We all wonder how to incorporate AI into our workflows! Soo I've recorded as special episode on AI in FinTech Product Development! Join me with Jas Shah, founder of Bitsul, and 3rd time guest in the show as we dive deep into how AI can streamline innovation and create more purpose-driven fintechs.

We take a very practical approach so you'll learn frameworks for implementing AI, from automating admin tasks to enhancing customer research, and discover how to build sustainable, customer-centric solutions that create lasting value

If you enjoy this Purpose Driven FinTech pod, please subscribe in YouTube, follow in Spotify, and leave a 5 star rating apple podcast. Remember to connect in LinkedIn to keep the conversation going.

Let’s get into it!

👉 You can find Jas here

• LinkedIn: https://www.linkedin.com/in/jas-shah/
• LinkedIn: https://www.linkedin.com/company/bitsul/
• Website: https://bitsul.co.uk/

👉 And you can find Monica here:

• LinkedIn: https://www.linkedin.com/in/monicamillares/
• YouTube: https://www.youtube.com/@moni_millares
• TikTok: https://www.tiktok.com/@moni_millares
• Website: https://moni-millares.mystrikingly.com/

We cover:

[00:00] AI as a catalyst for customer impact

[03:05] Purpose-driven FinTech with AI

[06:01] Incorporating AI in product development

[10:06] Practical AI tools for product teams

[14:00] Crafting effective AI prompts

[21:42] Using AI for customer research

[29:16] Building AI-powered solutions

[32:34] Cost considerations and ROI

[38:38] Common mistakes to avoid

[41:48] Ethical considerations in AI

[45:54] Future of open banking

[50:00] Key changes for sustainable AI

SEARCH QUESTIONS

How to implement AI in FinTech products?

What are the best AI tools for product development?

How to use AI for customer research in FinTech?

Is AI replacing product managers?

How to measure AI ROI in FinTech?

What are common AI implementation mistakes in FinTech?

How to build cost-effective AI solutions?

Should FinTechs build or buy AI capabilities?

How to create AI-powered customer experiences?

What is the future of AI in banking?

How to use ChatGPT for product development?

What are the best practices for AI in FinTech?

How to automate product management tasks with AI?

What are the risks of AI in financial services?

How to build ethical AI products in FinTech?

How to use AI for product innovation?

What is the impact of AI on product teams?

How to create AI strategy for FinTech?

How to scale AI solutions in FinTech?

How to balance AI automation with human touch?

DISCLAIMER

Disclaimer: This episode does not constitute professional nor financial advice and does not represent the opinion nor views of my current, past or future employers. The guest has agreed to record and release our conversation for the use of this podcast and promotion in social media.

Jas Shah

===

Realistically, you could create a template and a framework. To do it within minutes, saves 5, 10x in terms of time, and then allows PMs, POs, CPOs, head of product to actually do some creative thinking and figure out, look, how do we get ahead? How do we build a better product? That's the biggest question most organizations are going to have.

And as long as you've got good principles, data structures baked into your software, into your tech stack.

Thank you. Thank you. Thank you. Thank you.

Monica Millares: Hello, Jas! We are going to talk about, of course,

the topic of the year, AI. And more specifically, how do we incorporate AI in our ways of working,

both to build efficiency gains and to incorporate into innovative products and propositions as such

Right now, AI is kind of like, everyone has a fresh start, but some people are already ahead of the curve. So, what do you think about that?

Jas Shah: Yeah. I'd say not everyone, there's, there's no even, even. footing, AI doesn't equalize everything because obviously there's some prerequisites to be able to use AI.

But I think, I think it definitely makes it easier for you for someone to build something quickly and utilize AI to do things like, I'm sure we'll get into it, like create a backlog, for example, that I think an administrative task that. POs and PMs will do and might, you know, I might spend a few days on it, but turning a roadmap into a backlog can sometimes be a few weeks of work.

Realistically, you could create a template and a framework to do it within minutes. If you, you know, if you give GPT or perplexity the right prompt, you could build all that out. So, definitely accelerates a lot of stuff, but some people are obviously. Further ahead, those people with, you know, reasonably solid data structures and plans and mechanisms will be further ahead, but makes things easier and faster to build.

