S1E7 - Unlocking Gen.AI Potential in Financial Services With Murtaza Kanchwala

In this conversation, Murtaza Kanchwala from Amplifi Capital discusses the integration of Generative AI (Gen.AI) in financial services, sharing insights on early applications, challenges faced, and the importance of regulatory compliance. He emphasizes the need for effective team organization and the selection of appropriate AI models to enhance productivity and efficiency. The discussion also touches on the future of AI in finance, predicting the emergence of personalized AI assistants and the ongoing evolution of AI technologies.

Takeaways

Murtaza has been involved in financial services since 2014.

Gen.AI applications started with content generation in 2021.

Feedback from users was gathered through personal discussions.

AI hallucinations posed significant challenges in early implementations.

Internal AI solutions were prioritized before customer-facing applications.

Testing AI requires a different approach than traditional software.

The maturity of Gen.AI use cases is improving over time.

Amplifi Capital is building an AI Matrix platform for Gen.AI use cases.

Choosing the right LLM is crucial for specific use cases.

Regulatory compliance is essential in financial services AI applications.

Sound Bites

"AI is delivering something artificially."

"The landscape is moving faster than we think."

Chapters

00:00 Introduction to Gen.AI in Financial Services

06:27 Early Applications of Gen.AI in Finance

11:07 Maturity of Gen.AI Use Cases

16:51 Building the AI Matrix Platform

22:48 Regulatory Landscape for AI in Finance

29:37 Building Effective Squads in AI Projects

35:18 Future of AI in Financial Services

What is Data Matas?

A show to explore all data matters.

From small to big, every company on the market, irrespective of their industry, is a data merchant. How they choose to keep, interrogate and understand their data is now mission critical. 30 years of spaghetti-tech, data tech debt, or rapid growth challenges are the reality in most companies.

Join Aaron Phethean, veteran intrapreneur-come-entrepreneur with hundreds of lived examples of wins and losses in the data space, as he embarques on a journey of discovering what matters most in data nowadays by speaking to other technologists and business leaders who tackle their own data challenges every day.

Learn from their mistakes and be inspired by their stories of how they've made their data make sense and work for them.

This podcast brought to you by Matatika - "Unlock the Insights in your Data"

Aaron: 00:15

Today, we're joined by Murtaza from Amplify Capital, an organisation dedicated to making financial services more accessible. And today, we're discussing Gen AI, data, and in a highly regulated space, technology, teams. Let's dive into it. Welcome to Data Matters Live, where we discuss all things data. Today, we'll be discussing gen AI and data in financial services, And we're joined by Muteza, CTO of Amplify Capital, whose main focus is, accessibility to financial services.

Aaron: 00:52

And, Murtaza you've been there and done that. You've been developing AI long before it was even, you know, fashionable and and trendy. So I'm very excited to dive in and and get your experience and and particularly in financial services. So thanks for coming on the show.

Murtaza: 01:08

No. Thanks a lot. Thanks a lot for inviting me to the show. Awesome. And, yes, yeah, I have been involved in data for a very long time.

Aaron: 01:15

Yeah. I mean, that's and that's probably a great place to start, you know, to kinda get the background and and, you know, how did you find yourself in in these kind of AI projects within within financial services?

Murtaza: 01:28

Sure. In fact, I've been means, of course, I'm an engineer by heart, but I've been into data probably more than 2 decades where I have been involved in lots of data projects. And then, as we're involved in data projects, you know, it's it's defacto. You'll probably somewhere down the line, you'll touch on AI. And financial services, got involved in financial services since, 2014.

Murtaza: 01:55

I've been in in financial services for probably a decade now in in variety of organizations, payments, trading, including blockchain, FX, invoice financing, insurance, and banking. I got involved in Gen AI in 2021. So I've been involved with traditional AI, and then Gen AI, I started touching in 2021. In 2021, when we started touching on on Gen AI, we were exploring the early models of GPT 3, but I'm trying to really explore as to, you know, what what we could do with these, LLMs. Organizations, of course, they were not aware of, you know, what what the the capabilities these models offer at the time.

Murtaza: 02:40

So we and luckily, you know, I was part of the innovation team where we had enough room to explore and research on on what we can use. And we we started building, initial applications where they were mainly driven for content generation, and, we try to give these applications to users within finance, legal, marketing, where they can they they they generate a lot of content. But also they

Aaron: 03:11

So they they actually started with content generation. So what what give us an idea. What sort of content do they have in mind?

