Latest / 10X AI with Julius Neil / Beyond the Algorithm: Why Behavioral Science is the Secret to Profitable AI in Banking
Transcript
- Julius Neil Buenconsejo: Good day, everyone. I'm here with special guest from Nairobi, Kenya. So it's one part of the world that I would like to explore also about the AI adoption and AI transformation when it comes to private sector and the public sector. So Peter Onyango is a data scientist he's an expert when it comes to ⁓ banking and AI transformation. Peter, can you introduce yourself to our viewers listeners? PETEE: Thank you very much, Julius for hosting me here. This is actually my very, very podcast. I've never done this. But thank My name is Peter Onyango. But I am in banking I do study artificial intelligence right now at the University of Liverpool in the United Kingdom. Julius Neil Buenconsejo: Yeah. Mm-hmm. Alright. you describe yourself as an AI who enjoys working at the intersection of machine learning and real-world business challenges. So can you explain to our listeners and viewers what those challenges in machine learning that you are trying to solve right now? PETEE: Yeah. So right now, my work in the banking, or rather in the financial sector, one of the things, or rather one of the main challenges that we have right now is trying to improve customer experience. And trying to move away from the old days of ⁓ interacting customers on a one-to-one basis using relationship managers. So we are at an intersection of using technologies like machine learning, ⁓ AI, ⁓ considering tools like natural language processing, etc. to be able to find interesting ways of actually connecting to customers. And when I say connecting, it's basically just understanding how our customers transact, what it is they do, what patterns of data they exhibit. And we now sit behind the scenes trying to figure out what is the most relevant message and at what relevant timing can that customer actually be reached out to. So it's just a way of being clever on how to scale banking with the tools that we have right now moving away from old ⁓ to new technologies. ⁓ Julius Neil Buenconsejo: Okay, so you already described that you're moving away from old tech to the new tech. Which type of technology are you now in your current acquisition or customer service platform? PETEE: So right now, we work ⁓ with a ⁓ scientist. So ⁓ what that basically means is using machine learning, we create predictions, if you will, of what a customer is likely to do. And behavioral scientist is now tasked with finding out the contextually appropriate ways of reaching out to a client. What that means we try to not over-message our clients. We try to not bombard them with how will I call it. When you use a recommendation engine and you have quite a number of predictions, so we try and make sure that we prioritize the type of predictions or rather the insights that actually reach out to clients. So ⁓ it's a dance. It's an intersection not necessarily ⁓ steeped technology, but we try and use not Julius Neil Buenconsejo: Yes. PETEE: other techniques that go into understanding customer behavior better and well ⁓ that we can reach out to customers. Yeah. Julius Neil Buenconsejo: So based on your journey of this transition from going from old tech to the new tech, what ⁓ of behavior that ⁓ you when you found out when working with the behavioral scientists? What types of customers that your company or those people who are in Nairobi, Kenya are when it comes to behavioral patterns? PETEE: So one of the things that I've come to realize in the old ways of banking is we assume that customers ⁓ what products we have. And insights we've had, whenever you actually write a recommendation engine or a model or a predictive model, if ⁓ you will, thing is ⁓ when we ⁓ work in with a behavioral scientist, we are able to see that the insights that we've had, whenever you actually write a engine or a model or a predictive model, the thing is ⁓ we now ⁓ work in with behavioral scientist, we are able to see that Julius Neil Buenconsejo: Mm-hmm. Mm-hmm. PETEE: just reaching out to customers with basic information, if I can call it that, actually has a multiplier effect just reaching out to customers with basic information, if I can call it that, actually has a multiplier effect on customers interacting with you and your platform. So it's not necessarily the heavy mathematics and the heavy statistics that actually gets you over the line. It's using that heavy mathematics and heavy statistics to now get Julius Neil Buenconsejo: Mm-hmm. Mm-hmm. Mm-hmm. PETEE: back or rather uncover the basics of what customer relationship is. So yes, you could have a very nice model, a logistic regression model, whatever it is, XGBoost that has the best metric ever. But if you actually now use that to improve business outcome, and improving outcome is basically reaching out to customer, just ⁓ making that every touch point in their life has been ⁓ noted. you have a very high chance of multiplying them interacting with you as compared to you interacting with them. Julius Neil Buenconsejo: Okay, so many data scientists focus purely on technical excellence. So you're deeply interested in workplace dynamics and decision making. Where did that curiosity come from? PETEE: It actually came from a project that had been handling. Again, if you remember, I mentioned the word assumption. So as data scientists, we are taught to assume, or rather to prove or disprove theories or hypotheses. But the one thing that nobody