So we know it's really hard to build and grow fintechs data. Actually help customers move their financial wellbeing. That's actually my definition of, of a purpose driven FinTech. But A, it's hard, but B, now we have AI, right? So AI should come and help us a little bit. But before we go into the how it will help us, what's your thinking around?

Jas Shah: Yeah. I, I mean, I think you've nailed it with the, like, It will help improve efficiencies because the, the, when I'm speaking to PMs and when I've worked at companies as a CPO or Fintechs as a CPO before, the thing that I always hear in terms of feedback is I don't have time to speak to customers, or I don't have time to do this, like, like, couple of hours of discovery or, you know, On a new part of the product, I have to, you know, run, run scrum sessions and fill out JIRA tickets and, you know, be on refinement and do all this preparation for refinement.

It will replace the admin part of our work to allow us to do the creative part. And I agree with that. And when you apply that into product thinking and create purpose driven. Fintechs, I think that's the ideal way to, to frame it, to go, look, it can write a JIRA ticket. It can create you a template. You can come up with the creative part and then let AI do the a gen AI layer with an LLM, create the actual ticket itself, and then, you know, ask it questions as you would an engineer, like, you know, pretend, pretend you're an engineer.

How do we build a better product? I think that that's where the, for me, the key part of AI is, is going to change, change product roles.

It's just like, let's use practical tools that are available out there in the market, 50 membership or 20 membership, but across the product team. So that becomes an expense, right? It's an, it's an investment cost wise, but it's also an investment time wise, because we're going to change how people work. We are building the case, based on everything that you said, that AI tools should help us have commercial and customer impact.

We will be able to measure commercial impact because we have our KPIs. How could we measure the impact of AI? on customer impact?

You're making. And as impactful product, as impactful and sustainable product, long term while saving costs, which means you can reinvest that into going back things like customer discovery. But I think it's, I think it's difficult to at the moment do that level of attribution to go, look, okay, we've got this.

You know, we've got this metrics framework now, now we have to add another layer of how is AI specifically helping it? I think it's difficult to, to disconnect the two. And I think you just look at, even if there's a commercial gain with AI, monetarily speaking internally, I think that itself is a long term benefit and creates long term impact, assuming that things like retention rates are the same all things being equal.

Monica Millares: Yes. And I want to add to that caveat, caveat, big caveat, if your commercials go up because of AI and we are assuming that that gives customer. Kind of like has customer impact.

We are assuming that your fee structure and your, and your basically like your, your unit economics and your pricing are done in such a manner that they don't penalize customers for, crazy stuff. So we are taking that as a baseline so that we don't misinterpret this. Exactly.

Jas Shah: Yeah, exactly. Yeah. I mean, it's a, it's, it's a, you know, science experiment, right?

Monica Millares: Exactly. Okay, so let's Deep dive. If we look at product teams, that actually I want to move away from the, from the word product teams, because I think by now most of fintechs work in a squad structure. So it's not just the role of the PM, but it's the role of the squad to deliver these innovative use cases, to deliver the OKRs, basically to help customers and companies succeed.

But when we think about building the product as such, if I am Someone in, in the squad building product, how can we incorporate, AI tools in the product development process?

And I think, I mean, for, for me, I'm a I call myself an AI centris. So, I'm not super bullish and using AI for everything. I don't really use AI in my newsletter. I don't really use AI in anything else other than, maybe validating or collecting some sources and then doing some, doing some refinement of ideas I've already got.

And again, if we go for ideation discovery to design, to build, to launch scale, I think on the ideation discovery side, actually. Chat, GPT, Claude, perplexity, all of that kind of stuff is great to do initial competitor analysis. Like, you know, give me the list of all PFM apps in the UK that are similar to, you know, that fit these jobs to be done, listing all that stuff out coming up with ideas, collating sources white papers and things like that.

What are the first few bits of data I need to capture to make sure that the decisioning for that lending product is as efficient as possible? There's lots of people using it as a starting point. And I think it's an easy entryway into, into incorporating AI activity.

