Remy Blaire: Financial institutions are embracing AI, and banks have graduated from GenAI to agentic AI, which can be used for fraud investigations, anti-money laundering practices, as well as customer servicing and credit operations.
But major implementation roadblocks do lie ahead.
Now, legacy architectural debt means mainframe infrastructure is slowing down agent execution. Meanwhile, systemic regulatory frameworks do require absolute auditability.
At the same time, early implementations do show friction rather than productivity, which can result if autonomous tools are deployed without updating developer workflows or operational handoffs.
Joining me to weigh in on the tech AI landscape within financial institutions is Dimple Thakkar, Founder and CEO of PradimeAI.
Dimple, great to have you here. Thank you so much for joining me.
Dimple Thakkar: Thank you, Remy. Great to be here.
Remy Blaire: Well, when we're talking about the role of artificial intelligence within financial institutions, we have to keep in mind that it is a highly regulated space.
So give us the lay of the land with agentic AI right now.
Dimple Thakkar: Yeah. This shift from generative AI to agentic AI has been significant, Remy. And we are now talking about AI shifting from just giving answers to making decisions.
So, for example, in banking, the AML investigations and the fraud arena has spent around, there were $40 billion of spend just last year on investigating whether it's a fraud or not.
So now, adding the AI layer on top of it, imagine the flow would be, you know, AI going and investigating the fraud, investigating paperwork, looking at customer communication records and making decisions itself.
So from a customer perspective and banking perspective, this arena has changed a lot.
And we need to make sure that the AI trusted-truth factor makes its way very soon before we lose control over, you know, AI making decisions in banking.
Remy Blaire: Yeah. And indeed, because the utility of agentic AI is there, we understand the opportunity.
But there are also challenges, especially when we're talking about the heavy lifting here and heavy workflows.
So where do the challenges lie right now, and what can be done?
Dimple Thakkar: Yeah, I feel like the industry is a little misled here.
We feel that only customer service is the prime target here for AI, but not so much.
I personally feel that in the past 25 years, I've seen that fraud, AML and KYC can be the biggest area where AI needs to be monitored.
And the reason being is that there is people's money involved.
So when we look at transactions, when we look at processing, when we look at login failures or card declines, that's the biggest opportunity.
But I feel that fraud, AML, KYC for both commercial clients, as well as automation in these areas, you know, there's a lot that AI can do there.
But at the same time, organizations and banks will need to make sure that all of the paper trail, as we call for AI, is beefed up a lot more in terms of automated processes, reconciliations and tracing back to the data sources for each transaction.
Remy Blaire: Yeah. And we all know that we're not looking for perfection at the initial stages.
So when things do go wrong, especially in a highly regulated space such as financial services or even within financial institutions, what can be done, and what is actually being done right now?
Dimple Thakkar: Yeah, that's an interesting point and question.
So we are now getting into an era in financial services where there are autonomous agents that are being created by a swarm of agents alone.
And we heard that this week. In fact, as we speak, there is a meeting going on within the UNGA to discuss AI security.
So for financial services and healthcare, and some of the highly regulated industries, it means that there is a graduation that's required from not only talking about traceability of data, but also being able to prove that the data model outputs are correct.
And that means that we need to have all of the safeguards and guardrails in place, not only to track, but to make sure that they are monitored in real time.
And that, I think, is one of the biggest challenges that most of the big banks that I know of are facing right now.
When we talk about data sources or when we talk about the infrastructure, the cybersecurity, it's no longer that the processes within the traditional cybersecurity, you know, there were traditional processes that were being handled.
But now with AI, that entire landscape has changed.
So we will see this graduate to an intermingling of cybersecurity with the monitoring and the real-time traceability that we need for AI.
And I think that's the challenge that most of the financial services, asset and wealth management firms are facing right now.
Remy Blaire: Yes. And as you mentioned, of course, the role of agentic AI means access to data, and some key data that is regulated.
So what is your future outlook as we move forward? Do you think there must be a human in the loop, or autonomy, or somewhere in between?
Dimple Thakkar: Yeah, that's a great point and question, right?
So everybody is wrangling around that point right now.
As we said, there is already an era where there is automatic transaction processing that is occurring right now as we speak.
So I feel like that transition still needs to happen.
I do feel that most of the transaction processing, looking at loan files, things like communicating directly to customers, is happening already with agentic AI right now within banks.
What I do not see happening is making a decision on a loan or making a decision on whether the transaction is fraudulent or not.
That, I think, will cease to happen just, you know, for regulatory aspects. And that's how it should be.
Unless we can prove that the decision that's made for that loan for a particular ZIP code, or a KYC decision made for a commercial client, is accurate with a human in the loop, I don't see that happening for some of the large or midsize banks.
That's too much of a risk to take.
So I definitely feel that there should be a human in the loop, check-marking and making the decision accurate and making sure that that output is correct before sending the communication to the client, whether it's a fraud transaction or whether it's a KYC or a loan decrease or increase or a loan sanction for a client, because it means people's livelihoods.
Remy Blaire: Well, Dimple, this is a very important conversation. So I appreciate your time as well as your thoughtful conversation on this topic today. Thank you.
Dimple Thakkar: Glad to be here, Remy.
Remy Blaire: Yeah, thank you very much.
