Accelerating tech modernization though gen AI agents

[Music] Thank you for uh for joining this session. Um just a quick intro. My name

session. Um just a quick intro. My name

is Raj Becker. I'm a senior partner here at McKenzie. I lead a lot of our tech

at McKenzie. I lead a lot of our tech work, particularly within the wealth, asset management, insurance space. Been

in the industry for about 25 years, uh doing more of the same. and we're

pleased to be here with you guys today and talk about AI and modernization using AI. But before we do that, Dante.

using AI. But before we do that, Dante.

Yeah. Hey everybody, Dante Gabrieli. I

am a principal product manager out of Quantum Black. Um that is our AI

Quantum Black. Um that is our AI practice uh at McKenzie. So I work within our labs division focus on building uh technology and capabilities that we bring to clients to help

accelerate tech modernization and also adoption of cloud. super excited to be here and uh chat with you about AI agents.

Fabulous. So, let's get into it.

Um there's five things we hope you take away from this particular session. First

is we have observed in our client work 35 to 45% reduction in cost and risk when you use AI Gent AI for

modernization.

This is a massive deal for all the folks in the audience. You have your companies, you've been through these sessions as well as um opportunities to modernize capabilities in your own

institution. You know, it's

institution. You know, it's costly. We have observed 35 to 45%

costly. We have observed 35 to 45% reduction in both cost and risk. I hope

you take that message away. So, the

question is, well, how how does that happen? Couple of things. One, we know

happen? Couple of things. One, we know tech is real. You guys are all here.

You're at this conference. You know the tech is real. You've been to all the different uh breakout situations and rooms, you know, the tech is real. But

it's not just tech, right? There's also

this massive human element associated with it. The humans that are building

with it. The humans that are building these agents, how we scope these agents, all of that is coming from the human, right? If we don't get that right, these

right? If we don't get that right, these agents don't perform the way we want them to form. Point two. Point three is you can have the agents, you can have the humans who are building them, but if

you don't have a structured approach, to actually scaling this it becomes pilot purgatory right you can't scale it in an institution so that's point point four

the last point I want to make is so you've got the tech you've got the humans that are building it you've got the structured approach and the last element is change management people that are going to use these things or the

outputs of these agents all of those things have to be in place in order to scale this capability These are the lessons we've learned over

the last 18 to 24 months looking at agentic AI right so we hope you take these four or five messages away from this particular uh

particular talk um just very quickly why now what's happening and again you guys are in the eb and flow you're at this conference you see it our ability uh the advances

in reasoning have actually propelled us in this direction where we can now have automated agents doing a lot of these different things, but they're overseen

by humans, right? Um the value realization

humans, right? Um the value realization equation to this is actually quite compelling, right? What used to take us

compelling, right? What used to take us 7 10 years to modernize a particular system mainframe platform pick pick whatever you want to pick 7 to

10 years we can now realize in months and we start to realize value in months and we'll show you a little bit of equation at the end in terms of how we think about the business case and why that's so important as you go in these u

in this particular direction from a modernization standpoint. All

standpoint. All right, awesome. Let's get into tech now,

right, awesome. Let's get into tech now, though. Who's excited about tech here?

though. Who's excited about tech here?

Oh, yeah. I sense the excitement. Um,

very cool. So, want to take you to a few things, right? So, first is what we've

things, right? So, first is what we've learned technically about doing this over the course of the past couple years, right? We all know that Genai,

years, right? We all know that Genai, Aentic AI is accelerating at a pace and innovating extremely fast. what have we learned in real world applications where you know we're we're solving real

problems for clients and then what we want to do is go through a demonstration right let's see this live and in action so let's start here the key thing and Raj talked about this is about scale and

the key thing about scale is not running around and simply throwing code or throwing business processes into a chat interface and that being the accelerant

that's going to get your organization to a point where they're able to make step function change that's going to make the business successful long term and what you see on this screen right here is much of the learnings that we've talked

about. Now what's interesting here and

about. Now what's interesting here and Raj talked about this as well is that we don't see Genai as a silver bullet. We

see this as the the next iteration the next evolution of how automation is fundamentally changing the way in which we operate and organize our businesses.