Murtaza: 03:19

To be very frank, when we started in 21, we just had to, you know, come up with with concepts. So so we had, things like, you know, generating an email, generating, some blog posts, helping review, summarizing, documents on track.

Aaron: 03:36

So productivity mainly. Like, actually trying to get, you know, one person to do the work of 10 kind of idea.

Murtaza: 03:43

That's correct. Yes. Yeah. And, also, we wanted to, get, employees to to experience this step. Mhmm.

Murtaza: 03:52

Because the more they experience the tech, the more, feedback we can get in terms of, you know, where they where we can help with these tools.

Aaron: 03:59

Yeah. Yeah. How did you go about gathering that feedback? Like, yeah, it's obviously they can use the application. You could sit beside them, look over their shoulder, but probably you you might not get they're when when you're sitting there, they're they're sort of real what's going on.

Aaron: 04:13

So, yeah, how did you get their feedback?

Murtaza: 04:16

We actually the way the way we got feedback was is is personal discussions. So we didn't create, like, you know, you know, say send us an email. We we literally sat with them, and experienced their day to day work and explore exactly where you're you're utilizing these tools. Also, that gave us an opportunity to understand their day to day work or what they actually do on a day to day basis. But that that's the best way of collecting feedback.

Murtaza: 04:40

Then as it as it got matured, we started creating, areas where they could give us the feedback. They could also give us, like, rating on the responses, thumbs up, thumbs down, so we know, you know, whether it worked or not.

Aaron: 04:53

Yeah. Because yeah. Yeah. No. I think that's probably something everyone's a lot more familiar with now that that sometimes the output's astounding and, you know, you're exactly what you want.

Aaron: 05:04

Other times, it's it's kinda miles off and you can tell it's miles off. And the the problematic ones are where it sounds about right, but actually, though, it's completely wrong. What what were those early days like with the users? I mean, were they I imagine they might have been blown away because it wasn't common knowledge.

Murtaza: 05:23

Yes. Some of them were were blown away. They couldn't, you know, actually believe, the responses because they were very human like. Right? But also you are you're you're spot on.

Murtaza: 05:34

Right? Because in in the early days, we didn't know how to control the hallucination, and the l LLMs actually hallucinate significantly. And it it the responses are so promising that some users were blown off, but some users were like, that's not correct. You know? Yeah.

Murtaza: 05:53

Those were good early feedbacks we got.

Aaron: 05:57

Yeah. That like you said, you touched some hallucinations there. Yeah. The the whole technology is trying to find the the best fit, the best sounding thing. Almost like, you know, an exceptional liar, you know.

Aaron: 06:12

It's it's really designed to sound convincing like what would sound exactly right here, you know, without any regard for the, you know, actual facts. It's just what would fit exactly right here, which is, obviously really challenging for financial services and and a and a domain where, you know, making stuff up is just not on. You know, that's, that's tricky.

Murtaza: 06:34

No. Absolutely. Yeah. Yeah. You cannot make things up.

Murtaza: 06:38

Yeah. Yeah. You know, so, one of the areas we were we were helping, the team was, we had loads of agents, right, insurance agents. So we we we were thinking of, generating product literature, and you couldn't hallucinate that as if you you created the wrong product in future that could have a big impact. So Yeah.

Murtaza: 06:58

Yeah.

Aaron: 06:59

That's right. So, but as you mentioned, the, you know, some of the use cases generating content, and then you mentioned the the way that, you know, if you're offering advice, did these services find their ways directly to the customer, or was it always via a a person inside the bank as more of a a tool, but they still had the final say? What what what did it look like? You know, the one the ones that you've deployed?

Murtaza: 07:28

Initially initially, there were internal use cases where we created, multiple, AI assistant or AI bots, with the vertical knowledge stores and then some, ability to generate some content if they if they wish to. But then it did go to the customer. So the initial use cases were indirect to the customer. So the internal employees would deal we were engaging with these bots and then having a human touch with the customer. And then we started looking at exposing these interactions directly with the customer.

Murtaza: 08:04

That took a while because we wanted to make sure, it has the right guardrails. It's matured enough. The conversations are right. There's lots and lots of testing you need because every time a response would be different potentially. We also had to look at, you know, how do we actually store the conversation history because the conversation could get quite longer.

Murtaza: 08:27

Right? And and and those conversation history, then could cost you a lot of money just to have one one one conversation due to the total limits. So those did happen, but we had to go through a mature cycle to understand this this, use case will work. It doesn't hallucinate. It has the right guardrails before we actually went into production.

