During this webinar, industry experts discussed the top AI use cases in banks. We’ve included a short transcription of the webinar, beginning at 13:39 of the webinar.

Stuart Neilson, Citi: So the question is, which AI use cases have you found to be the most compelling? This time we can start with Javed.

Javed Ahmed, Metis Corporate Training: Yeah, there's certainly a lot of compelling use cases, from underwriting, risk management, compliance, the scope for analytics is pretty broad. I think one of the things that's really newer is this focus on using best practices to ensure that things like biased inputs and bias data don't make their way into lending decisions. And so there’ll certainly always be a frontier of better risk modeling, better exposure and the fintechs are attacking that frontier pretty aggressively right now and adding additional layers of disintermediation. 

I think within the bank, one thing that's really changing is that we can systematically use our machine learning models to ensure that we're not incorporating biases into our decision-making and that's an area that is growing very fast. People are not doing it as systematically as they really could be. 

Stuart Neilson, Citi: Right, it has certainly always been true that in human-based decision-making processes and machine-based decision-making processes, you have to worry about the bias. There are ways that you can deal with the bias and certainly in a lot of the financial industry context, you absolutely have to worry about the bias. But then intelligently designed artificial intelligence will be able to do that. Any other comments, Ruchi or Anusha?

Ruchi Gupta, HSBC: I can say something. Basically, I think one of the use cases which I have seen in AI is in terms of streamlining in risk management. Especially in the development and challenge of our models. Any time you develop a champion model, which is actually used for all your processes, you always develop a challenge in order to see what you could have done better. So the AI definitely allows you to use different kinds of techniques to see what I could have done to make my model better. 

This also gives you sometimes a lot of insights about your data by different technologies or something like this you can use. This is one of the use cases I've seen a lot.

Anusha Dandapani, Barclays: From my perspective, to add on to that, AI use cases are very effective when it comes to credit card fraud detection, identifying anomalies in credit cards and other different actions. Second, is surveillance and outlier monitoring use cases of AI. That is what we are focused on right now, to identify anomalies and bad behavior and the things that they think are humanly impossible for us to do. 

So all these use cases are augmenting our human decision-making process. We are not getting rid of the human involved in this process. We are just helping other humans to make better decisions. It's basically enabling them to make smart and intelligent decisions. 

Certainly other use cases that I'm interested in and also optimistic about are cyber security and preventing adversity attacks that can come about to an institution. Also doing anti-money laundering network analysis. We're trying to understand what is the state of the transactions that are exposed to the underground network that you don't want to be part of. Also very interesting coming use cases are customer segmentation and driving insights about your customer or 360 degree or holistic view of your customer even before you end up working with your customer. These are important and also a set of use cases where there is human involvement.

Explainability is a very important consonant in the regulatory things that we are exposed to, but how to go about improving the explainability is based on the testing and the stuff that we build around the decision making models. We do that through including transparency and testing them effectively while also identifying potential human bias that can get introduced in the kinds of algorithms or the choice of the models that you choose to solve that problem. Good monitoring is also important. If you don't monitor these models, there is this concept of models drift that can get introduced over a period of time. So model monitoring will improve the trust that is on these AI based models.

 

Learn more and watch the full video on YouTube: https://youtu.be/tDhJJ1Dd-IA.