Learn How University Credit Union is Transforming Banking in the Age of AI

Really appreciate everybody taking the time to log on here with the interface team. We’re really excited to spend time with you guys today. There’s a couple of things that we’re really looking forward to not only informing you around AI and what interface is doing in this space, but we actually have a few things that we’re really excited to show you a little bit later in the call today.

So to get started, what are we going to cover? So we’ll go through some introductions.

We wanted to spend at least just a few minutes and do quick deep dive on AI and have a conversation around what is generative AI look like today in the space? What is it? What is the technology?

Why is it even important? Why are we even here to talk to each other for instance? Then we’ll dive down a little bit deeper. We’ll talk a little bit about foundational AI models we want to talk a little bit about domain specific versus larger based systems. Really quickly talk about what some of the latest advancements are.

You know, cover, you know, what’s in it for you for instance and what some of the bold projections are moving forward. Right? How is AI really going to shape what is it that we’re planning on doing, right?

So we will then show you some surprises and unveil a little bit around what interface AI is doing today and talk a little bit about some of the new products that we’re going to be bringing to the space and then we will follow-up with some Q and A. As a reminder, as part of this, there is a chat window that you can leverage in order to ask some questions we will compile those and get back to them towards the tail end of the presentation.

Alright. So who will be speaking today? Myself, Brian. I’m relatively new here to interface. I’m the director here, the solutions and sales engineering team. And I am joined by Srinivas and Jay, who is our founder and CEO.

Srinivas has had a ton of experience working in AI, working at Microsoft for instance and has spent a good chunk of his life working in the credit union space. So We bring a lot of experience or working with both large and small customers and trying to make sure that our customers get the most value out of technology that we sell.

Also joined with us is from the University Credit Union down in LA, we have Estella Nagahashi. She is the EVP and COO of University Credit Union. She’s been a great partner of ours.

I think, you know, I had a quick conversation with her and really loves working with interface. She has over twenty years of experience working in the retail banking operations and services space, working with customer relations, lending operations.

And she’s worked on a number of projects and initiatives.

From the very beginning, through the execution to delivery and is already starting to see a significant amount of value here with interface.

So quick stop. Let’s talk about interface really quickly. We are the ones bringing this webinar to you today.

Now we’ve been in the space for about eight years now. We’re headquartered here in the Bay Area.

We do bring a wealth of existing knowledge in the space from a lot of the large players from Salesforce and Apple and Microsoft.

We’ve already got wide range of customers in their credit union space and the community — excuse me, community bank space.

And what we’ve tried to do is take advantage of natural language engines and AI to increase member experience typically by reducing the amount of time it takes in order to get just something done or to get to a person to speak to them to get something done.

Now, we’ve been endorsed by the Kuna strategic services and Allied Solutions.

Uniquely in this space, we are actually starting to see over a million requests actually a day now, helping individuals answer different questions, rich questions, simple questions.

We do this using an AI solution that is not just targeted to one simple channel like chat for instance.

We support things from chat to phone calls to web.

We assist in helping you with member issues and queries. But we also have products that assist your employees internally as well. All in all, we have about sixty different offerings to help you make sure that you are leveraging AI where it makes the most sense in your business today.

Right. So, Brian, appreciate the introduction.

So, we are here today. This is free.

We are here today kind of talking about some of the most, like, important technology breakthroughs in our lifetime right, generative AI. We’ve been hearing that across all the news outlets we rely on, and it’s pretty exciting what’s really happening out there. And we wanted it today in this presentation, kind of tell you what generative AI is all about.

What made it happen, and then kind of show how it’s gonna make a huge difference to banking, and then some products and how interfaces aligned with that generative AI possibilities to create more value for our customers and future customers, right? So, you know, Genative AI is grown, and it’s a league of his own. So if if you look at this slide here real quick, Chad GPT is a math or hundred million users, the fastest compared to any other technology ever existed.

Right? So, you know, it took two months which attribute to get one hundred million users.

In comparison, if you think about Apple iPhone, which we all thought it’s a breakthrough innovation in two thousand and eight or seven or nine, you know, it it took fourteen years for it to get hundred million users. Right? So that’s like a a little bit of a comparison. So there is something really exciting happening here, let’s dig deeper, understand what’s really going on, why is it crazed.

So Now, there are three concepts we often, you know, read. Whenever we talk about generative AI, there is generative AI itself.

And there is other terms like large language models, and there is another term like GPD. Right? So what is you know, my goal is to gonna kind of give you an idea about what each of these are and how are they related?

So you understand the terminologies like as you kind of progress through this journey of generative AI in the next decade or so, you’re more grounded on what is the possibility of these technologies, where are they heading and whatnot. So let’s visit let’s visit each one of them real quick. So, you know, generative AI, what is generative AI? So if you are like me who understands like top down, so you got mission learning, which we all understand, which is kind of a, you know, it’s more like a a set of It’s like a set of algorithms and techniques that help you train machines to do certain things that are not really exactly programmed for.

You know, it’s a machine learning kind of learns through some of the techniques and programs, right? So there is a subfield in machine learning called deep learning. Right? So deep learning is basically inspired out of architected, inspired out of like a human brain.

Right? So the name itself comes from deep learning. It has multiple layers, it understands and a problem from top to down, kind of breaking down into you know, iteratively breaking down to smaller problems and understanding them better. So, generative AI is a a class of kind of set of models and techniques that are largely based on deep learning architecture, which is in turn is an architecture based on how a human brain works, right?

And specifically, these subset of models and techniques, generative AI, that’s based off of deep learning architecture mostly, specializes in creating original content. Like a mind blowing content, like all the things we have been playing around with transmitting, you know, essays or there are other tools that could help you create really stunning images on the fly and things like that. But, generally as a set of models that kind of focus on generating original new content that is based on underlying data set, right?