ever teaches you is how humans actually interact with this technology that you are putting ⁓ across their face. So yes, you have built a very nice model, a very nice system that does everything that it was supposed to do. But the one thing that you overlooked was actually interacting with the human being who's supposed to use these two. And the one thing I've come to see is that this human being comes to this chain at the tail end ⁓ this system is already live. Yes, I know say we have Julius Neil Buenconsejo: Mm-hmm. Okay. PETEE: technical terms like user acceptance testing and forth and so forth. But if you think about it, UATs usually ⁓ done a subset of people or of individuals. Not everybody who's going to use that tool actually has a view of what is being built. So it was very interesting when trying to increase the adoption of this particular system that we were working in. there was quite a lot of resistance from my colleagues and us in my department and vice versa. Cause we could not really understand why they were not seeing the advantage of what it is we bringing ⁓ to them. that's now ⁓ made me now understand there's a whole regard to change management that we do not understand or rather we do not know. Julius Neil Buenconsejo: haha PETEE: I was very, interested in how then the interaction between technology and humans actually plays out. And with that, that's how ⁓ I ⁓ have now come to the intersection of trying make sure that there's quite a lot of adoption behind tech. So I become a tech evangelist, if you will. Yeah. Julius Neil Buenconsejo: ⁓ you, since you are doing your masters now, can you describe ⁓ is your research about or what is your dissertation or case study in your master's degree right now? PETEE: Yeah, so I'm yet to do my dissertation, but one of the ideas that I have that I'm going to toy with is how to use agent AI to improve customer experience. And while I'm saying that, you look at it from a scalability perspective, mostly with the number of customers and the number of staff members who actually take care of customers, staff members being relationship managers, what we see is that there is The ratio is tilted to one side, right? There's an imbalance in the ratio. So the only way we can try and figure that out or solve for that try to ⁓ or supplement the relationship manager with a tool beside him. So how is the agent able to now do that what an RM does, a relationship manager does, make him a virtual relationship manager in terms of just understanding his customer's portfolio, understanding what his customers want, and then Julius Neil Buenconsejo: Okay. PETEE: The agent is now able to package all that into one, into one sort of ⁓ communication, be it an email, an SMS, or whatever media we use, and then reach out to the client. So I'm trying to see how we can now use such kind of tech to scale operations compared to relying ⁓ an individual human being reaching out to ⁓ thousands at times of clients. Julius Neil Buenconsejo: Okay, so when you're doing all this sub-behavioral analysis of all your customers, what are the top things that you think that you did wrong before just by assuming that the customers wanted this? Top three things that you discovered during your journey. PETEE: Yes So we bombarded clients with everything and anything. We didn't necessarily understand their business patterns. And pretty much it, to be honest. ⁓ So we ourselves just sending ⁓ emails SMSes about all the products we have. Now we're at a point where we are properly understanding the probable products that customer will take up. And in doing so, Julius Neil Buenconsejo: Okay. PETEE: just now understand ⁓ to reach out to customers. So the seasonality pattern of it all. those are actually those are the only two things that to be honest, we found out. Yeah. ⁓ Julius Neil Buenconsejo: Okay, really? Okay, so yeah, interested in that. So how do you predict now the behavioral patterns based on the season? PETEE: Yeah, so old techniques. ⁓ still have to use the same old time series analysis. But in time series analysis, that is the technical bit. Put the technical bit to the side. I think that the one thing that actually takes up most of your time in doing seasonality, or rather, understanding the behavior of a client, is actually interacting with the client themselves. So how do we do that? Julius Neil Buenconsejo: Okay. PETEE: you either now rely on the relationship manager, where they come up with what you call call reports. Call reports are basically reports that they have put in a particular CRM. you now ⁓ AI or ⁓ you use Natural Language Processing, to understand the patterns between what the RM ⁓ communicated the customer. So if you augment use AI or ⁓ if you use NLP, Natural Language Processing, to understand the patterns between what the RM ⁓ communicated with the customer. So if you augment Julius Neil Buenconsejo: Mm-hmm. Mm-hmm. PETEE: That type of analysis with the time series analysis that you see from data patterns, you are at a much better place of uncovering behavior or behavioral insights That type of analysis with the time series analysis that you see from data patterns, you are at a much better place of uncovering behavior or behavioral insights about your customer. Julius Neil Buenconsejo: Okay, that's good. So have you already implemented this in your organization now? These new systems that you have PETEE: Yeah. Yes, yes, yes. have, though some of them are still working in progress. I may not necessarily be at liberty to