Murtaza: 08:51

And we did go into production, through for some of the social media channels like WhatsApp, Facebook, and Instagram.

Aaron: 08:58

Yeah. Yeah. Yeah. Yeah. I I also found that testing and the guardrails a challenging aspect because it's actually quite counterintuitive to any kind of normal testing.

Aaron: 09:10

You know, you you very much expect an input to generate an output and and and to test that. There's, you know, with Gen AI, you can have an input and an output, and then do you test again, same input, different output. And that's just, you know, a concept that's quite hard initially to, like, get your head around, well, a, how to deal with it, but also why that might be the case. And, you know, same with the guardrails, you know, you kinda don't really know initially how to test it and go how to go about making sure that's doing the right thing. And and we what what did you do in that regard?

Aaron: 09:45

Did you do it all internally through your own research team? Did you bring in externals, like, you know, these are the early days where no one really knew. So how did you overcome those challenges?

Murtaza: 09:54

It was mainly through internal research teams. We did engage, you know, you know, one external research organization, but most of the engineering actually, pretty much 100% of the engineering, we we looked at, internal. Pretty much 100% of the engineering, we we Mhmm. That, internal. And you're you're absolutely right.

Murtaza: 10:09

It's it's it's a fairly new. Right? And, whoever came in to guide us in terms of, Gen AI or how we should do Gen AI. We we found that, we had the same knowledge or or in in some scenarios, a bit more.

Aaron: 10:26

Yeah. So it's it's fascinating to see how Gen AI is growing and and, you know, the hype around it. And one of the things I I kind of wonder often wonder, you know, have we reached the peak already? You know, are we are we, you know, starting to see all the use cases we will see? What what's your opinion?

Aaron: 10:45

Is it is it are there more things it can do in financial services or or in the wider world that that we haven't even dreamt of yet?

Murtaza: 10:54

I wouldn't say we've reached the peak as of yet. We've definitely, got a lot more matured in terms of understanding these tools and these models. There is a lot, you know, I I think we have we have we have we've just gone with, we had the hype cycle, and we've reached we reached a stage where it was quite overwhelming with these tools. Now we've reached a a slight level of maturity where we can understand the models. We can understand the technology better.

Murtaza: 11:29

We have lots of tools available in the market. We can, create, proof of concepts a lot faster in in matter of days or weeks to test that this use case is valid or not. So what I what I see now is the majority of the use cases will be a lot better. Businesses will be able to identify where they need to apply GenAI, and in which processes. And you will probably see a shift from original content generation or conversation to being solving, more business problems, underwriting, for example, legal and compliance, operations, processes, and so and so forth.

Murtaza: 12:13

So

Aaron: 12:14

There's there's kind of 2 areas for me that seem to be the indicators that we're reaching a different level of maturity. The first is kind of I saw, GitHub in their, Copilot as now allowing users to choose the language model. So, you know, you have some control over which particular model you want to get advice from. And I think that in in this use case we're discussing could be thought of in terms of tone. Like, if it's a legal use case, there's probably a legal tone that you want to adopt or want want the generated content to adopt.

Aaron: 12:50

The other is actually the knowledge itself, you know, so we, you know, can imagine in financial services want to use a, high quality knowledge base, trusted information, 0 hallucinations. You know, there's a very much knowledge component that seems to be getting more mature. And, you know, the 2 together, I think you'll probably end up with that's where we'll get more growth. And and then, like you said, it becomes available to different use cases that we we we would know to be, challenging today that that probably you can't use the, you know, the straight OpenAI API generated response. You need to augment it with both that that tone and the and the knowledge base.

Aaron: 13:37

I wonder then, where's Amplify Capital putting its effort? Where where are you sort of starting on this, you know, leveraging the the benefits?

Murtaza: 13:48

So with Amplify Capital, we we originally when I joined, the first thing I did was to identify, you know, which areas we can bring efficiency, because it's always about, you know, moving the needle. Right? Mhmm. And where we can get the fastest return of investment. And and then, we started identifying the the use cases.

Murtaza: 14:09

So we listed out, you know, areas where we need, efficiency, and then the use cases, which we ended up with 13 use cases. But then what we did after that was we validated each of those use cases. And the way we validated each of those use cases by just putting an IRR in there and looking at, you know, what is the percentage, benefit we are gonna get, in terms of productivity, cost benefits, so and so forth. So there's a there's an algorithm which we apply. And it was quite fascinating because, the use cases which we believed would be game changers and and would be fascinating were got the lowest score.