Based on the trained data set, it understands the features and it uses them to create completely original new content. Right? That’s the class of models that does that is called Genative AI.

Large language models is a term used, which is a subset of these models that are part of generative that specifically focus on language generation. Like for example, you know, if you’re in GPD, asking to write an essay or a poem, and things like that, it comes back with a creative writing.

That is a subset of models, but that’s still generally AI, which is a subset of models You call it large language models because they’re focused on, you know, kind of a language generation. And then there is GPD.

What is GPD and comparison to JENDA and large development more? GPD is a specific proprietary implementation of a pre trained model. It’s like a system that is pre trained and offered specifically by open AI that you could kind of use out of the box, and and and if you’re familiar with chat GPD, chat GPD is a kind of a variation on top of GPD to make it more conversational.

Right? But GPD is is kind of a pre trained system, right, while LLMs are like models and techniques and architecture, GBD is a specific pre trained system that you could use, right? So that’s kind of the difference of these terminologies, right? So You know, real quick, what is making genetic AI possible now?

Right? There any such huge innovation doesn’t happen with just one breakthrough. Right? It happens through multiple breakthroughs, you know, kind of happening around the same time.

Right? So one of the notable mentions of the breakthrough that enabled GPD is a research paper published by team of researchers in Google in twenty seventeen titled Attention is all you need. You know, that is one of the notable reasons of the breakthrough of how GPUs are so good today.

And Tinction is all you need. Sometimes, my wife says the same to me too. So Tinction is all you need. So, you know, attention really means here the focus, like you’re placing variable weights on wherever the words appear in the sentence, right?

So, because based on where a word appearance is sentence, the meaning changed. Right? So that’s what attention here refers to. And this is also not a new mechanism.

This also existed in the past But what this research has found out in twenty seventeen is, this mechanism was applied along with many other techniques. But they realized this is the most important technique, right? In twenty seventeen, they proposed in a paper saying attention is all you need. So kind of making a proposal, everything else is not that relevant or rather adding incremental value, but attention is adding the most value. So The impact of that is the performance of the system increased massively. So, the improved training efficiency Now AI is able to learn faster because it is no longer processing word by word to understand them. It is consuming the whole sentence.

Enabling it to learn from a lot of data that is available. As I said, many breakthroughs like this happen because of lot of other smaller breakthroughs are happening. They all coincide around the same time. The other breakthroughs including our smartphones.

We got lot of data now that is available for training and the Internet, and then the GPU is the power of processing itself has improved. Combined with all of this now, there was a right kind of a a intersection of all of these kind of, you know, GPU took off. Right? So the notable mention is kind of transformer’s model.

And there is another research that is happening that also led to creating stunning high resolution images we see. Like, you know, you might have seen all over the internet saying, show me an astronomer sitting on a horse. And like, you know, this shows you very high pretty like stunning images. This is based on an other innovation called stable diffusion.

This is basically marries the best of the techniques of GANs, which is generally our dorsal networks, and transformers, they come together to create this stunning images. Right? So lifelike images. So genetic AI in general can generate original content in text, audio like music or video or images, right?

And that’s kind of, you know, making us think you know, the AI is better than humans than things like that. Right? So the architecture of this system At a high level is pretty simple as we dig deeper, you kind of understand it. It’s actually quite complex, but it’s still quite easy compared to previously how hard it is to build such an AI system. So this is a high level architecture. You got hardware at the bottom, which we talked about. One of the innovation that led to the growth of GPUs is like the GPUs.

Which is largely powered by NVIDIA, they recently become a trillion dollar company. There is a reason for that. This is it. And so that’s the computer hardware at the bottom.

Then you have cloud platforms like AWS, GCP or Azure, you know, where these models are hosted. There are different approaches you could take to have these foundational models. Foundation models are like the best examples, as you can see here is like GPT. This is a pre trained system, that can understand, you know, generally all the concepts in the world, right?

So it’s a pretrained system is — because it is offered by — it’s like a proprietary offering by open AI, it’s called closed source Foundational model, right? The reason foundation model means is because it largely understands the world that we’re leaving, right? So That’s why it’s called foundation model, and it’s closed source because it’s offered by open AI. But there are a lot of open source foundation models as well, like stable diffusion we talked about.

Which is really the one led to huge innovation and creating stunning images and whatnot. So that gets hosted in your cloud and then the apps, which is, you know, b to b, b to c apps like interface kind of build on top of this, right here. But there is another approach, which is something What interface is kind of gone towards is building end to end. You know, where we take some of the some of these closed source models, and then on top of it, we build our own proprietary models and then build apps.

Gonna talk about that in a future in a bit. But that’s, like, the high level architecture, how all of this come together. Right? So if you take a step deeper in terms of what all needs to happen, even when you’re using an out of the box, like a closed source foundation model like GPT or something.

There’s a lot of work needs to be done to make it usable. There’s a bunch of questions I often get from our customers and prospects saying, hey, you know what? We got this GB API, can we plug it in and then just build a chatbot on our own? You know, absolutely yes, if you are building online mobile banking in house, and, you know, you could scale that up next level in terms of technical talent required, you could do that.

Right? So But if you’re not building online mold banking, this is probably like I would say ten x harder. Right? So But it is possible.

Right? So, you know, the there are three layers in how this technology stack work. Now, we kind of double clicking on how these apps get architected on top of foundational models. I just go back real quick, we are looking at this box here in how that all the things that needs to happen.

So one of the famous approaches in kind of training and building apps on foundation models is what we call in context prompting. Right? So, basically, GPUs and today, at max can only take you know, fifty pages of information to to learn from it. Right?

So you have foundational model, which understands the whole world, And you have to fine tune it to, let’s say, banking specific model, then you got about about fifty pages worth of content you can give.