speak about all of them, but we are in the process of implementing ⁓ such systems. One of the projects that I'm currently doing is to carry out sentiment analysis on the call reports that I mentioned from relationship and customers and vice versa. Julius Neil Buenconsejo: Yeah. PETEE: So what this now helps us understand is the sentiment our clients actually have about our products and even the level of experience that they've had interacting with us. it now, again, removes that assumption that I mentioned. The assumption that comes in rather the bias that is introduced when you only look at the data transactional patterns, but now you actually understand the customer from what it is they see from, may not necessarily be verbatim. But at least you have a view of what the customer said when they had that interaction with somebody on our team. Julius Neil Buenconsejo: then Peter: I Julius Neil Buenconsejo: so this one thing that I already discussed with you before our recording, ⁓ about AI adoption in as a whole. So how is it? For example, in the public sector, in the private sector, and then individual people who are using AI, is it already going mainstream? is it very less people that still knows AI? Peter: ⁓ Actually, quite a number of people do know about AI, but when you talk about AI, we are talking about tools like ChartGPT, Gemini, and the like, right? Yes, we do have developer communities that do go into the tech side of things and build applications using, heavy applications using AI. But from a general sense, yes, ⁓ do have quite a lot of individuals now using tools like ChartGPT, Copilot. Julius Neil Buenconsejo: Yeah. Peter: Gemini and they like to do, to basically do everyday things. So we have marketing companies now coming up with adverts. Of course we have some of us who use tools like that to code a bit better and faster and so forth and so forth. So yes. ⁓ fact, actually Kenya is one of the countries leading in the adoption of ChatGPT in Africa, if I'm not wrong, if not the whole world. Yeah, so yeah. Julius Neil Buenconsejo: Mm-hmm. Okay. Okay, okay. Peter: we are getting up there. Julius Neil Buenconsejo: Okay, okay. So when it comes to government led activities when it comes to AI or projects, for example, in my previous podcast, one ⁓ of experts said that Saudi Arabia have their own publicly funded institution that gives online courses to their citizens to be able to adopt and learn about AI. How is it in... Kenya or in Africa as an Indian. Peter: Yeah, so the thing with adoption in Kenya and Africa as a whole is we are seeing a number of countries actually coming up with their respective strategies on how to adopt, use, and build AI. Recently, Kenya came up with Kenya National Strategy that was designed or rather developed in 2025. That was last year. And the whole gist of this particular strategy is to give a framework. on how AI development is to be done within the country, how the ethics, they basically research on the ethics and the privacy matters of implementing AI, and also they go about discussing how to adopt AI. Generally, sorry, overall in Africa, we are seeing countries like South Africa, which is the leading country right now, I think, in Africa in adopting AI, and also just in some of the use cases they do have. Egypt, Morocco, Nigeria, some of the big powerhouses ⁓ the field of AI. As continent now, we have EU now coming up with a continental strategy that again also just tries to put a framework on how ⁓ the respective member states to develop and AI within the legal framework of ethics and practice. Julius Neil Buenconsejo: Okay, okay. So Kenya, specifically, so you're in the banking sector, so you know that AI is already being integrated into your processes and systems. How about in other sectors? Do you see some other sectors already doing this also, integrating AI into their systems? Peter: Yes, yes, yes. So for example, in education, we now see schools, quite a number of schools, universities rather, now come up with education curriculum, if you will, that teach AI and AI skills, related skills, data science, machine learning and the like. We have research and development ⁓ and startups that ⁓ basically on researching on how AI can... be permeated into the society. to note is ⁓ company called Oshahidi. Oshahidi basically deals with data to ⁓ in humanitarian efforts, right? So these ⁓ from a local perspective, we do see quite a number ⁓ institutions or other industries from the financial or banking that are heavily using AI and trying to Julius Neil Buenconsejo: Mm-hmm. Okay. Peter: work with AI within the regulatory framework defined by the national strategy that I earlier mentioned. Julius Neil Buenconsejo: Okay, so when it comes to startup ecosystem, so different countries have different needs when it comes to AI. So ChatGPT is a worldwide, generally acceptable AI platform. So how about in Kenya? Are there startups also doing AI that is specifically localized or customized based on the needs of as a nation? Peter: Yes, ⁓ quite a number of startups that operate in Kenya lie within the financial space, FinTechs if you will, they predominantly use AI and machine learning and the like to advance credit facilities to the local populace. So when I say credit, it's basically how can we advance loans to unbankable. Julius Neil Buenconsejo: Mm-hmm. Mmm. Peter: in as much as we are a country of 53 million in population, just not of 53 million, but not everybody is bankable, to speak. So what these fintechs have come to do