Okay. So what you should see here is that we have have started to to implement these capabilities at clients but we've done it in such a way that we've taken a microservices approach right you can see the various services

that we've put in here we've introduced uh things we've learned around conversation right so you see the conversation service over here on the right this is a key thing and you'll see this when we do the demonstration it's

really important that we don't let the agents just go on and do their own thing but that they're able to come back and ask us questions questions, okay? That

they're able to work with subject matter experts. And we're getting the most out

experts. And we're getting the most out of our subject matter experts and not simply um sitting there idally while we let agents do their work. We liken it when we talk about

work. We liken it when we talk about this inside of our teams to to children, right? So, do a lot of you have children

right? So, do a lot of you have children out here? Maybe. I see a couple hands.

out here? Maybe. I see a couple hands.

So, what's interesting here is when you think about agents, right? We think

about them. They bring a lot of like humanlike personalities to it. Not that

we're saying agents are humans, but that's what's been intriguing to people.

So, I liken it to my son. My son's

name's John, right? When I tell him to go and clean his room, and you probably done this, he's 17. Um, what he does is he folds up a couple shirts, moves it over to the right, and then starts doing

something completely different. I come

back, I'm like, "John, what's going on here? You didn't do what I asked you to

here? You didn't do what I asked you to do." And we iterate on the process,

do." And we iterate on the process, right, John? move the shoes and get them

right, John? move the shoes and get them back over to where they need to go. This

is very similar to what we've learned with agents. They're able to come back

with agents. They're able to come back and quickly get us information, but using the subject matter experts to be able to refine the approach and make sure that they're moving in the direction that you expect is key to

making this very successful long term. So, how do we how do we simplify

term. So, how do we how do we simplify this? How do we build out like a

this? How do we build out like a taxonomy or an ontology that really breaks down what are the components that drive agentic orchestration at scale?

And what you're seeing here is similar terminology to what you'd see within the tech space, things like tools and agents. But then how do you start to

agents. But then how do you start to organize these agents in such a way that they work together, right? In the same way that we work today where we would

pull in a group of experts, give them an objective, a problem to solve, it's the same way we want to think about agents.

I was chatting with some clients and some different folks within our our organization. And I think what we all

organization. And I think what we all loved about cloud was the fact that it gave us access to compute in an ephemeral and an elastic way. Right?

What was awesome was the fact that as a development team, I could quickly spin up infrastructure, not worry about the historical slow IT service management processes, spin it up, get get some

compute, spin it down when I'm done, and I've only paid for the time that I've actually used that compute power. Agents, we've been talking about

power. Agents, we've been talking about it is very similar to like ephemeral and very lowcost workforce management. How

do I go out and spin up a set of experts very quickly, have them execute on an objective, get it to me way faster than any human would, and then have the ability to actually iterate with them,

and only pay for it during the time that I've actually used them. It's it's it's a great shift in thinking, and allows us to think about how we scale work forces very elastically um without the overhead

of traditional HR processes. And that's

what you're again seeing here. And then

we think about flows. How do teams work together? Right? If I have, and you'll

together? Right? If I have, and you'll see in the demonstration, um, a set of agents that are responsible for understanding the business logic that exists within, uh, Cobalt and a mainframe, how do I then pass that

information over to another team that's starting to think about what is the future state design of the business process and turn that into a cotification within a modern stack that

then enables the business long-term, right? And this is what Raj was talking

right? And this is what Raj was talking about. It's very key that the business

about. It's very key that the business process is the the main driver, not the technology. Technology is an enabler and

technology. Technology is an enabler and historically uh technology uh the business process has been uh shoehorned to meet the needs of the technology.

What you're going to see today and what you're going to hear us talk about is how the business process, how the change management, how the goals of an organization are the key to transforming businesses and the technology is an

enabler that's going to support that long term.

So what does that look like practically?