For it to understand and then process the request. Like, for example, that’s what we call, you know, prompt, like a few short prompting. Where you have using a foundation model at the bottom, and then you have room up to give fifty pages of content on top of it. And you prep the system using that.

You’re not changing the model, you’re prepping the system, and then you send the customer query into that system. Right? Then it’ll understand because the fifty page of prompting you did, perhaps the system to understand banking. Right?

So then you pass the customer request. Right? So And because it’s only fifty page, there is a lot of optimization you can do.

Or rather you need to do, right, which is based on embeddings, vector databases, which makes the data that goes into the system more optimized and and more compressed in a way, so you can have more data sent to prep the system before it process a request. So there are a lot of nuances. We’re not gonna go into the detail. The slides will be available for you to dig deeper and things like that.

So what are the benefits of generative AI? Right? So the benefits of generative AI, as we kind of spoke briefly about this, is it’s large federated model, which understands, you know, kind of the the world in general, and it’s crowdsourced, and there is a lot of human in the loop going on to train them better to make sure it is not really hurting anyone’s beliefs, and and, you know, thoughts. Right?

So, that’s kind of the benefit of that. It understands overall model. And it is also really good in generating original content. Right?

And it’s also getting smarter over a period of time, right? So, and then the drawbacks of the system are it is not domain specific at all, it lacks deeper understanding the domain. But for it to train as well, there’s a limitation.

At least today, what we call context window, which is the fifty pages I was talking about, there’s a limitation on how much you can train about the specific domain. And that is there’s a lot of research going on to improve that, you know, but also the technique of in context prompting itself is kind of broken. Because it’s really not changing the foundational model for it to learn the domain, but rather it is just prepping the system, right? So the other things are, the genetic AI is today is not factual.

So if you use strategy video or something, if you sit right next to an expert who understands a domain really deeply and then have him or her ask questions, they immediately would identify that it is actually making plausible information, but really not accurate or factual, right? That’s kind of a drawback too. And also, there’s a lot of compliance risk Right? With respect to how PR data gets handled, and, you know, how the compliance could handle, and how the security can handle, and things like that.

So those are some of the drawbacks.

So, yeah, I wanna show you an example of foundation model being used as is. Right? So let’s say you’re using foundational model as these, you say, tell me about Apple. It says Apple is a fruit grown on a tree.

Right? But a domain specific model that is trained in some way either through fine tuning or in context prompting, you know, would say something like this. Tell me about Apple. It says, I found three transactions from Apple Store on your credit card.

And that’s kind of the difference.

The cleaning of that And is There’s a lot of complexity as we spoke. So But there is, there is a Do we discard the foundational model and build everything on our own? Plancer is no. That’s probably not a good idea. So the best approach is to use foundational model as a foundation.

Probably just the reason they named it. And you verticalize on top of it by creating domain specific data. And the way you could do it in multiple ways, then we’ve taken a very unique approach doing that, which is real quick to store you.

So we have built a domain specific generating a a model on top of the foundational a model.

Right? So it’s like a layered, you know, generative AI. So the one at the bottom largely understands the world around us, And then we use a small domain specific model, and we trained that. On top of the foundation model so that it understands banking really well, and it has no limitations of fifty pages, right?

You could really understand very thoroughly the whole domain. Right? So And then, what is this domain specific model on top of foundation model is doing? What all we trained in that, so it’s about accuracy.

First of all, we cannot, as a financial industry, we cannot afford hallucination, we need accuracy. So we have fine tuned these domain specific models to provide accurate domain specific No, banking domain specific responses.

And they’re also fast. You know, one of the biggest challenge of foundation models alone, if it is used, is it’s quite slow and expensive. We’ve kind of fine tuned to make it faster with the layer approach with a lot of caching and whatnot. And it is out of the box now right, with the foundation model and banking specific domain model, you can take it out of the box and use it.

And it is compliant. With the consumer production loss. And it also is scalable, unlike foundation model today, is really challenging. I’ll give an example.

If you’re submitting, you know, if you have, like, a thousand documents, like, let’s say your policy and procedures, you got all of this. If you we want to prep the system with in context prompting with foundational model with this thousand document. Each API request will call you several hundreds of dollars.

Right? But with the domain specific model, we’ve kind of removed the constraints of how much documents or information could be fed, and it could be fast as well as cheap. Right? So that’s kind of the fine tuning that happen.

It also ensures for our prevention, system reliability, which is also some of the challenges we have with the foundation model out of the box. So what are the benefits of this for financial institution? Why do you why do you need to care about this? Right?

So Generative AI may be the biggest innovation in the history so far, and it is very transformation of financial institutions specifically.

There are five areas where it’s gonna make a massive difference. Number one is you’re gonna reimagine customer and employee experience, like never seen before. We’re going to show some, you know, demos today, and hopefully, it’ll blow your mind away. So, there is nothing like that you guys seen. Right? So, it also ensures cost efficient operations.

With automation and all of that. It also ensures better compliance. We’re going to visit each use case with examples to see how these all could be accomplished, and then it also helps with improved risk, and then better forecasting, SBB, you know, everything that is happening. It helps with better this might as well end up forecasting.

So, I want to take a couple of examples here. So to kind of talk about how you could reimagine customer employee experience generated AI. So, historically, our traditional approach is when you are processing your loan, you know, bank evaluates based on present parameters, decision making manual repayment is managed by the customer.

But we generated the AI.

AI would be capable enough to understand the nuanced financial and personal data of the member or the customer. And make personalized recommendations and proactively manage repayment by monitoring their current financial condition.

Right? So, you know, that’s what we all need, like a peace of mind. We shouldn’t As consumers, we shouldn’t be worried about, okay, am I going to make sure there is enough cash to make certain mortgage payments or car payments.

The AI will take off those stress off of you as customers and as your customers are members, and then helps them manage their finances and personal finances a lot better, right? So, And then I wanna quickly show the second use case, which is cost efficient operation.