Murtaza: 14:47

And one of the example I can give you is is, Voiceport. Because when we when we, when when we started looking at these users, we said we we should have a Voiceport, a human like Voiceport, which could and we rate it quite hard because that's a great, you know, way to engage a customer. But, when we looked at the actual return, it dropped, in terms of the detail because most of our customers, they either send us an email or they would prefer to chat because these are lending customers. Right? Or up in a phone means, of course, depending on the different financial product, it could be different.

Murtaza: 15:20

In an insurance world, it would be slightly different where, you know, they they do have these interactions. And then once we did that, we we just didn't, I think, Lance, your second part of the question is, like, you know, LLMs, which LLM was the right LLM. You didn't jump on saying, okay. We're gonna use OpenAI 3.5 or 4 or 4 o or or or Haiku. We validated, what would be the best LLM for that particular use case.

Murtaza: 15:47

And then last but not the least is is, how are we gonna engage with these LLMs, which is where your, you know, data security, privacy, all of those things come in. Right? We need to make sure that we are regulated as well all the information which we are because this information is gonna cut through multiple networks. Right?

Aaron: 16:04

Yeah. Yeah. Yeah.

Murtaza: 16:06

And that's when we we decided to, you know, build our own platform.

Aaron: 16:11

Okay.

Murtaza: 16:12

So Yeah. We've started the journey on building the platform. We're calling it AI matrix.

Aaron: 16:17

Calling it?

Murtaza: 16:18

AI matrix.

Aaron: 16:19

Okay. Nice. It's

Murtaza: 16:20

a interesting name within the organization. But the platform the the fundamental thing about the platform is it is a central place for all JENI use cases.

Aaron: 16:31

Mhmm.

Murtaza: 16:31

And, it will harness the knowledge from, our data platform, which is also we we've just, gone live with our data platform. And that data platform is gonna, constantly send us the information. We'll curate it, and we'll start building these knowledge stores. What we're looking at is automating the the build of the knowledge stores, if we can automate that. And then, of course, prompting.

Murtaza: 16:58

So we have a prompt engine. We have a LLM service. The LLM service, we have some, we have, business logic around it, which can, identify which LLM to use at what point in time, where the channel is, where's the the request coming from. And it's all building blocks pretty much. It's all building blocks Yeah.

Aaron: 17:17

Yeah. Which

Murtaza: 17:18

we'll bring it together.

Aaron: 17:19

And, you know, that that makes I think a lot of organizations are at at that same point where they they kinda know that there's a huge benefit to be had. They kind of know some of the use cases, but I think are predicting that there'll be so many more that it's worth investing in a more strategic platform approach because you don't know what the demand might be, but you expect it to be a lot more than it is today. And, you know, the kind of, you know, for me, the the LLM choice. I went to a talk where one of the people who were assessing them and publishing the kind of results went from sort of tens of of LLMs that they could talk about and understand to 100 of thousands where there was actually no point anymore. They they couldn't there's no point in sort of trying to, you know, describe them in a in a in this kind of, you know, fashion way you get someone's opinion about them.

Aaron: 18:18

You know, you really have to look at them a lot more statistically. And so they stopped publishing their blog about LLMs, you know, specific LLMs. And and, you know, much becomes much more nuanced where the benefits come from. And, you know, kinda like your name, you know, the matrix of of those plugging those different things together. Well, there will be a lot of combinations, I think, and specialized use cases.

Aaron: 18:38

Maybe many of the organizations running, many LLMs, many data sources. Yeah. This is interesting.

Murtaza: 18:45

Yeah. No. Absolutely. And and, again, you know, this has been coming from experience. Right?

Murtaza: 18:51

Because it's very easy to go to the tool, the best LLM and generate your your content or responses or or do your analysis on on whatever use cases you have. But, you'll be surprised, like, some of the smaller elements are actually more smarter if you use it in the proper way. And also you have, just as we are talking about elements, you have vertical LMS now. Right? So you have LMS for banking.

Murtaza: 19:17

Like, you had one from Bloomberg, Bloomberg GPT. And, again, it's a very small model, but it's it's pretty precise in terms of finance. And then some of the, responses you get is actually more accurate than some larger models out there. So, yeah, that would be the advice I would give anyone who's trying to explore, any use cases. Try to pick the LILM which is right for that use case and has domain knowledge.