Right? So what you're going to see here, and I'll show you this, and then we'll hop into the demo, is it's very natural language, right? I want to be able to

language, right? I want to be able to describe what I want a set of agents to do without having to think about the syntactical um requirements of traditional programming languages,

right? I want to do this quick. And the

right? I want to do this quick. And the

key to that is this is if I can make it very humanlike and I can do it in such a way to where it's natural language, I can start to expand the use cases to

users that may not have the technical depth as engineering teams. And again, that's what we're seeing within businesses. We've deployed this with

businesses. We've deployed this with business analysts and folks like that that are actually using this to synthesize conversations with uh various users and stakeholders to within minutes

fully understanding and synthesizing what's happening and what the goals are across the organization. This the the step function change and acceleration is real as Raj said um and we're very

excited about it. Then how do I define my experts? Right? Again, we talk about

my experts? Right? Again, we talk about this workforce management approach. In

the same way that I would create a wreck today to go and hire someone, why would I not have that same uh approach and actually defining the sets of agents and team members that I would pull together

so that they can work, right? And that's

what you see over here on the right hand side. Right? In this case, we have a

side. Right? In this case, we have a planner agent. Think of this as the the

planner agent. Think of this as the the person responsible or the agent responsible for organizing the group, ensuring that they're passing the objective along. They understand the

objective along. They understand the skills and capabilities of each of the agents so that they are able to uh fundamentally work in a way that's most effective. So I want to do now is I want

effective. So I want to do now is I want to walk through a demo for you. So let

me set the stage for the demo. One, it's

recorded. Uh that's on purpose for anybody who's done live demonstrations.

Uh they can go crazy. Um, plus agents are conversing here. So, as fun as it would be, uh, probably to do that, we would use up all of our time to just watching these agents

talk. So, the scenario we're going to

talk. So, the scenario we're going to talk about here is a payment processing system that's running on a mainframe.

And in this case, the the the the organization uh is getting a lot of feedback from their customers within a call center, right? Um, and it's not great feedback. and they're coming to us

great feedback. and they're coming to us going, "Okay, look, we know our customers are upset. We know they're frustrated, but we really don't know where to start." And the key thing is we

don't want to modernize the entire mainframe um in order to achieve greater customer satisfaction. How do we build

customer satisfaction. How do we build something that's going to solve the problem but doesn't require us to stay on the main frame? So, we're going to walk through three stages. Okay. First

we're going to talk work through the current state what we call a reverse engineering phase. Um that is to

engineering phase. Um that is to understand the business process the problem we should target and what the actual technical architecture and software architecture is of the

mainframe. After we understand that what

mainframe. After we understand that what we want to do is work with the business leaders the product managers as well to say let's reimagine the business process. What would this business

process. What would this business process look like if the technology weren't an inhibitor? Okay. And then

lastly, what we want to do is say, okay, can we commission a team of agents to accelerate the buildout of the code that will implement this new capability that's going to solve this problem. All

right, this will move fast. Um,

shouldn't have to pause it, but if I do, I will here.

Okay, so what you're seeing here is uh the training document for all the call center agents. So we're doing is

center agents. So we're doing is commissioning a set of agents that are going to go in and actually read the document and uh put it in a flow diagram for us so that we can quickly understand. You're seeing that there's

understand. You're seeing that there's two agent teams here. The blue dots represent agents we've defined. Uh the

silver dots represent tools that we've given them beyond LLM access. You see

the conversation happening over here on the left. And this is a key thing. We

the left. And this is a key thing. We

want people to be able to quickly understand how the agents made their decisions and how they came to the conclusions and outputs that they were there. You see up in the top right um

there. You see up in the top right um all the artifacts that the agents are creating. So what we're doing now is

creating. So what we're doing now is we're scrolling through the conversation again time-lapsed here. So this takes about 3 or 5 minutes if you did it in real time. And the agents have worked

real time. And the agents have worked together. We have critics within there

together. We have critics within there that are, you know, analyzing what was done. And we're going to pull up the

done. And we're going to pull up the document. You've got the flow process

document. You've got the flow process which shows the customer uh the call center and the actual um payment uh solution. The key thing and you might

solution. The key thing and you might have missed it is that the customer always has to call the call center before to get their their balance. And

this is going to be key as we go through the actual uh logs to determine what problem to focus on. So we've

commissioned a whole new set of agents here. We're saying look, we've got a

here. We're saying look, we've got a bunch of call center logs which are representative of all the calls that have come in, the customer satisfaction score, resolution time, things like that. What we've asked these agents to

that. What we've asked these agents to do is to analyze that document and surface for us what's the key area that we need to focus on. What's the problem that if we can solve, we can get that

customer satisfaction score up higher.