Monica Millares: Yeah. And I think you've raised a good point.

Jas Shah: I, I don't, I don't have a list of prompts. And again, I, I think it's because I. I'm, at the moment, I'm treating AI like, like, I don't, I don't want to hand off expertise and kind of outsource it to AI. So I try, I'm trying to keep as much of that, you know, thinking creativity as possible. And for me, it's like, it's like, I don't know, I think the comparison is like learning a new language.

And it's almost like, instead of learning the new language and actually understanding creative process, it's, I just use a translator. So whenever I go to France, I'll just use a translator. I don't learn French. So I still want to be able to speak the language. So I don't, that's why I don't have a set of prompts.

I might be in a couple of years outsourcing that creative thinking and just having a fixed set of prompts that I could send and get feedback from a GPT from. So, this is just my personal, I think. I'm probably in the minority being the centrist and not having certain prompts, to be honest, as a product person, but I like to go up.

I want, I don't need to go Google that now. You know, let's let, let the let the GPT do the work and then I'll verify afterwards. Yeah, I don't tend to, I'm very much in the middle at the moment, so I'm not like, I don't have a set of prompts. I'm still, I think in a year's time maybe I'll have a set of prompts, but not, not at the moment.

Monica Millares: I, I love that because then basically you're answering the big question, the big fear that people have in their minds that it's like, hey, will the AI replace my job? And what you're saying it's basically, of course it can do. bunch of stuff, but we should retain that judgment and critical thinking and then use AI to help us enhance that rather than outsource critical thinking straight away into the machine.

And just, you know, what, what step 1, 2, 3, 4, 5, 6. And then go through that thinking, and then maybe go steps two and three, I could use AI for, but steps one, I want to, I want to kind of still tune my skills and still, like, exercise that part of my brain. And again, that's why I say it's like, it's like learning a language.

And that's my, I still want to have that expertise and knowledge in the background and still hone my skills. And then lean on it when I think, Oh, this is not, you know, this is not a high skilled thing. This is more of an administrative task. I'll, you know, I don't need to spend two hours again, writing tickets or, or like formatting a roadmap.

Thank you. I'll let, like, I've done two bits of formatting, replicate this formatting across all these 60 features.

Monica Millares: Cool. And I think you just touched on skills. We as innovation teams, not only we need to incorporate AI into our ways of working, but we are expected to build solutions for customers that are AI enabled, customers are also increasing their expectations, right?

Jas Shah: Yeah. I always go back to experimentation. So I think the best way to obscure yourself is to start experimenting. And again, you know, try and incorporate AI in your own efficiencies day to day. If you're not using AI yourself to, to again, generate tickets or, you know, standardize a roadmap, then you won't really know how it works.

Sage have got an AI assistant and I think that they're using AI in, in that way. They're like, well, how can I improve experience for the customer? Like, how can this feature be better and more impactful using AI? And then they've started to kind of. Step into that, into that realm. And I think that that's why I'd say to anyone who's who's thinking, Oh, how do I, you know, how do I bake it into what I'm doing?

How do I bake it into features I'm building for consumers? It's experiment yourself first, and then look at does AI make this feature faster to deliver, more effective in terms of efficiency gains, less bias and more cost effective for the customer as well.

Monica Millares: I like that approach of experimentation and coming back to what you said at the beginning about many FinTech leaders say, I love, I want to speak with customers.

It's about everything that needs to happen before we talk to customers. And it's not about even the planning. It's about the sourcing, the customers, and then ensuring that they are there. And then it happens to session. That's amazing. And then it's everything that goes after from. Basically consolidating the research across all the customers, creating the insights and then incorporating those insights into actionable items for the roadmap or for the company.

Jas Shah: I mean, I don't need to answer that question because I feel like you've just answered it in describing the process.

So, because, so when you get like, and this is the challenge, like sourcing, sourcing customers, you know, creating a template for like sign off for compliance, profiling, then actually doing the research and then collecting it and organizing it, categorizing it and putting it into different ICPs, all of that can be improved.