Murtaza: 19:46

If it has domain knowledge, it makes it easier for you because then if you augment with your own knowledge tools, the, generation of the responses are more accurate.

Aaron: 19:57

Yeah. And, of course, there's no no requirement for a company to release their LN. But, you know, I'd say one of the possibilities is that you use your own history of trading, your own history of responses to produce a very good LN that sounds like and generates responses like the company. And, you know, that that could, you know, use all the email history, for example. I think there's a there's a a point probably worth touching on is the kind of, you know, the the call slot.

Aaron: 20:24

You know, if if we're all generating a lot of content through AI, you don't want that content coming back into your element and, you know, essentially degrading the output. So I think that's probably another challenge that companies or organizations will have to overcome. You know, they'll have to deal with what is AI and what is not AI and what is part of that what what they use as their training base. That will become more and more difficult to to even determine what was AI generated and what and what wasn't. And, you know, I see that as a as a big old challenge.

Aaron: 20:54

I don't know if you're you're thinking about that stuff already or or trying to deal with that in your platform.

Murtaza: 21:00

We we try to. Again, it's very hard as you said. Right? But we we we we try to deal with that through, our knowledge tools and card readers, which I mentioned earlier. Identifying whether it's AI generated or not, I think for us within this domain, it's probably less important.

Murtaza: 21:17

More, I think, is in the academic space.

Aaron: 21:19

Right.

Murtaza: 21:20

And I think in the academic space, it's very, very important whether it's AI generated or not. More of in our side, I think more important is if you can get the the accuracy of what we are generating, what is the use case? And does it adhere to the regulations and compliance? Very, very important because, we're dealing with we we we are in finance. Right?

Murtaza: 21:43

So it's it's

Aaron: 21:44

Yeah. Yeah.

Murtaza: 21:45

Money. So we need to make sure also, they're dealing with customers, and we need to make sure that yes. Yeah. We to answer your question, yes. We are looking at it.

Aaron: 21:55

But,

Murtaza: 21:56

in terms of, higher levels, it's more important we get the data security plans right. Yeah.

Aaron: 22:02

And and perhaps a a a wider topic for another day, but I actually wonder how regulators and, you know, more, industry governing bodies, how well they're keeping up. You know, that that seems like a a massive challenge for, you know, someone who's probably typically quite slow and then, you know, you know, they're watching and and coming along afterwards. You know, imagine that they are looking at AI and financial services organizations being more AI facing services, the regulation will have a job to keep up. I don't know what your experiences so far. Are they are they ahead?

Aaron: 22:36

Are they behind? Are they are they are they pushing you in a in a certain direction? What what is what is the regulator approach at the minute?

Murtaza: 22:43

I would say the regulators have started taking this seriously now. Mhmm. You, they are, accommodating AI, that this is this is AI generated, and this is the areas where we're using AI. And we have the EU act, right, which is coming in in place. So the EU act or EU AI act, I don't know exactly what they call it, but it's coming in, 4th April 25.

Aaron: 23:10

Right?

Murtaza: 23:12

And and the way they have done it is they've identified as, you know, number of risk categories as to which risk category your use case fits into. And based on that risk category, you have to provide certain details of, you know, how you're using, data and how you're you're using AI and what are the nonhuman elements to that, like, use case. And, I think the higher risk areas are more in the human talent management space. So if you're doing anything in that senate in in that space, then, yeah, you will fall into high risk category. So yes.

Murtaza: 23:52

Yeah. They they I I can see that, they're getting more serious, and it's gonna get, even more serious once we step into the new year, once a new comes in. I mean,

Aaron: 24:02

that that's just driving everyone, isn't it? That that is really quite impressive. Yeah. The the amount of effort and, you know, the pace at which all these things are moving is just Yeah. Yeah.

Aaron: 24:10

It's quite incredible. So these these data and Gen AI projects, they can be massive undertaking. So you can pick the by the sounds of the type of investment you're making. I wonder then how you organize yourselves. How how do you create the teams that can be successful?

Aaron: 24:27

What what what's your approach?

Murtaza: 24:30

So, yeah, very, very important. The the team, what I believe is, you need a squad which is, has all the skills needed to execute that use case, which, of course, is a variety. It's a range of skills you need starting from data science, data engineers. You need actual core engineers as well because it's very heavy heavy engineering, because we have the models ready. You need to make sure you have the right infrastructure people as well within sort of, like, DevSecOps team, business members involved in building that use cases.