So here, let’s say, you know, your loan officer today in the traditional sense like loan officer gather, you know, data manually from various systems to ensure you know, pull the application together as well as make sure it’s regularly compliant.

It’s a very time consuming labor intensive process. With generally AI, with a train on the historical data, you can instantly provide You can instantly get responses that the AI would have learned off of aggregate data, that will help your loan officers to ensure there is any compliance risk to be addressed and streamlining the process, improving the overall efficiency of turning out loans, right? So I’m sure you’re still in compliant, right? So that would significantly prod an impact of top line for financial institutions.

And then the third use case is fraud detection.

You know, this is the place that could be, you know, customer experience and employee experience and fraud could be the biggest innovation and biggest bug for money. So here, most of the fraud detection system today are preset rules, right?

Creating a lot of false positives, you know, kind of leading to a lot of resources spend, you know, there are a lot of research that talks about ten to fifteen percent of employees in a bank or a credit union are just focused on magic fraud, and managing false positives. Right? So And there’s only In spite of all of this struggle, you just end up being effective three percent with generative AI, it’ll learn these patterns and continuously evolve and identify these anomalies and help fighting fraud a lot more easier, more autonomous way. Right?

So that’s the third use case. And the fourth one is managing the risk. So, you know, we’ve all gone through recent struggle of what happened with SBB and other banks is, monitoring risk is so important because there is And so complex, because there is insights everywhere. There is unstructured data everywhere.

Someone has to kind of analyze all of that to proactively predict the risk that may be coming, right? So, you know, Genative Air play a huge role in managing risk for financial institution, and you know, leading to increased profits and, you know, better balanced sheet and whatnot.

And then the last use case is finance team operations in financial institutions, right? So, finance team, like, you know, go through quite a lot of labor intensive, like manual process forecasting reporting, interpreting tax codes and whatnot.

With general AI, one of the things that it would do really well is not only look at detect patterns and streamlined reporting, but it will be really good in understanding and interpreting tax codes and compliance.

Right? And then give you a suggestion in size what needs to be done. So these are, like, five areas you’re going to see huge applications that are leading to quite a bit of ROI and whatnot.

So some board predictions on our end. Right? So, you know, AI is gonna be enterprise wide.

As in every employee and a customer of yours is gonna use AI in some form or capacity gonna be enterprise wired across banks and credit unions. And it is going to help your customers and employee complete tasks and find information ten x faster.

Right? Are we going to show some demos to kind of ground you on reality of that is possible?

And we, as a, you know, AI company being around for eight years, we’ve already seen significant benefit. That we’re able to deliver to our customers with, you know, in the range of forty to fifty to sixty percent call automation, chat automation, and things like that. With generative AI, that is going to go up to ninety five percent right? And we have a lot of things that are built to enable that, we’re going to show you like live demonstration of how you can get there. And then, you know, everything your employee and customer want to achieve, either will be augmented by AI or automated by AI.

So that’s the future we’re going to live in in the next three or five years. Right? And so prior interface is kind of the leader in making some of this happen. So now the question is, like, we understand broadly how genetic AI could be very impactful, what is interface AI doing?

And how is it aligning with these predictions and you know, and whatnot. Right? So I want to show how interface AI’s products is going to transform your customer employee experience for good, like with generative AI. So the fundamental concept before we go down that path is to understand What is really happening?

There is a with generative AI, there is a paradigm shift in how customers and employees interact with machines.

I’ll give you an example. With software, like online banking, mobile banking, or IVR, what happens? Let’s say you wanna send money. Right? So you’d say, you would go to your online banking, click on money transfer, that’s the blue box, and then you click on from account, that’s the yellow box, and then you click on two account, that’s the blue box again, and then you click on amount, and then that’s the yellow box again. So you’re going through points and clicks a task oriented approach completing one step at a time to complete a task. It’s the same for your employees, but different systems were the same for your customers.

But there’s a pattern shift in how it’s gonna happen with generic AI. Which is, you’re gonna say, I wanna send one hundred dollars from my checking account to my mom now.

Right? That’s that’s blue box and yellow box. And then AI does all of the steps, and you just confirm and get it done.

That’s goal oriented approach. The way we interact with machines is fundamentally going to change from task oriented approach to goal oriented approach. We tell the goal to AI, AI figures it out how to get there. We’re not going to tell it every step of the way how to do it. Right? There’s a massive shipping how we interact with computers. With that being said, so this is our past with online mobile banking you know, SMS systems, our customers and members have various channels that connect with various back end systems, right?

And on the employee side, it’s a mess. You know, your call center staff could probably have fifteen to twenty systems open on the screen.

Juggling between all of them just to process a customer member request, right? So this is our past. There’s no AR here. I’ll show through the years of eight years what we have achieved, which is kind of present. Right? The present is with some of our customers who adopted multiple solution, AI has become pervasive.

As in, you still have customers reaching out to you in multiple channels, but there is a layer you introduced, which is kind of the AI brain or system of intelligence you introduced in your stack, where it is automating those interactions or providing upsell cross sell, providing financial insights connecting to the back end system. This is present, and which we kind of made it happen for the last eight years for many of many of you, the customers on the call and and prospects, hopefully very soon. The same for employees that is like this intelligence layer.

But with genetic AI, the future will be very different.

Right? So let me introduce you to what we call interactive banking intelligence.

So where your customers is going to have just one multimodal chat GPT like interface.

And they will get everything done through that. Same with your employees.

They’re not going to have fifteen, twenty systems open, they’re gonna have one chat GPU like interface.

Right? And they log in, and they chat in natural language or speak, And this interface is gonna talk to all the back end system they used to have open and complete task on behalf of them.

Right? That’s the future. We call it interactive banking intelligence.

So today, we have demos for both of them and these are products that are already live powered by generative AI across our customers.

