How AI Is Reshaping Financial Services with Paresh Ashara

Summary

In this episode of AI Speed, Evan Cholfin interviews Paresh Ashara, VP of Data Analytics at Quinte Financial Technologies, about how AI is transforming financial services. They discuss AI adoption challenges, innovative use cases like document discovery and automation, and the future of AI in banking.

Takeaways

Financial institutions are risk-takers in tech adoption but need a clear ROI to scale AI.
AI can automate mundane tasks like document processing, saving days of manual work.
High-quality data and transparency are critical for AI success in finance.

Soundbites

"Business moves at AI speed, not internet speed."
"AI can reduce days of manual work to minutes."
"Transparency and trust are key for AI models."

Chapters

00:00
Introduction to AI Speed and Guest
00:28
Evan introduces Paresh Ashara and Quint Financial
01:19
Why Paresh was drawn to AI in finance
02:29
AI adoption trends in financial services
04:08
Quinte Financial’s approach to deploying AI
07:57
Core problems for financial institutions with AI
09:41
Paresh’s current focus and solutions
13:28
Document discovery and compliance automation
17:04
Automating loan application validation
17:46
The biggest challenges in AI adoption in finance
22:06
Trends in AI: Hype vs Reality
26:28
Vision for AI in 12 months
28:21
Closing Remarks and Future Outlook

Video

https://youtu.be/3tuuz0b8bns

Transcript

SPEAKER_00 0:20

Welcome to AI Speed, the show where AI-powered companies talk about what actually works in the market right now. Business doesn't move at internet speed anymore. It moves at AI speed. And the people who figure it out how to turn models into money will own the next decade. I'm Evan J. Chamilton, founder of Luxhammer and growth partner to high-performing brands. Today I'm thrilled to be joined by Anadash Sashimara, VP of Data Analytics, AMI and Automation at Quint Financial Technologies. Manadash is working at the forefront of applying AI and financial services, helping institutions move from data to decisions through automation, analytics, and the shift towards vertical AI. His role sits directly at the intersection of strategy and execution, making sure AI actually delivers results inside complex organizations. Thank you for being here.

SPEAKER_01 1:19

Thank you, everyone. It's my pleasure joining this podcast.

SPEAKER_00 1:23

So you've built a career around data analytics and AI and financial services. What drew you specifically into the space and what made you focus on applying AI at the operational level?

SPEAKER_01 1:36

Certainly. If you look at, I actually have been in the IT services industry for almost past 27 years now. And all throughout, it's been working with financial services companies. And what I have seen over these years is the financial services industry is the most regulated industry. When it comes to technology adoption, they are the first one to try out new tools and technologies as well. So they are risk takers in terms of technology adoption. But at the same time, they try to create a balance with respect to what the regulatory compliance needs are in terms of ensuring that it is balanced in terms of all the business decisions that they take. So gradually, if you see the past and especially over the past uh three years since the launch of tools and uh technologies around, it has actually given a new meaning to the financial services industry and not just financial services industry per se, but across industries also. There are specific use cases which the companies have um identified where it can play a very significant uh role. With that uh kind of uh focus, uh the come most of these financial uh uh institutions have started at least exploring uh what it means and how they could leverage this as a part of their day-to-day operations and the business decisions that they make. What I have seen is there is an interest from the uh from the banks uh and other financial institutions in terms of uh exploring these uh AI tools. But um going it uh beyond the exploratory stage, it does take uh some more effort and decisioning uh from the top leadership to adopt and put it into production. There are ways in which this could be implemented and managed uh to an extent, but again, there are uh in terms of uh what they look at as a value add with respect to their business uh functions, is what going to decide the fate of uh all the exploratory work that uh they do in the AI space. So that's that's where I see uh the traction will have to take place uh uh uh uh going forward, where uh the decisions are uh done by the business owners, they have to take active um a role in um uh determining the fate of their initiatives. And unless uh they see that there is a real ROI uh is coming in, they would be reluctant to put this into production.

SPEAKER_00 4:07

Absolutely. So, from your position leading data and AI initiatives, what makes Quint's approach unique and how it actually gets AI deployed inside financial institutions?