Murtaza: 25:03

So you have the knowledge of the process as as you go along because because the technology is so powerful, if you don't have the right, you know, business member as part of the squad, it's very easy to to to sway away and start building something which might turn out to be something which is not not usable. So so very important. And last but not least is is to buy it business buy it. So you need to make sure that you have the business buy in, and the right funding in place. Of course, you know, the fundings are now slightly different to what we when when I started the journey, the fundings, we we had a very heavy infrastructure funding needed as well at the time.

Murtaza: 25:47

Excellent. But now, of course, the infrastructure funding doesn't need to be that heavy because you have these models available through AWS or Zoho. And you don't have to jump on a a large model to to start with where you, you know, spending a lot of money during the development and testing side.

Aaron: 26:06

Yeah. I I can see that.

Murtaza: 26:08

But, you definitely need funding in terms of the right skills and talent. Yeah. And and, one of the other things which you need to make sure as well, which which we have learned, is it's, not a traditional, IT deliverable. It's it's a mix of technology and AI. So it's, it is both coming together and, also driven by business because the business is looking for, bringing that change within the organization.

Murtaza: 26:41

Therefore, whoever is gonna lead that delivery of that project, I believe, should be a senior member, within the organization and should be technically savvy as to how the how the underlying technology works because it's constantly changing landscape. It changes every week.

Aaron: 26:59

Yeah. So if

Murtaza: 27:00

you're not up to it, you might go with the traditional approach and realize 3 months, 6 months down the line, what you've delivered is already outdated.

Aaron: 27:08

Yeah. Definitely. So it there's quite a few interesting things, you've you've touched on there. The the it strikes me that the, delivery model, at least you mentioned, is not, you know, kind of typical technology project. So it sounds like it's more you can imagine how.

Aaron: 27:23

It's more aligned as a as a service delivery. Because, you know, artificial intelligence is delivering something artificially. So you actually, you know, kinda it sounds to me like you you hit the nail on the head. It's a business person involved is there to deliver the service, you know, so they had to deliver what the outcome is. And that that's kind of an interesting way of looking at it.

Aaron: 27:46

And, you know, I can certainly see that the infrastructure demands previously, you know, sort of 5 years ago when when we were starting to invest in this space and look at building things, it's all about model management and how you actually go about, you know, managing these things. We're like, you said, suddenly, they're available fairly cheaply. You know, you know, that that just changed the game entirely. And, you know, you mentioned it's changing weekly. I I wonder what what's how do you deal with that?

Aaron: 28:18

You know, that that what are you seeing that, you know, that what do you do? I mean, because you you could end up so distracted that you don't actually end up delivering anything. So how do you how do you sort out the noise and and and end up with a good focus?

Murtaza: 28:31

Yeah. And and that goes back to the original leader why we started with the AI matrix because AI matrix is is, like blocks. It's like Lego. Mhmm. Mhmm.

Murtaza: 28:42

So so as the landscape changes, as the models change, if you have a new model come in without, you know, changing a lot of your infrastructure and and your back end services, you can bring that model into as a Lego and start using that model.

Aaron: 29:01

So just make your technology more flexible. Just make it possible to to plug in and take take pieces out. That that seems quite clever.

Murtaza: 29:08

Exactly. Right. Yeah. And and as you mentioned, you actually got the the thing right. You know?

Murtaza: 29:13

It's it's it's, more of a service we are delivering rather than a project that, you know, this is finished because it's not gonna finish. Yeah. It's the constant going ongoing life cycle. So because you're delivering the service and to be ahead of the curve and making sure that that service is delivering value at all times, you need to make sure, you know, you have a very flexible platform so that you can deliver that, service to your end user whether it's an internal or external.

Aaron: 29:43

Yeah. Yeah. Yeah. Exactly. Yeah.

Aaron: 29:45

The other thing you touched on there was the kind of the squad idea. So it's not not a totally brand new idea, but I I don't think it's that commonplace in financial services. And it strikes me in it to build a squad, particularly in a domain like this where it's quite hard to find talent that knows all the different parts. And that's gonna bring its own challenges. So how are you going about building up the squads and finding the right people and actually creating the right teams?

Aaron: 30:10

That'd be interesting to to dive into.

Murtaza: 30:12

Yeah. So luckily, now we have that in place. So we are in execution mode. Initially, yes, it was, more of education. I think it it it was not about, you know, getting, the the the the business.