They’re gonna envealed two products today. One is for the customers and one is for the employees. We’re gonna start with the employee one. Right? And that is frontline assistant.

Print line assistant is some of you may have heard about it. This is two dot o. Print line assistant is a generator AI for multimodal. What is multimodal mean?

Is It takes inputs and outputs in multiple farms, text, voice, video inputs, and then outputs also video voice, and text, and anything it could be. Like, multimodal means it uses multiple ways of interactions to get things done. And it’s gonna be like a tragedy like interface, and there are three main features we’re gonna demo very soon. One is the bull pin, which is the moment if your member is talking to AI on the call, if AI didn’t understand and hands it off to your staff, this front line assistant that is used by a front line staff is gonna automatically open a policy or procedure based on the question they asked to the AI they were on the call.

Your customer was on the call. That’s called bulletin. And, you know, very soon brand new flow. And then there’s nexus, which is like one stock shop you understand three sixty degree view of your members and customers calling for your staff, and there’s co pilot features, right, where it goes back to our prediction, which is AI is going to argument or automate for your staff.

So I’m going to introduce Brian here real quick to take do the demo, Brian.

Thanks, Ray. Really appreciate it. And and, you know, we should probably start with the istla’s asking some of the questions, Brandon, switch over. So, Estela, I know you’re starting with you already started friendly assistant implementation at UCU.

So can you please talk about as a little bit what is the vision for front line assistant here, and then, you know, what is the specific use case and what are we expecting next one year?

Yeah. Sure. I think all the things that your presentation already kinda covered it, but let me just to reiterate and then just to be highlight a couple of things. So at UCU, we wanted to restart this journey.

We had a successful launch with our membership. So we wanted to bring that similar tool to for our internal team members. So frontline assistance, our goal is to really provide access to a knowledge base tool in in one tool, like you mentioned. Because we realized that our contact center is using really eleven to fifteen different tools. So we wanted to just really consolidate it And that was a part of our focus on investing in people by providing augmented intelligence. You know, it’s really aim to enhance and improve.

Their intelligence in the weather than replacing it. So that our team member, our staff doesn’t feel like, oh, well, this bot’s gonna replace me. They’re really there to enhance them and then improve our knowledge, and so that the way that we support our members. And then to your point, it’s speed and quality. We need to get get the information quickly and fast to our members, but really just that empowerment to find that solution without having to tap in people. We realized during COVID, we no longer had neighbors to tap into, but we had a box. So that was the the the great savings for us.

Some of our use case it’s really not it is a knowledge building tool for our retail team. It’s where we are starting our journey. So it helps our contact centers, as well as our branch employees to find policies and procedures to any knowledge that they need to successfully serve our members. But not only that, we are expecting our team to be prescriptive and predictive tostring you mentioned again, you know, not only providing the information and support, for the initial request from our members, but predict, proactively anticipating next question and fulfilling that. Right? So we talked about those. And then, ultimately, we want to be able to use the tool for all areas of the credit union to to to encourage a self serve all across functional team knowledge.

And expected success in a couple of years is really just the reduction of the support team just waiting for somebody to answer the call or in that whole time, the anticipation. When you are when you place member on call. That ten seconds, twenty seconds seems like a lifetime if somebody don’t answer that. So through the our through our bot, our ten team members can get the faster learning up to speed accuracy. Right? That’ll have to go through ten different policies and procedures to figure out which one is the most most up to date information. And then the savings gonna also come from the one call resolution, that which will increase our employees job confidence because they can get to our member solution right away, and the member satisfaction as with all.

And, you know, lastly, it’s the data driven evaluation. We get ton of data from our bot, from so we know where the learning gap is, where our procedure content content gap is, and what questions are our members asking. What what are they asking for, what, how to prepare for so that the data evaluation and the decision making is what we’re looking forward to. So it’s thank you for sharing those insights and the vision of the possible environment system.

Again, you know, who would like to now show your demo of use industry credit unions, French line assistant, again, this is an AI system that is set up for Frontland established is a call center branch tab within US Credit Union. This is industry first, generated with AI powered, multimodal, chatty bitty, like AI and system, Right? So we’re very excited to kind of show you this to you real quick. Let me switch my screens here.

K.

Hopefully, you can see my screen now.

I’m gonna quickly log in here.

The first thing you’ll notice as soon as you log in, we talked about a few features.

Mulpan mixes and co pilot. I’m gonna talk through each one of them here in the demo. So first thing you notice is like a dashboard which has a list of all the calls that are active. So as a call center staff, you can open one of the calls that you picked up, you’ve masked the caller ID, there’s also a way to automatically select in the system to configure, but right now let’s say you manually selecting the call you picked up as an agent.

As soon as you open, the screen you see is what we call assist. Right? So, this is this one stock shop for your call center staff to get all the insights about, you know, what is position history so far with AI until the call got handed off to STAab. What is the authentication statuses?

The AI already authenticated the color or not. And then you’ve got AR system on the left. So one of the things you’ve noticed here is within the nexus, there is this, the feature called BullPAN, where if you look at the convolution history, it seems like someone called in to the AI, and I was asking about, you know, issue with our online banking. The AI was not able to assist.

It just forwarded the call what happened is the seamless handoff between AI and the staff, which is, in this case, it calls into staff, it proactively queried based on question AI, didn’t understand and found the policy process document and kept it ready for the positive stamp. All they have to do now is to click on this, and they get the specific text highlighted in a specific document, in a specific page that shows how to address this question. Right? So we’re gonna look at a few use cases here real quick.

So, so, you know, I’m gonna ask you a follow-up question.

Let’s say reset o l b password.

And every time you do a follow-up question. It searches within the same document.