And again, what you're seeing here is the agents conversing. They're working

through the document. We've got a Python coder in there, you know, writing the code to analyze the document. We've got

a data mapper trying to understand what are the types of calls and what you see here is we get a nice markdown document that says hey payment or billing payments is a key problem. Now go back

to the business process. The actual

problem was customers always have to call the call center to get their payment information which then results in lost uh or late payments. So what we need to do is give users access to the

technology. But before we do that, let's

technology. But before we do that, let's understand what the software architecture is today. So here's some some awesome cobalt, right? Um all of us are excited about that. So we we give

the agents some access to the cobalt and we say what we want you to do is give us a high level software architecture that defines the databases, the batch

processing, the front end, so the BS BMS screens um and and who is accessing these systems. So again, we have another agent team. This goes back to what we

agent team. This goes back to what we were talking about before. How can I quickly spin up experts rather than pulling from my my human labor force and have them work on these uh these

problems in an iterative fashion as fast as possible. So you'll see here that

as possible. So you'll see here that again we land on the same session screen. They're conversing and up in the

screen. They're conversing and up in the top right uh we're starting to get additional information. So the agents

additional information. So the agents were able to work through it. Here are

the UI components. Here's the uh database. Here's the batch processing.

database. Here's the batch processing.

So on and so forth. So we've got three things. business process, software

things. business process, software architecture, and the problem. Now, what

we've done is we've had conversations and interviewed business stakeholders.

What should this process look like going forward? And we've recorded that. So,

forward? And we've recorded that. So,

rather than traditional like let's create some user stories, let's build this requirements document, why don't we just use agents to actually transcribe the audio and then use that as a way to

start to refine what this process should look like? And again, we're not talking

look like? And again, we're not talking what should the tech be at this point.

We're simply talking what should the process be because that's the key driver. So the agents are are working

driver. So the agents are are working through this. Now what'll be interesting

through this. Now what'll be interesting here is you see on the bottom right a question just popped up. This is super important. The agents took their first

important. The agents took their first stab at listening to what was said and they build out a diagram and we're like close but not right. So this is where our experts come in. What we want you to

do is remove this vendor column that you have and we want you to use a particular set of logic to determine when the actual uh notification should go out to the customer. So as a human you see us

the customer. So as a human you see us typing that in here. We're providing

feedback to the agents. They are taking that feedback and reconvening to actually introduce the updated information into what the business

process should be. So they continue to converse here. Um, and we'll give it a

converse here. Um, and we'll give it a second. They're talking some more.

second. They're talking some more.

Imagine if I made you guys watch this in real time for 15 minutes. I'm sure that would be a lot of fun. So, you see that they've made the changes, but we've got one more here. We're saying, "Hey, you've le you've left the user out. You

left the customer out. Can you go ahead and actually update and make sure that there's a swim lane that shows that the user should be involved?" So, you see in this case, our product manager typing it

in. They do that. the a the agents take

in. They do that. the a the agents take the new information and what they're going to do here now is create what will be the final business process. Okay, so

we've got a new link here. Got some

plant UML. Boom. There you go. So, what

do we want to do lastly? We want to design a new software architecture. Now,

this is key. What we're what we're telling the agents is look, we want something cloudnative. We want that to

something cloudnative. We want that to be integrated into the mainframe because the last thing I want is to spend, you know, two years updating the mainframe to solve the the notification issue

that's creating a horrible customer score uh customer satisfaction score.

So, we asked them to to come up with this. We've uh we've obviously told them

this. We've uh we've obviously told them to use Google Cloud. Um and you'll see that here um in a moment. But over here on the right, and we'll pull that up again. What you'll see is that there's a

again. What you'll see is that there's a new notification service that sits external to the mainframe and it's going to integrate uh with the mainframe to solve this problem. So we get to to

customer value to client value faster.