This is their income level. And I'm going to ask it, like, when I ask it questions, respond as if you are this person. Customer, and you can, you can also generate, generate a scale of feedback that way. But again, you've answered the question, like you've gone through those steps and AI will improve every single one of those steps in terms of actually processing the data and and standardizing it.

And then just creating a summary of like, I sat in on customer research interviews, creating interview notes when you're speaking to. When you're speaking directly to an end user or potential customer, one of the things you don't really want to be doing is just typing. So having an AI AI recording software, record, take the notes.

It makes it infinitely easier for one person to do it.

Monica Millares: Yes, definitely. And I want to go back to the point that you said, Hey, we still need to do the research, talk to customers, but having said that today, there's tools that can help us even with that, if, if, if we don't have the time or the capacity or capabilities I recently spoke with the founders of a company called.

They use ChatGPT, Clor, and LLAMA, and all the list of LLMs, and actually they optimize it such that depending on who's your persona, they will choose the LLM that is most fitted to do that. So, for example, if your cohort is women, there is one of the LLMs that has specialized, or at least optimized, to have, basically to work for BIAS, right?

So, So it's like, I love that, but I haven't used them yet.

Monica Millares: no, no. I had the demo with them, but it's in my, it's in my to do list geeking out time.

Jas Shah: Yeah. So, and then I think we're, you know, we're coming full circle. It's in your to do list. You have to, you probably have to, to try and incorporate more AI To replace admin so that that space to do the creative stuff and to like use a tool like this is is more available.

It's like, well, what can I do to give you 1 hour back a week? So you can go experiment with a user research tool. Or you can review some of the transcripts from the past 10 customer success calls to figure out if, are there something in the backlog that could, you know, that could solve a problem that 10 people have asked for 10 people have called in to resolve.

What, what can I do to, to free up the time? And the answer is at the moment is implement AI as much as possible for tasks for like administrative tasks so that they can, you know, they have more time to do all that stuff. Well, it is a big challenge. It's a big challenge for myself.

Monica Millares: Yes, it is. It's yeah, it is a challenge.

That's why we're having this conversation. Exactly.

Jas Shah: Yeah.

Okay. But what's the use case? This one, this one. Let's assume that we've gone through discovery and we've found basically what is it that we're building and it solves a problem and it's commercially viable when it comes to building AI solutions and designing, not from a, not specifically from a customer perspective, but you know, like the ecosystem, it's complex, all these features are very complex.

So when we start designing. Data structures services, you know, all the, the technical parts behind the product. In plain English, what are the other considerations that we need to take into account when we're building an AI enabled solution that usually we couldn't think about now in a non AI enabled solution?

It uses inherent knowledge, programming and rules, but it's still AI. It's just very, very simple. You know, sometimes I call it dumb, even though it's called an expert system. That's the definition of it. It's a bit of a dumb AI, but the principles are the same. If you go one level up and try and build a machine learning algorithm, it still uses the same underlying structure.

As long as the structure is good. I think the biggest decision for most organizations now is, you know, Do we invest the time and effort right now to build an AI model ourselves? Or do we leverage someone else's leverage chat, GBT or Claude's API? And I think that's the, the biggest challenge, and I think it's the biggest challenge because there's a, there's a huge like.

And if you build, you know, if you build your whole fintech around AI, But that is not, you know, it's open AI's, you know, resources, then there's a potential that if they put the costs of each transaction call up, all your cost basis goes up. So I think that's the biggest debate right now is, do we like, do we see AI as a fundamental part of our business in the next couple of years, in which case we should start building ourselves?

Providing you have the right expertise as well.

Monica Millares: Of course, I think that's a brilliant point. And I want to expand on that because basically you're talking about like commercial viability and technical viability because like, yeah, we can create super fancy, exciting, innovative user experiences, but if they don't make sense commercially, or it's extremely expensive to build them, then we cannot build them.

Right. And then. They will work with a squad. They come up with a proposal with the, with, with the end to end user experience, the proposition, how long it takes to implement. And then when I go to the, to the leadership team, they say, and this is viable because we make money this way. And we have these costs in the cost structure.