is to plug that gap. And how they plug that gap is to use machine learning and AI to basically understand the limits, credit limits that can be fronted to a particular individual. Case in point, we have ⁓ facility called mShori. Julius Neil Buenconsejo: Go ahead. Peter: of Fulisa, the way Fulisa works is have a mobile wallet. So on your mobile wallet, in the event you ⁓ the funds in your mobile wallet, it automatically kicks in, the limit kicks in so that you're able to continue transacting within the particular limit or guidelines that they have, let me say, predicted you for, or that has coped you, scored you for. So yeah, ⁓ most the ⁓ startups Kenya. Julius Neil Buenconsejo: Yeah. Peter: lie within that financial space. Julius Neil Buenconsejo: Okay, since you are in a banking industry, you work in a bank, so bank has the resources or the finances to be able to this type of projects. What do you think are the challenges of the ⁓ companies when it comes to AI transformation in their organizations? In Kenya, basically. Peter: Yeah, so one of the things that may actually hampers is ⁓ cost of infrastructure in as much as yeah, there are quite a number of initiatives that are being led by the government. So ⁓ an example, the has a project to lay over 100,000 kilometers of fiber optic cable ⁓ by 2027. ⁓ that being said, do know how the compute computing ⁓ AI is very, ⁓ intense. So you have to run up costs in terms of accessing services like AWS, data centers and the like. So as much as we do see quite a number of individuals other startups or other embracing AI, the one thing that actually will come and bite us in the back, if you will, is the cost the infrastructure that actually needed to ⁓ scale up ⁓ some these initiatives. Julius Neil Buenconsejo: Okay, okay, okay. So having said that, know, when it comes to AI transformation, it costs a lot of money. In your company, how do you measure the ROI now? So since you have this behavioral scientist, they have already made a model for the framework, how to understand your customers very well, how to give them all this information during the season, off the season, and all those other, you know, behavioral patterns that your customers have. So... What do you think about that? Peter: Yeah, so that's a very good question. It lies in the intersection of business and technology. So it boils down still to the KPIs that have been set, the key performance indicators. And the key performance indicators is not necessarily on how many models you will build, but it's ⁓ built around measuring the impact or the customer impact that we have. So for example, how many have we had? And when we measure the number of interactions, how now does that translate into the balance sheet, right? How we interacted to X number of customers and out of those customers, have they been able to transact, right? And from that, it also now boils down to the financial metrics. So you have to have a mix of both, a mix of the technical in terms of what you're measuring, what you're and also on the financial bit or side of things. Because at the end of the day, businesses exist to make money. So yes, you will have the best models and what have you, but if it's not making money downstream, yeah, ⁓ not necessarily being impactful with your particular innovation. Julius Neil Buenconsejo: Based on the projects that you have now, for example, the first few roll outs were you able to get some customers to go on board with certain products or packages that you are offering now? Peter: Yes, yes, yes. So what we see is product recommendation engine that we have built is actually doing very, very well. Again, what ⁓ I, I said earlier, the ⁓ assumption we had when onboarding some of these clients is that they understand how the nature of banking and banking products. But now when you sit behind the scenes and analyze their transactional patterns and recommend a product to them, that... ⁓ reach of recommending that product and notifying them or other nudging them as my behavioral science, scientists colleague will say, ⁓ them to take up that product actually ⁓ works wonders. So it doesn't necessarily have to be very, very technical, that piece of ⁓ analyzing recommending a product based off ⁓ what we ⁓ the customer take up. actually has had a positive effect on customers interacting and trust acting with us. Julius Neil Buenconsejo: What is the behavioral metric that ⁓ your behavioral scientists are using or the variables to be able to see? For example, some of the variables that you have, maybe one or two or three variables that you saw, that they are measuring based on the algorithms or the technology that you are using. Peter: Yeah, so one of the most ⁓ ones that I've seen is ⁓ from the product world, monthly average users, daily average users. ⁓ when look at that, ⁓ the of customer inactivity. So when you look at all these metrics in tandem or rather in togetherness, we're able to see Julius Neil Buenconsejo: ⁓ huh. ⁓ huh. Peter: from the point we interacted with the client, right? Or rather we have initiated that interaction with the client. How has their average use of our products and channels been? Has it increased? Has it dipped? they ⁓ least done one transaction with us? Are they towards inactivity, dormancy, and what, or ⁓ if you will, activity, right? ⁓ Julius Neil Buenconsejo: Mm-hmm. Okay. Peter: which is the better of it all. So the whole idea of it is just, a nutshell, if it just to put all that into one, the behavioral scientist is more