So now lastly we're going to take this last agent team. What we're going to say is like look we want this running um in Google workflows and we also want this written in JavaScript uh for the back

end. So we commission a new set of

end. So we commission a new set of agents with those capabilities and now what they're doing is taking the business process. They're taking the new

business process. They're taking the new software architecture and they're going to work together to build out the YAML file that's going to be uh what we put in there for the uh Google workflow and

we're also building the JavaScript that is the actual business logic that's going to execute on the notification system and uh the the other things as well, the balance checking and those types of things. So, what you're going

to see here now is we're going to take the AML file and hop into the IDE.

You'll see the configuration here, and we're going to open up uh the Google Cloud uh console, right? We're going to drop that in here. And over here on the

right, what you're going to see is the entire process now in cloudnative form.

So um again fasttrack through the entire process and there are iterations that would occur even further as you're working with your clients but again the key is the business process and having

the technology support that in the future and make sure that the business can evolve. Raj turn it back to you man.

can evolve. Raj turn it back to you man.

[Applause] That was pretty powerful what you guys just saw, right? Prior to this, it was all humans

right? Prior to this, it was all humans doing all of that work, engineers doing all of that work. Right? Tomorrow, and

this is where the 35 to 45% comes into play, is you've got your tech workforce building these agents, nudging and training these

agents, and supervising these agents.

So imagine how that changes from a tech perspective, as a tech management function, right? I think uh Jensen Wong

function, right? I think uh Jensen Wong said this best when he said, "Look, the nature of the tech function is going to change as a result of some of this technology. We're going to be

technology. We're going to be supervising and training these agents and building these agents as opposed to doing the work that you saw the agents do just now." Right? That's the crux of

this 35 to 45% reduction in cost and risk. Right? We put risk in there on

risk. Right? We put risk in there on purpose, right? Because again, we all

purpose, right? Because again, we all know the more humans you embed in this stuff, the more risky it gets, right? So

if you can take that risk or at least reduce it, it's tremendously valuable for any institution. So this slide here just

institution. So this slide here just kind of articulates a business case at a very very macro level, right? There's

tremendous amount of modeling spreadsheets that go behind this. But if

you just look at it simply from a um sort of a cost per line of code standpoint, if you have a humans in the traditional way that we do

modernization, $6 a line of code. If you embed agents, it's about $3.75. And this is from our own

$3.75. And this is from our own observation as well as independent external research uh around the $3.75, right? And that's a 38%

$3.75, right? And that's a 38% reduction. That's 0.1, right? That's the

reduction. That's 0.1, right? That's the

the cost. Then you look at the time to modernize what you just saw these agents do. Again, imagine a human being doing

do. Again, imagine a human being doing that. That's probably months if not

that. That's probably months if not years worth of work to go through that, right? to document what's in our legacy

right? to document what's in our legacy platform to reshape it and then actually implement it years worth of

work. We believe and we've seen we've

work. We believe and we've seen we've observed we can take as an example and the numbers under the time are years 10 years to modernize let's say a set of

platforms in an institution down to roughly five. It can be

five. It can be done. It's being done.

done. It's being done.

Right. And then the last point is just around the uh the run cost savings from a um sort of a mainframe to cloud, right? Let's not even factor that in.

right? Let's not even factor that in.

And by the way, there's so many other benefits that we're not even contemplating in this business case, right? In terms of just streamlining

right? In terms of just streamlining organizations, right? Um skill sets that

organizations, right? Um skill sets that people would have, right? There's

benefits that you would acrew out of those things as well, right? We're just

taking a very simple view of the business case. So, we think it's pretty pretty

case. So, we think it's pretty pretty powerful. Um, let's maybe talk about

powerful. Um, let's maybe talk about some of the pitfalls and maybe give you guys some lessons learned around what we've seen in our travels. I think on the left

side, despite what you just saw, AI, Genai is not a silver bullet, right? Like, I'll give you a

bullet, right? Like, I'll give you a small story. um one of my clients

small story. um one of my clients insurance carrier they said hey can you just give us the tech right and we'll do

it we said okay why don't you just try it you can do a pilot they want to do a piloted claims function of this carrier they're still trying to figure

out the business case with the claims function itself haven't launched the pilot still just working through the people issue that we typically see to understand the business case and