Try to see if there is a relevant response and shows it up, again highlights the specific part of the document, and exactly the text you should be looking at. So, you know, as you notice, that bulping like you must hand off from AI to staff, seamlessly where it opened up the policy procedure, and then the staff could ask follow-up questions to get right documents popped up. This is not just limited to, you know, just seeking information for the call center staff. The AI here could also take action.

If there is any question that you could ask here that, you know, that could be facilitated through a backend integration, through an API or through an RPA, the AI actually gets it done. So that’s how you could have this as, like, this one single stop chatty bitty like AI assistant that could replace all the tools behind the scene, right? So any question you ask that requires intervention of those tools, the AI, the friendly assistant AI itself is going to speak to those tools behind the scene through RFP or API facilitate those interactions by mimicking user inputs, right? So, that’s kind of how friendly assistant could be used for a call center agent.

Now we’re gonna switch here real quick. Let’s say if it’s a brand staff, right? So through the branch, there are a member of a customer walked in and you want to inquire some specific questions related to that. So, like, for example, let’s say you want to understand procedure related to large cash order.

Again, AI is looking through all the information available, through documents, and integrations, and through RFPF, getting out, right way to address this question. It seems like there is a process associated with it, so you can click and open, and then again, it brings up the right document highlighting the text. So at this point, if you have any new question, so the best is to click on start a new search and then ask for the follow-up question. This time, I’m going to ask how do I cache non member checks?

In ASetching through all documents and integrations to find any way to facilitate that.

And soon should be able to provide a response with a document with steps and whatnot, so you can click on it, open the specific document in the steps, it summarized everything for you here.

You know, know, frontline assistant is compatible to integrate with your RPA system. If you don’t have one, it comes integrated with it. So what happens with RPA system is all these steps that you’ve seen the AI is gonna do it. So mimic exactly the clicks and, you know, you know, navigation that your staff would do, mimic all all of that, and take care of the whole request automatically.

So, that is a possibility with front end assistant as well. So, that way it just becomes one stop shop, charge a pretty like interface.

So, and if you want to manage all the integrations and documents, you go to manage documents, You can upload documents and integrations as well could be added to ensure it can tap into all the back end systems as well to complete a task or find information.

Right? So, you know, just going back here, that’s a front line assistant. Right? So, now, I would like to introduce you to a brand new product. There’s nothing like that exists.

In the industry today, and we’re the first one to do it, and we’ll call it SPEER.

SPEER is a generate AI app, our multimodal chat GPD like A Assistant for your customers and members, right, where the front line assistant was for your employees, particularly front line staff.

And spear, there’s nothing like this exist today. Some of the features here are, it’s multimodal, so you can interact with it through chat voice and, you know, visual elements like clicks and you’ll have widgets and all to interact through mouse as well as audio, and it also responds back to you with audio or like a generated video on the fly and whatnot including text, of course, That’s what we call multimodal, and there is plugins.

You know, specifically, this connects with the external systems and kind of marketplace.

Like you want to integrate with meridian link for processing a loan origination application, or it could be on an account opening and things like that. So there’s a suite of plugins we’re gonna show in the demo how they all come together to complete a task. And then in context Mestro is this is kind of the AI in line where it is kind of a a conductor, kind of figuring out through every conversation how to maximize the outcome for the members, as well as for the financial institution, kind of figuring out right financial insights to be provided, and doing upsell cross sell importing real time guidance and helping through the transaction or loan application, that’s kind of the third feature. And the fourth feature is what we call co pilot, where the AI co pilots along with the customer and member to complete a journey, right, successfully.

So let us show this product in action. Again, Spear is industry first, generative AI powered multimodal tagidity like assistant for customers. There’s nothing like this exists. This is going to replace online mobile banking. It’s kind of you know, AI powered banking in many ways.

So let me show you a demo. Give me one second.

I need to log into a remote desktop, where we only have access through that.

I think real quick.

Great. Let me go share my screen.

Hopefully, you can see SPEAR.

In action.

So here, you know, as you can see, this is a replacement to your website, online all banking, and it brings all other third party systems like online account opening or mortgage obligations or live chat system, all behind the scene hides all the complexity and gives you the simple you know, multimodal, you know, as in, there’s UI elements, and and and chat, you can chat and, as well as speak to it. Right? So we’re gonna go through a a demo showcasing you the entire life cycle of a customer from all the way finding products six is really applying to banking and then eventually achieving getting air with financial wellness and with any financial hardship, right? So let me go and get started here real quick. So I’m gonna start by quickly asking to apply for a credit card.

So it’s asking what credit card I’m looking at interested in cash back credit card.

So it shows me, again, as you can notice, it’s not just chat. It has u elements to render information that is rich, and also, which shows subsequently how you can even interact with it. So it shows me benefits and things like that, and I’m gonna ask you a quick Is there any offers for the college student, university credit union source students, so we’re gonna quickly check their any offers for college students.

It seems like there is, and there are a couple of offers here, and I would like to you know, I would like to compare those cards. So let’s say, yes. Okay. I see side by side, as you can see, kind of shows you two different cards that are asked to check out and that offers the students here real quick. And you know, that is helpful. And I could also ask specifically what are the main differences, if I want further explanation on that, you know, what are the differences in these cards.

Nard specifically summarizes the difference for me, so I don’t have to also waste time looking at all the details on the card there, and it asked me which one you prefer. So it seems like I am excited about cash back, I’m going to say cash back, credit card.

Now it’s automatically connecting to Meridian again, it’s all happening in one window, not switching between different applications, and it already started application process for me. So I’m gonna enter my name, Last name here, quick date of birth.

Any new income, Security number.

As you can see, as soon as I was entering social security number, it was all masked.

For a number, It gives me a summary asking if everything is accurate.

I’d say, yes, And this is the in context, mushroom action, right? So where I apply for credit card, and it it’s coming back, doing a little bit of an upsell saying I can get one percent API by default, with this credit card, you know, and you can earn up to five percent at least on APO and opening a checking account. And do I wish to continue and of course that’s a killer deal, and I’m gonna say yes.