interested knowing whether the customer is active or not active. From that particular point, of course, the other engagements do kick in and take over. Julius Neil Buenconsejo: Did you also analyze the time when they are logging in? Which time of the day? Most men will log in, for example in the morning, most women will log in between office hours, those things. Have you analyzed that? Peter: Yes. Yes, yes, yes. That was actually one very interesting project we did. And what that helped us understand was when was appropriate time to send messages to clients. So again, as I one of the things we do a lot is making assumptions. So of course, we used to bombard clients with blasts at a time of the day. But the moment we actually understood the ⁓ most opportune time. that some of these clients actually interact with us. That is when now you send, or rather you nudge them to do something with us, or you nudge them to do a transaction with us. And the funny thing is that at that particular point, just when the customer is trying to think of that transaction, and then they find an SMS or an email from the bank or from the institution, the probability is very, high that they'll actually interact with us at that particular point in time. Julius Neil Buenconsejo: that's really good to hear that you're getting a lot of success when it comes to your AI ⁓ transformation and integrating that into your system. now, going why do think are ⁓ your projects your AI transformation initiatives are very successful currently at this stage? What are the variables or ⁓ what is environment that your company or the banking system that you have that enables you to be able to be successful in... rolling out these AI initiatives. Peter: out gate, you need to have ⁓ a buy in. You can only necessarily just implement a project just because you learned about a particular model, but you have to make that ideology as simple as possible for a layman to understand. And when you talk about the top leadership, I'm not saying they do not necessarily understand what goes on behind the but you have to make it very easy for them to understand. how this particular project will impact the bottom line. So once you have that ⁓ top level buy-in or leadership buy-in, it's very easy for you to now cascade the impact of the, or rather the project itself to whoever you need to implement it to. So one of the things again, if you actually now remove employees and colleagues out of the equation and reach out directly to the client. you actually have a much easier time getting the impact or assessing the impact of your experiments. And what I'm saying so is when you try and go through ⁓ third party, if you will, to reach out to the client, you get some friction whereby somebody will come and say, ⁓ not necessarily understand this particular insight because I understand my customer better than what you're trying to put across. ⁓ Julius Neil Buenconsejo: Mm-hmm. Peter: is if you actually go to the client directly, the rate of adoption of your particular metric or initiative will be higher than ⁓ you do not necessarily go ⁓ a party field. Julius Neil Buenconsejo: Okay, okay. So you said that leaders don't need to understand algorithms but must understand the implications, right? So what does understanding implications actually look like in your perspective in practice? Peter: So understanding implications is one, understanding the finances. The pill that I had to swallow is understanding how algorithms into money. I'll just ⁓ put out there. Yes. You actually to understand how to ⁓ translate your metric, initiative, algorithm, what have you into the ROI. What is the return of investment in me? bitter pill that I had to swallow is understanding algorithms translate into money. taking up this particular project, right? and above that. Yeah, think that's pretty much it actually. You'd actually just need to understand the money. ⁓ Yeah, once you understand how we get money from initiatives, it now becomes easy for you as a data scientist to now go back to the methodology that you want to put in place. Julius Neil Buenconsejo: No I didn't. Peter: It now helps you understand whether you need heavy algorithms like LLMs computer vision, if you like, ⁓ you can actually just stick to the basics of linear regression, logistic regression, XGBoost and the like. And to that bit, the explainability of it ⁓ all. You be able to explain the results of your model. ⁓ Nobody understands these models work apart from us who are technically trained. Yes, we understand that we need to... optimize the metrics, the hyperparameters and what have you. That's fine. But the layman does not necessarily understand what linear regression or logistic regression and respective assumptions hold. So you have to understand that from a business perspective. Put it in English, you will, ⁓ lay English, if you will, and ⁓ that now your work easier in trying to have people adopt your initiative. Julius Neil Buenconsejo: Yes, correct. Okay, then you describe also AI's transformation as a sense-making challenge. So can you the idea for leaders listening today? What is the sense-making challenge? can you unpack this idea for leaders who are listening today? mean, can you elaborate on that? What is sense-making challenge? Peter: Yeah, so whole idea about making is how do leaders understand what ⁓ AI initiatives are, right? Yes, everybody talks about AI, it's the new buzzword right now, but AI is a whole ecosystem of things that is not necessarily limited to ⁓ tech