SPEAKER_01 4:17

Right. So we we work with uh financial institutions. Our focus is in providing automation opportunities to uh all our clients where we have uh operational services uh engagement. Now, let's say for example, uh we uh primarily have um support uh service uh uh engagements uh with most of the banks and credit unions in the US, where we help them in their contact center processes, in the loan servicing, mortgage uh validation, documentation support, KYC related, dispute management processes, all these various processes where we have uh human agents, human analysts who are actually looking into these uh activities, leveraging uh standard operating procedures and uh ensuring that uh all the business processes that require certain kind of uh checks and balances, QC or quality checks to be done on the outcomes of uh individual loan decisions, etc., are followed through. Now, if you look at most of uh these uh activities are um uh manual in nature and probably it may involve uh uh shuffling between uh different um core applications within the bank or other set of applications. Now imagine if you could determine a pattern in which some of these activities could be automated leveraging I. So that's where we are uh looking at building automated solutions, and as a part of uh one of uh the process areas that we have narrowed down too, we have built a platform called QIDesk, which helps uh banks uh orchestrate uh uh their contact center uh uh process through the omnichannel uh support. And uh here is where uh the email uh channel, uh the chat channel, and the voice channel can be automated leveraging GNAI capabilities. Now, as we all know, JNAI has evolved over the past um three years and especially in the past six months, a lot more um capabilities have been added um into GNAI models uh like uh the OpenAIS, uh Chat GPT, the Entropics Cloud, and uh Google's Gemini. So now these models are um uh also moving towards multimodal capabilities where they are able to look at different um uh uh uh types of uh media and data as such. It could be a structured, it could be images, it could be unstructured data. All these um uh capabilities are now available within these uh JDI models. So we are leveraging these capabilities where some of these uh process-related nuances uh could be uh looked into and automated so that uh the human agents who are working on uh these processes can be freed up uh to an extent where they can put that effort towards more higher value-added uh kind of activities rather than the mundane activities of shifting through lots of documents and um determining whether a particular uh value exists between the two systems or not, just by manually looking and uh navigating to the individual fields and validating those uh data points. So that is that is where we are trying to address uh the pain points uh and provide um the enhanced capabilities of LLMs and HDC AI frameworks uh which um can help uh automate uh this process uh to an extent and uh drive uh some of the uh productivity and efficiency gains uh for the financial institutions.

SPEAKER_00 7:52

That's great. From what you're seeing day to day, what do you think the core problems are for the financial institutions that are struggling with right now when it comes to data AI?

SPEAKER_01 8:03

Right. No, certainly see, if you look at uh the uh the entire genesis of uh leveraging AI um uh depends on uh the quality of the data that you have within your organization. And uh typically what um we see is in their core banking systems and other satellite systems, the quality of data that is captured is not up to the mark. And uh when it gets into their centralized uh data store, which uh could be a uh data warehouse or a data lake that they have set up where uh most of uh uh these uh advanced analytics and machine learning kind of uh needs come into picture, it becomes a little um uh challenging in terms of achieving the end objective. So uh tip um so having good quality data which can help you build your models around, which can help you then take more informed decisions, is what uh is uh lacking in most of the organizations. Until such time, the uh financial institutions put that effort in ensuring that they have uh good quality data in their core systems, which is the customer onboarding systems or their loan origination systems, the loan management systems, the loan servicing systems, the credit underwriting uh process uh related uh satellite systems. All these uh systems they need to have good quality data which can then be used to create these um uh models uh which can then help them predict um the future outcomes and basis which they can take decisions uh in terms of their um businesses as such.

SPEAKER_00 9:41

So, in your role, where are you personally spending most of your time right now?