So it seems like it also responded and gave me an idea about if there is any peace associated with it, which is really helpful.

So I’m going to quickly go ahead and confirm I wanna apply.

So now it is again connecting my reading link, because I’ve already applied for Cretaitis, fetching all the information back from it, and prefilling whatever is required for checking account, only asking questions that, you know, it didn’t ask me earlier that is required for specifically for opening checking account. So I’m going to enter a couple of details here.

My email address.

Again, he comes back with confirmation, whether those all look good, I’m gonna confirm.

And submit it. So as you notice how, you know, our journey started about finding the products that are very personalized to me and then having many options, being able to make a selection on one of the products that makes sense to me, and then being able to apply and being able to get upsell into adding another product, all completed in one window in a very consistent way as if I’m talking to a customer.

As if I’m talking to a representative in the bank. Now, I am completed it, so I’m a customer now, as a member, I would like to transact. So let’s say one of the questions I had here real quick. So let’s say, I would want to do a stop payment on a check, say, stop payment.

And I checked.

So then it is asking me, hey, for this, you need to log in, Could you please enter the check number first? So, okay, I’m gonna give the check number. Give me one second here.

Then one thing you notice is it is connecting to UIPAP. It’s a RPA system, I’m not even touching it. You can see my hands are here.

You know, RP system kind of completed stop check payment, right? So you know, it automatically connected with my online banking system on the back end, and mimic the user action to whatever a consumer would do to take actions, complete all of that, and then report it back to me check his stop payment for the check has been issued. Right? So, you know, that’s kind of the, like, awesome aspect of it really, like, you know, any process that is humanly possible today can be automated with either integration, with the third partition through API or through RPA, right?

So hopefully you got an idea. And imagine the power of that RPA working in front end assistant do with fourteen different systems, and you’re chatting with the front line assistant, front line assistant could, in turn, connect with all this systems and execute task, mimicking your, you know, your staff behavior on those systems and completing while they’re just talking to AI agent with a simplified interface, right? So, that’s kind of the power of that. As you could see, there’s that was a plugin in action, where it was connecting to RPM, and all of that.

I’m gonna continue to show you a lot of exciting plugins. So let’s say we move forward here, and let’s say I want to verify a transaction that has happened recently.

Can you verify a recent, about fourteen ninety nine, charge on my credit card.

I’m asking a specific question, which is really hard to do in an online banking traditional U. S. Setup. It’s determined I’m not logged in. So it’s, yes, authenticate, and brings up a login. I’m just gonna enter my member number, It’s the same.

One second. I’m entering the download system.

Okay, if there’s an entered, then it’ll ask me, it’s a multi factor authentication, so it’ll ask me, either to send one time passcode email or phone, I’m gonna prefer phone, send it, and it asks for a one time passcode, so I’m going to enter the same, it authenticated, and it comes back, says the last transaction’s automated payment of my whole subscription.

I didn’t know that. So that was helpful. So, now let’s say, as I decode this, I would like to kind of understand when was the last time I paid? And I’m kind of curious to understand my expense, let’s say I’m continuing to dig deeper here. So one of the questions I may want to ask is, when was the previous time I paid? Or comes back and says this paid on May twenty six. Our was helpful.

I may wanna unsubscribe here, so I’m going to ask if that is possible.

So it seems like it has given instructions how to go ahead and subscribe with Hulu. And then now, I have gotten attention to the auto payments. I wanna check all the auto payments that are set up on my system. Right? So I will ask, can you show me all auto payments.

Okay. Great. So it shows me all the auto payments. And again, I got surprised by one transaction.

I was able to quickly get to it in traditional online mobile banking it takes forever to dig through them if it is a month old or two months old. So it was able to immediately fetch and tell, yes, you did the transaction on May twenty six. It’s not a fraud. Then I got curious about other transaction, dig a little deeper, and now I have all auto payments available, including guidance on how it can cancel Google subscription.

So and and now let’s say I got auto payments figured out. I want to do a little bit deep dive on my credit card expenses. So let’s say, show me my credit card expenses, for last month.

Right? So, as you could see, the whole, the benefit of this chatty bitty like experience is the whole screen gets cleaned up and responses, the response to your questions is the one that takes up the full screen. So your attention could be specific to what you need to accomplish, right? So instead of all the clutter you’d see in online banking, there is, the moment you ask question, the holds can clean up just to tell you, hey, here’s action you need to take Right?

So you could actually check these for multiple years. These are interactive. That’s what the multimodal is. It’s not just shared and wise.

You can interact with new elements and whatnot. There is a lot more exciting stuff we’ll show you as well. And now I got this. You know, I’m gonna say, ask for further of these expenses.

So, for example, can you, as it by the type of spend?

So, again, comes back with different kinds of categorization across accounts. Again, you can interact with these two elements. And then it also shows, hey, you’re twenty four percent over the last month. This is the in context, Mestro, like how to optimize the experience for the consumer.

So it says, do you want to set a budget? You know, that sounds good. So said, yes, it understood my concern, and it is helping me, you know, there’s a reason I’m digging expenses probably, I’m spending more. So it’s a it’s a budget.

And then, you know, you will send an email — will send an email notification.

What if, you know, I need to get like some sort of text notification fight or go or a budget. So I’m gonna ask, you know, can you spend me text instead.

Great. So it is already set up. It was my mobile number, it is set up. So now you understood how some of these transactions are almost impossible or online mobile banking, you’re able to dig deeper.

I quickly get answers to it, see the data, take actions. It is guiding you to take those actions, and that’s the magic of in context, my stroke, constantly figuring out how to make your experience better, I’ll help you achieve your goal better and think a potential member, you’ve gone through kind of finding right products and services and comparing them. Having relevant products being chosen, and then you apply, and you become a customer, and then you already you should stop payment, and let’s say you continue to bank for other questions, let’s say you want to pay minimum balance for your credit card.