Chat GPT and the like. So the whole idea is for leaders to understand the impact of these technologies. to their bottom line. So it's them to understand the nature of business that, of course understand the nature of business that they want to run, but now they have to incorporate that with the vision of the company. So how do we incorporate the vision of the company with what technology we have ⁓ for the and ⁓ marry those to be able to evangelize what it is, whatever roadmap. ⁓ that we'll have to be able to take advantage the ⁓ technologies that we have. So it's not necessarily a matter of them pushing down AI down your throat, but it's a matter of them understanding how AI and its respective tools tie into the vision of the company and of the business itself into taking us to the next level. Julius Neil Buenconsejo: Anchoring into that, so for example, if you are advising a CEO in Nairobi right now, launching their first AI initiative, what are the three conversations they must have before deploying anything? Peter: Change management is very, very, very key. And in that change management, it boils down into this. ⁓ and his top leaders, him or her and his top leaders have to understand what AI and what AI can do for them. It very, important to have change agents or change champions who will now help him and help the CEO evangelize the ⁓ message there on what AI is to his respective individuals downstream. And three, it is very important to have employee buy-in. It is very easy for a CEO to actually just cascade the vision and say, this is the direction we are in. But one thing that actually aids in adoption, or rather faster adoption, or scalable adoption, you will, is if there is employee buy-in into some of the technical initiatives that we have. And being said, that enables... every single individual employee to have a say in the nature of work or rather the proposed nature of work. So that also reduces friction in the adoption of AI and the respective AI tools. So those are some of the things that I would actually recommend to a CEO. Understand AI and what it can do. Don't just cascade it down. Have change champions all over the place. and ensure mostly and importantly have your employees buy into initiative. Julius Neil Buenconsejo: Yeah, so you already about the employees buy-in, but the problem with that is must be some psychological safety, right? Because psychological plays a role in adoption. So this is already the buzzword in AI now. So how does fear of job loss ⁓ or of wrong or being misreplaced silently kill AI initiatives? For example, in your organization now, when you are just implementing in some systems, Do some of you or do some of the employees already thinking, okay, maybe in the next projects, they will use AI to replace me doing all this repetitive tasks that AI can do better in less time. Peter: Yeah, definitely. I like that you say psychological safety a new buzzword. Change is ⁓ hard. And not managed well, it ⁓ will the doom of your project. But if managed pretty well, it will actually spell the success of your project. So being psychologically is not something you just say and everybody magically just becomes psychologically safe. ⁓ You actually to have Julius Neil Buenconsejo: Yeah. Correct. Yes. Peter: towards it, know, as much as yes, I understand that it may be very, very hard to have individual employee buy-in, but it is more costly to not have their buy-in now you have ⁓ of sabotage, instances of psychological as you said, and so forth. So it is for us who are in positions of ⁓ evangelizing and respecting and their respective tools to be able to help guide that particular change and dissolve into ways that is easy for people to understand the nature of what it is we're trying to propose and how it augment and not replace them. So it's ⁓ again, change is very, very hard. If I was to and tell you that I'm moving you out of your house right now. without necessarily telling you that I'm moving you into the app. Of course, you'll be up in apps, but if I actually explain to you why we are moving into a bigger house and so forth and so forth, it will be very easy for everybody down the line, whether you go top down or bottom up, to embrace AI change and psychological. Julius Neil Buenconsejo: Hahaha ⁓ So from your experience currently, so how is AI adoption in Kenya different from what we see in the US or Europe or the rest of the world? Peter: the thing that I see that ⁓ let say we are trying to do is when you look at the US and Europe and some of these countries out there, they start embracing AI at a very young age. You can imagine countries like China, Japan and the like, right? they start experimenting with AI. from a very young age when still in primary school. Over here, we are yet to embrace such kind of systems in our education, right? Though, as I had mentioned earlier, we do have curriculum in tertiary institutions, universities, colleges and the like, but what these other countries do differently is that they start educating their populace very early on ⁓ at the of their lives. compared to us now. When you put that into context, by the time we catch up to that particular race, countries already have years of experience over us. So by the time we start evangelizing what AI and AIK can do, somebody already has 10, 20 years of education on the same. So clearly, ⁓ we won't be at par in terms of adopting such kind of techniques. you compare that with other countries. Julius Neil Buenconsejo: So since you're in the banking