SPEAKER_01 9:45

Right. So, as of now, as I said, uh the complete focus is on um uh provide and creating this um uh the solution that we have and helping our clients understand uh where it would make uh sense for them to implement and try out the solutions so that um they are able to do a lot more uh with what they already have rather than uh linearly increasing the headcount as the volumes grow. Probably that may not be a scalable uh model uh here going forward. So we are trying to address that uh particular uh pain point where um we help our customers deliver more uh with less or whatever they already have. So and uh also not just uh one uh point solution. So as I mentioned, uh in terms of uh the platform QI desk that we have uh we have built and uh we are launching it um next week. We as a part of that solution, considering that it is an LLM uh-based uh model platform, it has capabilities where we could um uh deliver uh a lot more use cases. So the other use case that we are um actively looking at and which we have also developed uh some uh capabilities is how you make uh your documents um uh discoverable within the organization, right? Because typically, uh as I mentioned, the financial uh institutions and especially banking is a highly regulated uh industry. As a part of uh uh account opening uh or loan um uh borrowing um uh process, there are a lot of uh documents um uh that are exchanged uh between the borrower and the uh bank. So in a typical mortgage application, for example, uh you might have uh anywhere from 15 to 20 documents just for one application to start with, right? So imagine thousands of uh such uh loan um uh mortgage applications and then uh the corresponding uh number of documents that uh the banks and the credit unions they have in their document management system. Uh and obviously, as I said, they are regulated uh uh entities. So every six months uh OSS comes uh for uh a compliance check, right? So they probably they would uh look at uh okay, show me a mortgage application uh where which has actually been complied as per the um uh the the uh uh the fairness act. So they have to pull out uh individual applications and then show it to the uh auditor, uh saying that here is the loan application, these are the uh various attributes around the application, and these are the criteria on which the decision of underwriting was taken and the loan application either rejected or approved, right? So they have to justify this, there has to be a traceability of this justification of whether a particular loan application, how was it uh reviewed, what was the end outcome, whether it was approved or rejected, right? And if approved, on what uh criteria on what basis it was approved, and if rejected, on what basis it was rejected, right? So on um the Fairness Act um uh related um uh aspects, banks have to have uh this kind of uh uh uh traceability made available to the uh compliance offices. So imagine going through even going through five uh such uh instances, uh you would uh manually shift through all these documents and individually narrate down to where that information is and then retrievate. So this can be completely automated by leveraging uh the JNAI uh tools now. So we have built uh on our QR desk uh we have uh uh a solution which can actually help you make your documents uh discoverable just by writing a plain English language query. So all these uh documents are ingested uh within our solution, and uh the the analyst uh who is supporting the uh compliance officer can then just run a simple uh English uh query saying that show me the loan application. And probably they might have identified specific mortgage applications which are representative of uh the scenarios. So maybe they will already have uh the mortgage application number or they have the uh the name of the borrower with them and uh with uh maybe an address information. So using that query, they can easily retrieve uh the very specific uh application case and the related document. The platform has capabilities which can actually retrieve the right document. Um this is the query as well. And it can also have capabilities uh which uh that it can summarize uh the overall uh query uh outcome as well, so which can then be used to support uh all the necessary inputs that the uh compliance officer would want them to capture as well. And then additional in inputs could be with respect to also for as a uh day-to-day operations uh need, they could also look at uh what are the various covenants that they have covered as a part of their uh general um mortgage uh uh applications or the mortgage uh agreements that they already have in effect. On the commercial banking side, it could be again related to validating a few things, right? As a part of Ember, I was recently having um one prospective discussions with one of our existing clients uh where they were looking at uh is there an opportunity for them to automate uh the process of uh current process of validating all the documents that they receive from uh their commercial customers. So commercial customers are typically the organizations, right? So as a part of their loan uh applications, imagine because they would have a lot more documents compared to a retail uh loan, right? So or a consumer loan where uh it is a mortgage uh application, will have a set of documents compared to the commercial loan, where which is uh a term loan, might have a different set of documents altogether. So application form plus supporting documents. So, how do I automate uh validation, validating this entire application form to see whether I have all the relevant information provided by the commercial client or the uh the organization who is applying for the loan? Because it uh the application forms are uh slightly lengthier compared to the mortgage loan applications. So, so how do I automate this uh process? So JNEI capabilities can actually deliver that kind of uh use cases as well, where it can actually help you validate and identify what information is missing, and at the same time, draft a response to the customer asking for the missing information as well, which otherwise the human loan analyst would have to do it manually at their end. And then imagine the effort and the time that goes into analyzing all this information and then uh drafting a response. So you are talking of uh maybe days in this case because you have a huge set of documents that you have to review at the same time the application information that has been provided. And and with JNAI capabilities, all this can be ingested and uh draft can be uh made ready within minutes compared to days. So that's uh where we see a lot more uh uh effort uh will uh go in for uh the banks and the credit unions to take up uh uh such uh uh processes where they can uh move the needle in terms of uh the return of investments uh from the AI implementations. And that's where my focus area is, uh and I think it will continue, at least for the foreseeable future, where um uh it becomes uh the AI becomes uh pervasive uh in all that we do.

