Transforming Financial Services With Generative AI (Video)

Benjamin Wootton

Benjamin Wootton

Follow me on LinkedIn
Transforming Financial Services With Generative AI (Video)

In this video, we demonstrate a real world example of how LLMs can be used in the financial services industry. Our aim was to bring to life the potential of how LLMs can be used in a real world business context.

The scenario we use is a financial services business who are making use of LLMs in four different ways;

  • Using them to query real time databases in natural language, giving them a new investigory tool;

  • Querying a knowledge base, which in this case is a set of regulatory documents stored as PDFs;

  • Generative document production, whereby we create documents for business scenarios such as compliance and regulatory filings.

  • Simple interactions with a foundational model to demonstrate how LLMs can be used for research and personalising customer communications.

Video Transcript

In this video, I'm going to talk about how you can use Generative AI in the financial services industry to make businesses more efficient and effective.

We're going to be looking at it both from the perspective of using Generative AI and large language models for knowledge discovery, but also using it to generate business assets such as documents and carrying out business processes on our behalf.

Again, I'm going to be using this example of an international payments business who are sending transactions around the world.

We can see that each of those transactions have an amount and they have a beneficiary and they're going to different countries such as Japan, China and Turkey.

We're going to look at this from the perspective of a compliance officer or a financial crime analyst who is responsible for ensuring that the business remains within the law.

I'm going to start by demonstrating a concept called Retrieval Augmented Generation, which is where we are going to ask questions about our real-time data set.

If you can imagine that the financial crime analyst, for some reason, they're digging into transactions in Australia, maybe that's been a recent focus or there's been some recent fraud there.

I've asked the question, who sent the largest value transaction to Australia?

We can see that it was a Mr Sean Cole who has sent over $200,000, so relatively large transaction.

Maybe the other country in focus is Turkey, so I've asked the same question in a slightly different way where I've said, what was the largest transaction sent to Turkey and who was a beneficiary?

Again, it was around $200,000 by coincidence and this time the beneficiary was a Mr Robert Johnson.

What we've done there is use the language model to ask questions of our databases and the state of our world, giving us a new investigatory tool.

Next, I'm going to demonstrate the concept of knowledge bases.

Here, maybe as an analyst, I'm not sure what are the limits when I need to report payments to regulatory authorities.

I'm going to ask a question, what value of transactions need to be reported when the beneficiary country is Turkey?

What the LLM has done here is gone away and queried our proprietary knowledge base and informed us that transactions over $100,000 to Turkey need to be reported.

Whereas if we send in a transaction to Australia, it's actually $200,000 is the limit.

What we're doing is querying this document, this PDF in the background and asking questions of it in natural language.

This is all about knowledge discovery and asking questions about how proprietary data sets.

Now, if you imagine that we have identified that both of these transactions are over the limits by which we have to report to the regulatory agencies.

What we're moving into now is a generative scenario.

I'm going to ask the language model to raise a suspicious transaction report.

That is maybe a combination of characteristics about the person and the size of the transaction.

The LLM has responded and asked the question, "Give me some details about the transaction, including the customer name."

I'm going to reply in very natural, discursive English.

I'm saying the customer name is Sean Cole.

The reason that we would like to make this suspicious transaction request is that we do not hold details about his income levels, which is maybe something we found offline.

I've raised that request in very quick, discursive, natural, plain English.

The LLM is going to process that.

In the background, it has gone away and generated a document in the form of a PDF which has been filed with the regulatory authority in the background.

This is something which businesses actually need to do.

If you have transactions of a certain type, then I have to tell certain people about them.

It's just an example of automating the document generation and the business process and the regulatory compliance.

I'm going to do that a second time, ask the question in a slightly different way.

Raise a suspicious transaction report.

The customer name is Robert Johnson.

We're raising it because it breaches our internal controls whereby we cannot send payments of more than $200,000 to Turkey.

This is very fast dynamic where we're asking questions about the database.

We're consulting our knowledge bases in natural language.

Then we are completing business processes here, raising a suspicious transaction report with a regulatory authority.

If we now go and check in our AWS account, we can see that these PDFs have been created in response to those requests which we have been sending via that chat interface.

Here is the transaction report for Sean Cole.

We've illustrated that we are suspicious about this transaction because we do not have any documentation about his source of wealth.

If we go back to the other PDFs, then this will download it, open the document.

Here is the suspicious transaction report about Robert Johnson.

We are raising the request because it has breached our internal limits.

What the system has done is taken that very conversational, natural, casual request which the agent has raised and it has translated it into languages appropriate for a compliance officer or a regulatory agency.

The final simple use case of Generative AI is here we're going to have a simple interaction with a foundational model.

I'm just going to say, please can you raise a letter to Mr.

Sean Cole asking him about the details of his source of wealth, explain what it is, a regulatory requirement and he has to respond in 72 hours.

Here we're just using it in a very chat GPT style where we are asking the LLM to generate content for us.

I think this is an important tool as well, using it for automating customer communications, carrying out research and things like that.

Being able to move between these different modes is particularly powerful where I can use this natural language interface to query my databases, query my knowledge bases, generate and personalise customer communications and then automate business processes using agents.

Hopefully that all made sense and hopefully you can see the potential for how that could make the typical financial services business much more efficient, effective, it could improve customer service and help with areas like regulatory compliance.

If you'd like to know more or if you can see any potential use cases for us, do reach out, it would be great to have a chat about this technology and approach.

Thanks very much.

Join our mailing list for regular insights:

We help financial services businesses build and run advanced data, analytics and AI capabilities based on modern cloud-native technology.

© 2024 Ensemble. All Rights Reserved.