What can Artificial Intelligence do to detect fraud?
by Sanjeet Banerji, on Sep 25, 2019 12:24:00 PM
Estimated reading time: 3 mins
Artificial Intelligence (AI) though deemed a fiction in the bygone era has assumed practical importance today. The technology has umpteen number of use cases especially in the Banking, Financial Services, and Insurance (BFSI) sector. It not only helps in improving customer interactions and customer visibility but also in securing digital identity, deploying anti-money laundering (AML) practices, and fraud detection solutions. Enterprises have already started implementing AI in their businesses and started seeing results. Having said that AI is here to assist humans and not take over.
What challenges does AI address?
Let us take the example of a fraudulent transaction in BFSI. A fraud is a regular transaction unless detected. An enterprise loses USD 150 Million every month by way of fraud. The global revenue lost in terms of fraud annually is approximately USD 2.0 Trillion. In some cases, the revenue loss is equal to the GDP of some of the European countries. However, such fraud cannot be detected in absence of data; especially so, when enterprises function in silos.
AI not only helps in creating an integrated view of the business but also helps discover patterns thereby helping detect and curb fraud.
It is interesting to note that these intelligent algorithms enable you to expedite the time, which is otherwise taken to manually handle huge amounts of data or Big Data, and arrive at patterns. For example, contemporary mechanisms working on an enterprise system housing 300 thousand accounts and carrying out 1.5 billion transactions annually would require 4 years to identify just one pattern. On the contrary, AI does the task in a few hours. So to say, AI can not only be leveraged towards social good but also help better understand what “may happen” and thus future-proof the enterprise against fraud.
What AI does to detect fraud?
AI is a boon. However, it cannot work without data. Let us see a couple of use cases, which highlight AI advantages and enable movement towards sustainable transaction monitoring solutions.
Use Case 1: Detection of Money Laundering / Financing of Anti-Social activity through Pattern Mining
AI algorithms work on the concept of self-training. The more they ingest the data, the more they learn from emerging patterns, and become intelligent. In fact, they generate intelligence that is not available in your regular KYC documents.
By focusing on how the transaction is happening, who is sending the money, through which account, into which account, AI helps you decipher the relationships between the transacting entities and their behavior. It also highlights their propensity of committing fraud.
AI helps discover patterns in transactions, normal versus anomalous behavior, associations & relationships between entities, risk scoring of accounts, etc., thereby helping nip anti-social activity in the bud.
AI algorithms can be used in both supervised and unsupervised forms to outsmart the smart fraudsters.
Use Case 2: Monitoring accounts for specific pattern of activity
Accounts can be monitored and flagged for certain patterns. Accounts with low tenure of money stay, multi-currency transactions across geographies, sudden dormancy of high transactions, and unprecedented rise in activity after being dormant, can all be auto-scrutinized with the help of AI algorithms. The algorithms discover specific activity patterns and provide a view of the bigger picture. This is also an important use case in business forensics.
AI enables banking enterprises to take quick corrective and prohibitive actions. When a fraudulent pattern is identified, the algorithms can trigger a bot to stop the transaction. Alternately, they can trigger a bot to take corrective measures and delve deeper to check if the transaction is legitimate and allowed or is surreptitious and disallowed.
Contemporary AI applications
Following practical uses of the intelligent algorithms enable enterprises to build a “digital lens” for monitoring transactions:
- Artificial Neural Networks (ANNs) to detect the patterns and relationships in data and learn through experience and not programming
- Recurrent networks to identify patterns and detect fraud
- Deep Convolutional Neural Networks to perform Language Generation, Speech and Audio Analytics
- Restricted Boltzmann Machine to predict, comprehend text, and perform semantic filtering
- Support Vector Machines to classify transactions
- Generative Adversarial Networks (GANs) to create virtual simulated world, 3 modeling, and predict fraud
Integrated data is primary for using AI and enabling the digital scrutiny of voluminous transactions. The intelligent algorithms help in building sophisticated models to fit transactions in to different clusters. High volumes of data or Big Data help to detect patterns. The more data you integrate the more intelligent the AI algorithms become allowing you to identify and stop fraudulent transactions in real-time. However, it is up to enterprises to evaluate the cost of fraud as compared to the cost of implementation, time, effort, and energy.