SPEAKER_00 17:23

Yeah, that'll save a lot of time for sure for uh organizations. Um so what do you see is gonna be the biggest challenge for you personally in driving AI adoption, especially inside large risk, risk-sensitive financial organizations?

SPEAKER_01 17:40

Right. Certainly. So there are um um uh, as I mentioned earlier, I would say about um three uh different um areas, right? So first um thing is uh in terms of uh having uh clean data. Uh because unless um all these systems um has data uh which is um uh complete and uh standardized to an extent with uh all the relevant uh metadata information as well. So metadata is nothing but it is the data about data because you will need that uh for your AI systems to learn as to what a particular attribute uh means when a business user says so, right? So there has to be a data about data as well. So all this is um necessary when uh you are looking at uh implementing an AI solution. So data, quite good quality of data, and creating some kind of uh centralized store where you could get a 360-degree view about your customers, right? Because that's where you'll be able to create uh some form of um semantic view of your uh uh customers uh so that you could uh leverage uh that uh particular view to understand your customer better and then deliver uh uh some form of um personalized um messaging or cross-sell uh opportunities which can uh give you more business uh from the same customer. The second area is uh related to how do you take care or and manage your risk? Because uh at the end of the day, when you are looking at leveraging AI, obviously there are uh set of uh behind the scenes um models uh which are um uh uh actually executing and providing uh that output to you. So, how well you uh are able to explain this um uh end outcome or the output of this uh AI models, how your processes have been documented in terms of uh what kind of data is getting inside the model, and basis that, how the model is taking uh decision. And basis that, uh, what do you see as an output? And on what basis are you relying on that output to take your uh business decisions, right? So having that entire risk uh management um uh aspect and the trust factor around the model output is what um is uh uh necessary. And uh that is uh again a challenging uh area because uh most of these models today do not have uh that transparency uh built in already. So you will need to have uh additional work carved out for those uh requirements so that you are able to completely document uh how the model behaves uh when it uh says so. And the third area is uh more in terms of um the change management and operations uh process uh within the uh financial institutions, because uh of late uh people are worried and then um uh in the sense that um AI is actually uh going to take away their jobs. But um, believe me, with whatever we have experimented and the solutions that we have delivered leveraging AI, it is not there yet to the extent uh it is being made out to be. The more uh pronounced approach uh should be in terms of how do I use AI as a co-pilot uh to improve and substantiate my work capabilities rather than uh being looked at uh as a replacement. So uh the change management um uh to adopting such AI solutions within the organization is another uh factor which is uh uh actually uh impeding uh some of the adoption related uh decisions uh within the uh financial institutions.

SPEAKER_00 21:32

Yeah, absolutely. So uh I'm curious um what trends are you seeing right now that feel real versus hype?