Murtaza: 30:24

When I when I went through the concept, Amplify got it straight away. So so it, that was less of a a challenge for us, but more of, you know, how do we execute this now. The model we use is, squads and chapters, which is more of a Spotify model. Organizations like WYSE and and other organizations, they use different models. I think WYSE uses a component and events methodology model, which is again, it's it's a good model.

Murtaza: 30:54

But with with our model, we have scores and chapters. Again, scores are autonomous. So we as I mentioned, you know, we we we bring in all the skills needed for that score. And then they are autonomous so they can run by themselves. The advantage they have is they can prioritize deliverables themselves, based on the, business priorities.

Murtaza: 31:17

So if a business has that specific priority, they don't need to go through, you know, multiple, priority life cycles. They they can make a decision, say, okay. We're gonna change, and and we're gonna make this as a priority and deliver that service. Also, the the services, can be lightweight and can be delivered a lot faster for the business to start consuming and testing it. And also what we're doing is we we have these chapters.

Murtaza: 31:43

Chapters are more, you know, horizontal specialists working in that specific area. And as we deliver these services, they're consumable services, they're APIs, or they're they're the services which is a business. They are available not only to other squads but to chapters as well.

Aaron: 32:01

Oh, I see. That's interesting. So Yeah. You know, that that, you know, but mainly obvious to everyone, but, you you know, as a company, you're in high growth mode. And this this kind of way way of organizing with autonomy, with, you know, shared assets, that that that's kind of because you're in high growth mode, you know, that that won't work for everyone, of course.

Aaron: 32:22

But, you know, the the pace that you're trying to deliver at, that that makes tons of sense to be to be organized in that way where you don't have overwhelming governance to get anything done, but do have enough governance to make sure that it's on the right track and and aligned with the business. Yeah. That's

Murtaza: 32:39

that's quite good. Yeah. No. You're right. We we have that, you know, opportunity to to bring these structures and move faster.

Murtaza: 32:48

However, for larger organization as well, I would say it is possible because, in my previous organization, we did run not 100% but a similar concept of Scott's, but we didn't have chapters. So it is possible for larger organization as well. And other thing which I would suggest is in for larger organization, there's a concept of tribes. So tribes tribes is multiple squads into that trend. So if if you have fairly large organization, then, you can build multiple squads with multiple, specialty.

Murtaza: 33:25

Again, they should be having their own goals as to what their goals are for that squad. So they they they can build that score with that skills needed from, cross functional teams, and then they deliver. And then you can define a tribe where you can have multiple of those codes as part of that tribe. So then they have all those goals coming to a collective goal,

Aaron: 33:48

which could

Murtaza: 33:49

be a strategic goal they want to deliver. So you could execute that in a different way for a larger organization.

Aaron: 33:55

Yeah. I mean, it's funny, isn't it? Like, you know, we've we've both been in IT a long time, and it it it comes back to it always comes back to how you organize yourselves to get work done. You know, technology is challenging. Today, technology is changing rapidly in this area, But, you know, a lot of it is about how you get stuff done, how you organize people, how you you actually make this available in the end.

Aaron: 34:19

Yeah. Good to see that some things don't change.

Murtaza: 34:23

Yes. Yes. Yeah. Yeah. Means the underlying methodology still stays, you know, the same.

Murtaza: 34:27

You still follow the same agile practices whether you do, you know, scrum or kanban, the the actual delivery. It's, the the management of it. So, and if you if you step back, you know, a few decades back, we had, you know, project management practices like the PMI. They they had different kind of models as well. Right?

Murtaza: 34:48

These are more modern, methodologies and more and more execution models. They're more for the senior management to make sure they get the goals delivered and and managed properly. But when it comes to actual delivery, we still follow the same agile principles.

Aaron: 35:05

Yeah. Yeah. Indeed. Now predictions time, I suppose. We've covered the kind of opportunity and the pace of change and the team structures.

Aaron: 35:13

I suppose then that just leaves the future. What does the future look like in financial services and AI? And where do you see it being in the next 5 years or so?

Murtaza: 35:25

5:5 years probably is is quite a long, period for for this, era or this technology. As I said, you know, the the landscape is moving faster than we think. But if even if I say a few years down the line, because I'm still waiting for the launch of Optimus from Tesla. Hopefully, you know, we'll we'll next year. That's AGM.

