best of breed technology, helps take a bank’s anti-fraud strategy to the next level.

References

 

Like this article? Share it!

LinkedIn
Twitter
Facebook
WhatsApp
Email

About the author

Naresh-Kurup-sm

Naresh Kurup

Chief Brand Officer
Naresh drives marketing and brand communication for the category-leading banking fraud management product company Clari5.
naresh.kurup@clari5.com