SPEAKER_01 21:42

Certainly. So I mean the see uh as I mentioned uh earlier as well, right? So AI is uh there and it is actually becoming uh real and real, right? But then there are use cases where um it has good adoption uh rate, uh, where it is able to deliver more um uh bang for the public. And then there are certain cases where it is not up to the buck. So obviously, if you ask me from the financial institutions' perspective, typically the setup is you have a front office, you have a middle office, you have a back office. Front office, obviously, the institutions are a little reluctant in terms of bringing in AI and exposing it to their end customers 100%. But then there are a few functions, like I mentioned about Contact Center, where some of these interactions can be done leveraging AI. But again, when you talk about customer journeys for onboarding new customers, et cetera, there are opportunities available today which can help you automate your processes leveraging AI. On the middle office side, where it is more from risk management, underwriting related decisions, et cetera, there is a lot that has been done, if you ask me. Even before Jane AI capabilities, the pure machine learning-based models have been in existence, which takes uh the underwriting decisions of uh loan approvals uh and uh determining the credit worthiness of the borrowers. And then uh accordingly, the banks uh take those uh decisions of whether uh the loan should be approved or rejected. But uh with uh Jane AI, I see there is more that can be done on rather than only limiting it to predicting uh the credit worthiness of the borrower, uh, the entire uh summary of uh the underwriting process can be documented as well, which can be used as a uh as a as a proof or uh or a credit memo uh for uh the audit purposes where it is clearly added, uh it can clearly a particular uh underwriting uh decision has been taken. So it can automate uh that part of uh document uh creation. So that's the area where I see between the middle office of the banks and credit unions, a lot of this uh work um can be uh automated and is ready for uh adoption. On the um back office side, um, as I mentioned, we which is where um our switch board uh sport is as well, uh, where uh we provide uh back office supported from uh the contact center perspective, the uh loan servicing middle, um uh middle uh uh I mean uh the mortgage uh applications uh servicing the account opening uh related documentation checks, the quality check uh processes. All these uh processes can be automated to an extent where uh AI can um uh bring in the capabilities of uh analyzing these documents, identifying what is missing, what needs to be there. And also at the same time, on the fraud side of things, it can actually create identify some of these uh frauds and also create a case around uh what uh has happened on that particular uh fraud uh case. And uh it can also create uh a response uh which is compliant uh as per the regulatory requirements. So all these uh various uh processes uh are ripe for uh AI adoption, in my view, and I see a lot of um uh focus uh from the uh banks, uh credit unions uh here in the US uh going moving towards uh that. Uh so in terms of ranking, if you ask me, I would say back office followed by middle office and then the front office. Because front office, as I said, um is a highly has a high touch point uh with the customer. So instead of uh directly exposing it to the customers, the banks uh can find a middle path where um they can actually help uh their relationship managers to use AI output, which can then be used to have conversations uh with the end customers so that they are uh more productive in their discussions and uh can do a lot more um in terms of uh the business growth uh as well.

SPEAKER_00 26:05

That's fantastic. So if we were to have this conversation again in 12 months, what would need to happen for you to feel like it was a big win?

SPEAKER_01 26:14

Certainly, I would say probably I would like to see the platforms that uh we are launching having a wider uh adoption and implementation. So I would like to see all my current uh clients uh adopting uh that particular platform where they see value for uh their spend. And uh also, I mean, just uh doing uh grazing through the uh crystal ball, maybe uh in 12 months AI uh will have additional capabilities, and rather uh the GNAI-based um uh tools, technologies with will have additional capabilities where it would so happen that um the current um core systems and the other satellite systems uh that the financial institutions use will have more embedded and pervasive AI capabilities rather than being an add-on or um or supplementary uh function to it. So it will be more pervasive, I would say, and uh uh and and all decisions will be based out of uh the data that uh uh the financial institutions capture. So it is going to become more real time rather than say, okay, I have received the application, give me a day's time and then I'll cut back to you. So rather, I would say that in 12 months' time, you will have uh most of these uh processes uh where the decisions will actually happen in real time and without uh any human intervention as such.

SPEAKER_00 27:39

Well, well, that's it for today's episode of AI Speed. A huge thank you to Parash for sharing his invaluable insights into how Quint Financial Technologies is helping financial institutions turn AI into real operational impact and for navigating the shift towards vertical AI. If you're building or leading an AI native company or a service business that uses AI under the hood and you care about revenue, adoption, and market share, make sure to subscribe to AI Speed. Learn how the best AI operators ship faster, sell smarter, and stay ahead. Thanks for listening. Until next time, keep building, keep selling, and keep moving at AI speed.