sector, so in the Kenyan banking sector specifically, where do you see the most realistic opportunities for AI over the next maybe five years? Peter: Custom experience. I know right now we use AI to improve for fraud credit scoring, and the like. But custom experience is going to be a heavy game changer for us. And what I'm saying, so we have quite a number of people who are banked by quite a number of banking institutions. Now, all these banks. are providing quite literally the exact same product. A current account is a current account. A savings account is a savings account, whether you go from bank A to bank B or whether you go to bank B or bank C. But determining factor will be into how a uses AI to now improve how customers interact with them. So customer experience for me ⁓ is... to be a game changer for banks in the next five years. Julius Neil Buenconsejo: Do you think it's risky for African organizations to copy Western strategies without adapting them to the local context in Kenya or anywhere in Africa? Peter: And to that, we talk about bias. So when talk bias, we don't necessarily understand how some of these models or initiatives are built. And even so, if they were built, we don't necessarily see how, ⁓ don't understand what in the framework of building ⁓ LLMs the respective technologies, chat boards, chat GPTs and the like. So of course, some of the data that is used to into these systems is biased towards the person who's actually building. And when they do that, they do not necessarily understand our local use cases or other local instances and nuances. So there is cause for Africans and Africa as a whole. to now start coming up with AI that is around our local nuances, because we understand Africa better than anybody else. And to be honest, ⁓ their local context better than them. So if we do that, we remove instances of bias and instances of where ethics might have been infringed on when developing some of these tools. Julius Neil Buenconsejo: Okay, so currently you already discussed this a while ago. for example, how can leaders in Africa build credibility around AI when the resources and then the technical expertise are very limited, right? Peter: Well, that's a very interesting question. in my opinion, I think it's about educating, or rather education. So how do we best educate? Let me use this term loosely, everybody ⁓ AI, And in educating everybody, it's more a matter of helping people understand what AI can do, what it can't do, most importantly. Julius Neil Buenconsejo: Hahaha Peter: biases data can have or what can be introduced as a bias within the system. Sorry. We have to also understand or be educated on the risks and ethics of ⁓ implementing Once we an understanding of what AI is and what it can do, I think from that perspective, we are able to have credible systems or credible frameworks that now solution for AI. and AI in terms of for our local nuances or uses. Julius Neil Buenconsejo: Okay, so let's go to some of the technical side. you're exploring generative AI and multi-agent systems. So which of these technologies do you believe will impact African enterprises first? Peter: you talk about generative I think impact Africa and is probably ⁓ impacting Africa when you look at how LLMs are actually being used to the usage of LLMs When I talk about usage of LLMs, I'm talking about Tadjiputi, Gemini and the like. So the thing with these systems that they have now created opportunities that were nonexistent before, right? We are having instances where we have freelancers who are now coming up developing crazy systems, websites and the like. have agencies, as I said earlier, who are now using AI, AI rather, to create advertisements and the like. And even just informational campaigns. So we are seeing a rise or an up such ⁓ of a new ⁓ cases, you and generative to create ⁓ better opportunities the young populace in Kenya and Africa as well. Julius Neil Buenconsejo: Yeah, so since multi-agents are still emerging, how do you think they might transform complex institutions like banks or airports or agencies? Peter: . So ⁓ effect of using multi-agent systems is ⁓ might lead to job losses, right? it will be very, hard ⁓ to explain why you need a warm body to do something that can be by an agent or a multi-agent system in a matter of minutes. The beauty of ⁓ multi-agent systems is that it will now improve ⁓ experience of and also augment ⁓ work as is. So as I said, when you a multi-agency system within a banking perspective, you now empower the relationship manager to actually now focus the ⁓ nitty of his particular clients in his particular portfolio, right? And he's not bogged down in the day-to-day mundane recurrent things that ⁓ go maintaining. clients portfolio. So it now gives them a third arm, if you will, whereby they're able to now let multi-agency systems do the heavy lifting, right, and them to now concentrate on the most strategic views of business. Julius Neil Buenconsejo: So, since Africa now is more on the consumer, AI built, with consumer agents, so could Africa become a testing ground for any innovative AI use cases rather than just a consumer of AI built elsewhere? Peter: Definitely, we are actually seeing instances of where AI is, we are leading in cases of research and development, right? Mostly natural language processing to understand or rather ⁓ initiatives around local sets and local language, right? So we are not necessarily ⁓ being consumer as it is right now, but we are seeing a shift whereby Julius Neil