convince the business that this is the right thing to do. Right? So, I reiterate, it's not

do. Right? So, I reiterate, it's not just about tech. The tools are there, the tech is there. It's wonderful. It'll

get us to the promised land, but it's not a silver bullet, right? We still

have humans involved in these organizations. You have to convince

organizations. You have to convince people that this is the right thing to do and the business case will hold up, right? That's point one. Point two

up, right? That's point one. Point two

is aligning on the business outcomes. We are totally against

outcomes. We are totally against modernizing things for modernizing sake. It doesn't lead us to any business

sake. It doesn't lead us to any business outcome. Right? So we have to ensure so

outcome. Right? So we have to ensure so as again I give you a small story in the claims function. We are trying to

claims function. We are trying to modernize claims right? How property and casualty

claims right? How property and casualty claims are handled by these carriers. Right? We all see what

carriers. Right? We all see what happened in um in LA from the fire perspective. We want to give people

perspective. We want to give people those claims quickly, but today it takes months. There's no reason why it can't

months. There's no reason why it can't take days if we fix the claims function.

That's an outcome that we want to go for, right? So, we want to modernize the

for, right? So, we want to modernize the platforms to allow us to get that functionality to drive that outcome. Right? And that's the business

outcome. Right? And that's the business case.

The third piece is overlooking change management. So you know how I mentioned

management. So you know how I mentioned you know humans who are building these things but then the the last part of the equation is change management people that are going to use this stuff right

and I'll give you a small story. Uh by

the way I'm in fig so all my stories around um banking and asset management insurance. Um so this one is a asset

insurance. Um so this one is a asset management story. So if you're a

management story. So if you're a portfolio manager, you're managing gobs of money, right? Billions and billions of

money, right? Billions and billions of money, right? And you're selecting where

money, right? And you're selecting where to invest, right? So here, you know, the

invest, right? So here, you know, the tech department or even the business comes and says, "Hey, we're going to automate that selection for you, right?

AI is going to tell you where to invest your money and what kind of allocation I'm going to I'm going to have in my portfolio, right? I mean, immediately

portfolio, right? I mean, immediately organ rejection immediately.

Right? Because if you're a portfolio manager, this is all you've done your entire life, right? You're not going to be

life, right? You're not going to be replaced by an AI or a bot or an agent, right? So that whole change management,

right? So that whole change management, helping that portfolio manager understand where the value is in using this and the output that's coming out of these agents, how best to put that as

part of their investment process as an example and how best to use it is tons of work, right? So again reiterate the

work, right? So again reiterate the change management aspect of it is critical right and it's what we call frontto back rewiring an institution or

a company to actually use this stuff in a meaningful way right that's on the left side of the page right side of the page um what are some success factors one is

you you got to have the leadership and stakeholders bought into this right again this is a gamecher but you've got to bring the organization along Not just the people that are going to

use the output, but also people that are part of the institution as a whole, right? You've got to bring them along,

right? You've got to bring them along, make sure they're aligned, make sure they're motivated, make sure they understand the business case. Point one.

Um, point two, the way to start, and we've seen people go haywire on this thing, they'll just pick a use case.

That's great, but then you look back a year later, it's like, okay, what did that use case deliver? Maybe it proved the technical capability of it which is wonderful helpful absolutely we need to

do that but it didn't drive to an outcome to a business outcome. So the

business is left wondering well where's my benefit right I've written this check the tech folks got to do tech it's wonderful but what's the benefit right

so again targeting a specific outcome is actually pretty critical in making this stuff come to life and we do have a perspective on that right use cases sometimes too small transforming entire

business might be too big but transforming a domain might just be right right as an example so that's point Uh point three again you're modernizing

for organizational agility right we all know changing something or a feature or a variable in a particular system especially a legacy system takes

years right it slows us down from a business perspective from a market perspective right and things are moving at the speed of light right and we can just see that over the last few days in terms of what the markets have done and

tariffs and all of that right we we've got to be able to adjust to the speed of the market and we need our tech to be able to adjust as well. Okay, thank you.

We appreciate it. Um Dante and I will be on the side if anybody wants to chat.

More than happy to chat but appreciate the time.

[Music]