Sandeep Anandampillai is Founder and Chief Product Officer of Crediwatch (CW). He shares his insights on the three most important questions the Banking and Financial Services industry is asking about using Artificial Intelligence.
Q: What are some of the leading AI applications in the Indian Banking sector that you have come across?
Sandeep: Globally commercial lenders are incorporating artificial intelligence (AI) and blockchain for back-office as well as customer-facing processes. But widespread adoption of these technologies is yet to gain traction in India. I see a lot of focus on this from most financial institutions. Things are looking up. One of the reasons AI adoption is taking time is because we are still operating as a people-centric and relationship-based society. In other words, many existing relationships are based on individuals trusting each other (lenders – borrowers). In some of the envisioned use cases, AI is a black box, and its outcomes are not well defined. Understanding the decision-making process in a new technology is extremely important for users.
I see adoption of AI tools in customer-facing applications such as chatbots, facial recognition systems, voice-based chatbots, image analytics for document verification, and transaction fraud analytics.
With explainable AI, we are also seeing new tools play a role in risk analytics such as fraud checks, legal due diligence and Early Warning Systems (EWS). Explainability is critical for such processes because people can trust and depend on the outcomes from these models.
A few select companies are helping financial institutions adopt AI better by introducing explainable models. This might play a part in amplifying decision making in core lending systems.
Q. How could banks get better at regulatory compliance and reduce risk with AI?
Sandeep: Compliance is about adhering to rules and guidelines. But rules are fluid and subject to change based on situations. For example, the Covid-19 pandemic has seen authorities relaxing rules for tax filing and corporate compliance. Going forward compliance could be digitising processes and using AI to derive insights from large data sets. This could include alternative data to simplify compliance.
Also, data in digital formats is growing multi-fold and is available in abundance. Using AI to make sense of it other processes like assisted compliance can simplify many of the issues faced today.
Q. How is AI better-enabling fraud detection in this segment? Can you give us some live examples?
Sandeep: Facial recognition systems are being applied to match photos from various reliable sources to the person on a video to confirm identity. At CW, we use VideoKYC systems with facial recognition to ensure counter-party authenticity.
There are models that are running in banks today that analyse millions of transactions and pick out any doubtful transactions or abnormal transaction patterns. These models are trained using large transaction training data and are limiting fraudulent banking transactions.
In addition, there are models that look at geospatial data in correlation to financial and transaction data that help identify fraud or risk patterns. For example, there are models we work on that look at geo data with shipping and export transaction data to check if goods movement has touched sanctioned countries as this is a legal risk.
In my view, Explainable AI brings simplicity to interpret and interact. When coupled with deep machine learning models, one can manoeuvre and automate critical functions such as underwriting, risk monitoring, supply chain financing, and invoice discounting effectively and efficiently.