Murtaza: 35:52

But in financial services, I would see you, what I can see in the next few years is probably, you would have a personalized AI assistant Mhmm. To guide you through, depending on on the products you're looking for whether it's

Aaron: 36:11

It's more like for your job or for your particular type of

Murtaza: 36:14

customer. Profile. Right?

Aaron: 36:15

Yeah. Yeah.

Murtaza: 36:16

And then depending on what what products you're looking for investment, wealth, loans, whatever that is, they can guide you, recommend you, they know your your, history, and can recommend new product. But also at the same time, they might actually go and, research that product for you. So almost like a personal financial adviser for everyone? That's what I

Aaron: 36:35

think is. That's, I think

Murtaza: 36:40

it's not that far off. Maybe give it 2 to 3 years, we'll start seeing those. It reminds me of a movie, actually, Her. I don't know if you've seen the movie. Yeah.

Murtaza: 36:55

Yes. Yeah. It it was ahead of its time. And when I saw that movie, it's quite fascinating, where, you had your own personal assistant all the time. And, and I think we're not that far off, but, the again, with financial services, the the key again in is regulation, data privacy, security.

Murtaza: 37:17

So we will definitely have to make sure that those things are in place. And even if the financial services doesn't have full blown AI assistant, you will definitely see some kind of AI assistant, in the next few years. You probably would have experienced it today itself. Like, some of the, online services, financial services, they do have AI capabilities.

Aaron: 37:43

Yeah. Yeah. And and, yeah, I think what I'm seeing is quite a lot of effort going into making sure you understand that it's generated advice and, you know, that's an effect of it. So this is, clearly gonna keep changing as an area and keep advancing. If you had one piece of advice to give someone investing in this space, what what would it be?

Murtaza: 38:05

Well, I would say it's never too late, you know. Start start as soon as possible, because, you are touching AI tools without realizing it. So even on a on a day to day basis, even if you think that you're not, you actually are touching AI tools in some shape or form, either in your day to day work or at home or in in public spaces. So bringing into your organization, you should bring that as soon as possible. Now when you're when you're bringing that, again, make sure the use cases which you pick quick wins can bring in efficiency, at a very faster pace.

Murtaza: 38:47

So, you know, it can move the needle. Don't go for, you know, high risk and, big use cases because that will burn, your your effort within the organization and

Aaron: 39:00

you'll you'll

Murtaza: 39:01

see the value straight away. So if you're starting, pick something which is nice, small, narrow. It is, high value, low risk, and, start, showing the return of investment on that use case. Once you start that, you'll also learn through a lot of, the journey as part of building that, and that will help you to start picking up medium to high.

Aaron: 39:23

Yeah. And I and I think with your platform investment, you could probably also include get the foundations right. You know, you know, once once you start naming a couple of use cases with a solid foundation, you'll be able to expand from there.

Murtaza: 39:36

Yeah. No. Absolutely. Yes. Yeah.

Murtaza: 39:38

And and and, again, when you if you pick small, you will be able to build the foundations in in in a lot more smaller granular fashion.

Aaron: 39:49

Mhmm.

Murtaza: 39:49

Mhmm. And if you're picking small, you can validate the use case a lot faster before you build the platform itself because you have so many tools available. And you just bring one of those tools to just validate the use case. You don't have to use it. But once you validate it, you know, okay.

Murtaza: 40:06

You know, this is valuable. You start putting those tools.

Aaron: 40:09

Exactly. Then invest. You know, that makes that makes perfect sense. And, you know, as a as a software vendor, you know, obviously, I'm, excited by that. You know, if they haven't started yet, it's not too late.

Aaron: 40:18

The the technology investment. Yeah. It's it's clearly is gonna be part of what a future company looks like. So, you know, the the advice kind of remains the same. You know, get your data foundation sorted, get your technology around managing things, these things sorted, your teams sorted.

Aaron: 40:34

Yeah. This is, this has all been, excellent insight. So thanks, Matasa. Thanks thanks for coming on and, and sharing your your thoughts with us.

Murtaza: 40:42

No. Thanks a lot, Aaron, for inviting me, and, yeah, pleasure to share the thoughts. And, happy to take any questions if you have in future. But, yeah, thanks a lot for inviting me.

Aaron: 40:53

Thanks for joining us on another episode of Data Matters live. Muthaza from Amplify Capital shared his experience of Gen AI, data, and financial services. If you're interested in doing more with your data, why not contact us? We'd love to chat about how we can help. With our data platform and expert support, we think we can get you covered.