Probably someone from leadership will be like, how about the costs of AI? And they'll come up with one layer, but it's like, what are the two to three layers that we need to think about when it comes to cost assessment?

Jas Shah: Yeah, that's a, it's a good question. And again, it's, it's easier when you're outsourcing it.

Five pounds, then it's easy to go. Okay. Every customer uses it. It's going to cost five pounds, but the return is going to be X. So I think that's simplistic. I think it is, it is trickier to go internally and do level layers one, two, three. Again, if you've got a good data structure, and you've got the engineering resources to, to kind of build a model, then it really is kind of layer one still.

That's when it gets into layer two. For me, again, those are usually kind of AI readiness problems. Projects and it's sometimes it's a bit of, it's like a digital transformation project. Sometimes it's you're taking 1 step back. You're taking a big hit on cost to take 4 or 5 steps forwards. And usually what organizations who don't have the data structure in place will have to, will have to do that work anyway.

But I think it's, it's complicated to do it at a feature level. What I usually do is look, what is the return on investment, for example, for this specific feature? Is it, you know, is this, this is a feature that's just not going to, it's just, it's like an aesthetic feature. So it's something that maybe pulls customers in or make or differentiate our product, but doesn't necessarily see a return.

It's super tricky. I would always go, look, what are we going to get? And then maybe speculate on the upper band and then be pessimistic on the lower band and then go like, what's the average cost of an API call using chat GPT or Claude or perplexity? And then we'll just use that as the baseline.

Monica Millares: Yeah, and I think you used the right word, like it's super tricky and I want to go deeper.

Because we have like future vision. So we just talked about cost, but they look, they look super cool. By the way, I got distracted. So we are. basically analyzing if this AI solution, if we should build, the infra and what's the, the, basically we went into the detail of the cost assessment and you said, Hey, but this is, this is tricky basically to assess.

So my question is, what's the biggest mistake or sets of mistakes we can do as squads when it comes to doing.

So I'd say, again, the biggest mistake is not, not looking at what return are we going to get? And even if it's assessing that we're not, we don't expect to see a return on it. You know, this, this is, this is an internal efficiency gain. There's going to be no direct revenue, revenue inflow off the back of this feature.

I think that's the biggest mistake is not not doing full due diligence. For me, it's always the biggest thing. Oh, let's build this thing. Cause it's cause someone else is building it. It's like someone else might be building it because they're already set up to build it. Like again, RAMP, I've, I've got a document, AI powered document scanning feature.

So I think it's, it's doing due diligence and understanding, like, what are we, what are we looking to get out of this? I get it. It's measuring as well. Like, what are we looking to get out of this? How do we measure success? How do we, when do we review? You know, review after a month, two months, three months, four months, is it successful or not?

Whether that is revenue or whether it's an internal efficiency, and then going back and going, should we scrap it? It's not worth the cost of running it. I think that's, yeah, that that's the biggest, the biggest mistake most people make.

Monica Millares: And yeah, I'm like, all of that resonates a ton, I'm like, yeah, I'm like, yeah.

Needs as such as customers. So this is where the debate in AI comes in on ethics. If we say standard blank statement out there, if we were to say, Hey, many of the AI solutions were created, in a male dominated environment by privileged people who have access to AI and the education, not by the underserved, how do we use AI?

in a way that we incorporate the ethical side and we start building for underserved customers.

And I think they have, I mean, they did a better job of it than Apple did with the Apple watch years ago when they had a, you know, a sensor, which was Basically, the whole sensor was designed for a specific, skin color range, and it didn't work on darker skins. And so I think AI has done a much, the models that I'm using that have been trained over the past couple of years have done a much better job, frankly, because it pays to have as much data coming into it as possible.

to build a PFM app that gives insights to a specific group. It's easier to use AI to make that way more diverse because you can go create different insights for all of these different profiles and make it more personalized. And so I think that's where AI has got a bit of a, has got a big benefit to create inclusion is that you can create a product that fits so many different, so many more different profiles.

Not too high a cost basis, whereas previously you'd go, Oh, I'll, I'll start with this niche and then I'll slowly, you know, open the bounds and go to this group and this group and this group. Now you can do it a lot faster using AI as an accelerant.