So let me go and ask a question real quick.

Pay give me my name.

Balance of, say, nine dollars on my credit card.

So as you can see, it is kind of saying, we strongly recommend paying the full balance. Again, this is in context mystra coming to picture, saying you their interest for the rest of the accounts, the rest of the amount that is due, so it is recommending you to pay for all of it. I’m gonna say that’s okay proceed.

And then, you know, it successfully paid the minimum balance and now it is again, in context, myself coming into the picture kind of saying, hey, there’s another credit card you could probably switch to. It says lower interest rate, right? So and it also offers me a use your net production plan. So let’s say I’m interested a lawnmower, I’m gonna ask Tell me or about, you know, that prediction plan.

One thing you’re gonna wanna see is something very awesome.

I’m gonna tell you in a bit how this is working.

UCU depth protection plan is a valuable offering that provides member — So I’m just gonna pause a minute. I hope you saw a quick animation there. It’s just a plug in.

Where the AI was connecting to Simvasir, which is a platform that on the fly generates human like videos that is personalized, and what just happened now is sphere, our AI disconnected with that third party platform created a entirely new original video that is human like that in a personalized way explains the UC’s debt production plan, which is kind of amazing. So I’m gonna play this so you can listen in.

Give me one second.

Whereas with comprehensive financial security and peace of mind, designed to safeguard against unforeseen circumstances such as disability, involuntary unemployment, or loss of life. This plan offers a safety net for members and their families, In times of hardship, it helps alleviate the burden of debt by providing coverage for loan payments. With UCU debt protection plan, members can competently navigate through life’s uncertainties, knowing that their financial well-being is protected. It’s a proactive measure that demonstrates UCU commitment to supporting its members and ensuring their long term financial stability Okay.

You just saw a personalized pitch just to your customers and members about your offering that was created on the fly by the AI. That’s the power of multimodal, right? So AI figures out right way to convey the message either through UI elements as we saw we were able to navigate on clicking through some of them or through text or through a personalized video or an image, right? It is multimodal.

So now that I know about the edit production plan, So let’s say I wanna apply for it, so I’m going to go ahead, ask for, you know, how to apply for it.

And says, okay, I need to connect with a a loan representative.

Or I can meet a representative of the branch, so I’m gonna ask real quick for nearest branches, show me the branches near UC, let’s say online.

Okay. Great. So it showed up a map with closest branches available. Within ten miles, and I hope you noticed the co pilot feature, which is on the left side, automatically helped me with the next action I could take.

So the cool part is while the right side of the screen kind of focuses on clearing all the clutter and kind of focusing on the task you’re trying to accomplish, the input you’re trying to give and things like that, the left side of the screen always kind of guides you the next action you can take. Right? So as you can see here, I got the map with the nearest location on the left side, I got affixed to direction and whatnot helping navigate there to one of the branches right away. So it is also offering me if I wanna connect to a live representative So I’m gonna say, yes, and it is connecting to L2B chat.

And then it is connecting, finding a representative, and there you go, John is available to assist. So as you can see, this is like this you know, one stop shop, this can replace your websites, online mobile banking, hides all the complexity provides a lot of guidance through the process, does upsell cross sell, helps finding detailed information for your members and customers digging through transaction data asking, special questions and connects with all the third party systems, so you stay in one place, not you know, to jump between different screens and helps get things done. As I said, there’s nothing like this exist.

We’re so excited to introduce this.

This is spear and in the age of generative AI, there is a lot of possibility and this is a project we’ve been working for three years. And, you know, generally, AI made it happen. So pretty excited about this, and we’ll move to the q and a here real quick.

Alright. So let’s pull up a few questions here that we’ve gotten.

All right, is there a limit on the size of the financial institutions that adopt generative AI specifically? Is there an area where you know, there’s a threshold?

No. So because this is a federated generative AI, as in we don’t only use your data to train the banking specific models. This is a trained AI brain that is learning across all of the customers, like a QoS model, right? This is a shared data set that AI learns from, so any size of the organization could use it out of the box.

Hey, thanks.

A question here, is there any way for the sources used in generating responses?

Can they be provided or are they too vast to be provided? Sources are important for credibility and accuracy.

So it can be any sources. The way you use the source and train it is the important aspect. So, you know, there need not be any restriction of the sources that is required, but the way you use it and train leads to better accuracy.

Okay. Talking a little bit about implementing this, what are the potential risks and challenges associated with adopting generated AI?

Sure it’s not as simple as just opening it up and using it.

Yeah, so it is as simple as that in many ways, because we’ve taken the care of all the risks by creating this banking specific model for the privacy, you know, conferring salary risk and all of that.

Yeah. There’s another question I’m looking at it, Brian here. Does work with Glia, correlation, Bang John. So if in the process there, if you asked any questions that require a human intervention, it would, in the same, you know, chat GPU like interface, it would connect with Glia or Ultra at the back end, and the chat will start right there. That’s also a plug in that’s already built in. So, and with SPHER, you don’t need another online mobile banking system.

This is the replacement of your existing online mobile banking system. Right? There is a phase, right? Like, you know, you can start with putting AI system within your existing on a mold bank system, which is what we call pervasive AI.

This is what we’re doing today. This kind of future where the AI takes over rather than being just a box within the online mobile banking system. Right? So it does integrate with Glia and correlation.

The the banjo is an online or bank system, so in a way, it it can replace Bangjar.

Great.

Well, we are at the top of the hour. We will be making these presents notation slides available.

And we really do appreciate your time and we look forward to having additional conversations with you guys as your interest grows and leveraging generative AI, please reach out to us if you’re not yet and a customer, we’d love to help you get to the next level.