Buenconsejo: Mm-hmm. Peter: research institutions, universities and the like are now using different technologies to research in natural language processing, computer vision, in medicine and the like. yeah, at point the tide will shift ⁓ whereby now start seeing more innovative use cases coming out of Africa as compared to just ingesting what has been brought in. Julius Neil Buenconsejo: Okay, so if adoption in Africa succeeds, what would it look like in 10 years? Peter: ⁓ that's a very interesting question. I think quite a number of industries will actually transform. The instances of some of things that we actually are struggling with right now, interacting with customers or education and the like, will actually now cease, not necessarily cease, but will trickle down to a bare minimum. What that means is we'll... start seeing new types of jobs coming up. We'll have roboticists, we'll have natural language researchers, we'll have multi-agent system developers and the like. These are not job descriptions that are current right now in the grand scheme of things. But again, as we had earlier stated, when we start educating the populace or the everybody on what AI is and can and cannot do. We're starting seeing different job descriptions, different use cases coming out of Africa, and that now levels the playing field for everybody. it now becomes ⁓ way of bringing that gap ⁓ the West and us. Julius Neil Buenconsejo: Yeah, so before we go to the rapid 10x in our last section, so this is one question I would like to ask you. What advice would you give young African data scientists who want to influence strategy, not just build models? Peter: be as close as possible to the business. That is the one thing I wish I was told before. I started, I heavy into models and ⁓ data and all what you. I spent copious amounts of studying on but the one thing that I did not really understand was the business and how the business rate. And you quickly come to realize that if you do not understand the business. when I say understand the business, there's no necessarily mean going to business school. It ⁓ what is the organization offers and does. Once you get close to the problem statement, understand how to track progress, measure ROI, understand financials even, you'll be able to make ⁓ quite ⁓ some impact your data science models downstream as compared to just Julius Neil Buenconsejo: Mmm, okay. Peter: on your laptop and coding italy. Julius Neil Buenconsejo: Okay, that's a very nice advice to the youngsters. Okay, so we're going to our last round now. So this is the Firefast question. So these are 10 questions I'm going to ask you personally about AI. So what's one tool you actually use every day? AI tool. Peter: such as GPT and perplexity. Julius Neil Buenconsejo: Okay, so what's one task AI has already 10x in your life? what's one task AI has already 10x in your life? Peter: I code a lot faster right now. I code a lot faster right now. Thank Julius Neil Buenconsejo: Okay, so what's one human skill AI can never replace? Peter: Empathy, empathy, empathy, empathy. There is no way AI is going to replace. Julius Neil Buenconsejo: So if you can automate one annoying thing forever, what would it be? Peter: Elon Musk has already done it, driving. I would automate driving immediately. I wish to meet Musk and congratulate him on the good job well done. Julius Neil Buenconsejo: Hahaha! what's the biggest misconception about AI, right? Peter: it will kill creativity while in essence it's actually augmenting and improving creativity. Julius Neil Buenconsejo: What advice would you give someone who wants to become 10x of himself or herself starting today? Peter: Just start, just start, do it scared. You never know where you'll end up. Just start, do whatever it is you can. Yeah. Julius Neil Buenconsejo: What would you tell your younger self about the future? Peter: Be very courageous and be bold in your ambitions. do that. Be bold and courageous. Julius Neil Buenconsejo: Okay, so what's one book or resource that shaped your thinking about AI or innovation? Peter: That was Super Intelligence by Nick Bostrom. It helped me understand the connection between humans and artificial intelligence. Julius Neil Buenconsejo: Okay, so if you had to bet on one AI trend that will be massive in three years, what is it? Peter: agentic AI. AI is going to change industries. Nobody can tell me anything else. Julius Neil Buenconsejo: in one sentence, what does ⁓ mean to you personally? Peter: Wow, 10x is basically just being bold in your vision, your strategy, in whatever, in who you are as a person. As I say this, it's something that I'm still trying to understand, I'm trying to embrace. But it means that you also have the courage to just do hard things and do them well. You may not necessarily succeed in them, but embrace the failure. And in embracing the failure, at least you have done something as compared to not doing anything. Julius Neil Buenconsejo: Okay, that's a very good ending so that was a that was the rapid 10x round. It's so nice So if our viewers or listeners wanted to keep in touch with you, how do they connect with you? Peter: Hey, please reach out on LinkedIn, peteronyango, Instagram at psonyango, send me an email at psonyango.gmail.com. Reach me on any of those ⁓ I'll definitely reach out. Thank you very much. Julius Neil Buenconsejo: Okay, so, all right. Thank you for watching 10xAI. Remember, the future doesn't belong to AI. It belongs to the people who use it. So, see you next episode.