Monica Millares: Yeah. I'm loving the conversation. I think we could go for another half an hour or one hour.

Monica Millares: Okay, cool. Then let's, let's continue then. So,

let me ask you like three more questions because I think we're in a good role. Okay. So I think we could go a little bit deeper now. And one thing is to see one of the things that I love doing as a product geek is basically you see a piece of news that so and so FinTech just rolled out this feature and you check it out and you're like, Oh, that's cool.

And you start kind of deconstructing it or your, your. own fintech app that you're using, they roll out a new feature and you're like, Oh, that's cool. And you start analyzing it. So I'll put you in the hot chair. What is one of your favorite AI fintech products in the market now, such that then we can deconstruct it a little bit.

We've got savings account for you. I think that's 1 of the better use cases. If it is a life, this is just speculation. Yeah, because it it's again, it's kind of ticking all the boxes. It's like all of the things that people and I. Optimists and bullish people say, oh, it's going to take all the admin away.

You just like automate and hit the brain button and it will do everything for you.

Monica Millares: Wait, we need to deconstruct that because this is cool. All my friends have been asking for that feature since many years ago. They are like, I have all my systems, but I want to automate this. And basically this is what Clio is doing.

So if I step back, then basically you're saying, Hey. They've built a money management solution where I put in my salary, let's assume that it's my salary that comes in, and then they pay my bills automatically, and then they move money to my savings, and then they tell me how much I can save. Did I get that right?

But I'm not sure if they're doing it wholesale. And I think the other, the other challenge is not them. They're ready to do it. It's the underlying infrastructure is not. there yet. So again, we can go into the open banking, open banking, open finance debate, but yeah, open banking is really, it's account, it's a current accounts and savings accounts, really some credit cards, but it's not, it's not every type of financial account.

And then once there's a 360 view of everything and the, and the technological capabilities there and VRP is you can automate payment initiation. You could automate. You can do rate chasing. So for example, I've got a savings account. That's 4 percent in HSBC, but I've also got a savings account.

That's 5 percent in that West. It just goes, look, we have all of the metadata on all the accounts. We're going to sweep the money to the 5 percent account. Let's take everything out of here. We'll take the hit on. on early withdrawal. We'll move it into here for six and after six months you'll make more than you do if you have it here.

It's more the underlying infrastructure has not moved at a pace that most of them expected it to move. And I worked, I worked for one years ago and I was like, yeah, I did it. The issue isn't what we build, we can build all of this cool stuff, but if the account coverage isn't there, that's the biggest, you know, the biggest pain point is trying to get access to all the accounts, that's the biggest cost, but once Open Banking is a bit more Open Banking moves into Open Finance, moves into Open Data, that's where I think these apps will make a killing.

But it's not as easy as that. So what do they need to do to, so that they, then they can enable fintechs like us, like the consumer finance fintechs.

Jas Shah: I mean, who do we need tonight? It's probably regulators and it's probably also banks because really to create an interoperable protocol, you need to have full buy in. And I don't think we have full buy in from everyone because most, because most backs have kind of seen it as a, Oh, it's a bit of a threat.

And now I'm being penalized. So I think it's, it's a bit of the regular layer. It's a bit of, open banking, organizations. And then it's a bit of banks.

Monica Millares: Amazing. Yeah, let's finish with the very last question. So, what is the one change that we as FinTech leaders should embrace to build sustainable, customer centric and AI powered businesses that create lasting value for customers, teams, and investors?

You can use AI to prioritize features, launch. You can use AI to kind of tailor a marketing campaign growth. To, to kind of monitor metrics and give feedback, scale, spot, spot opportunities and optimization. But really the starting point is, is ideation, even for an existing FinTech, ideating, ideating a new set of features, ideating you know, a new niche for customers, using AI to, to make that process as efficient and, and kind of all encompassing as possible is, is I think the one change that everyone should make.

And it's one change of, one change that I've made as well.

Monica Millares: Amazing. Yes. It's been an absolute pleasure having you in the show as usual.

Jas Shah: Thank you, Monica.