Data science and machine learning are complex yet important concepts. So, what role do data science and machine learning play in the development of FinTech systems, especially in countries like Indonesia? The lack of credit cards combined with the substantial use of mobile phones in Indonesia represent a sweet spot for FinTech companies to deliver advanced user-friendly financial solutions in this country.
In this episode of Data Point of View, Laurie Hood welcomes Joel Samuel, the VP, Head of Machine Learning Engineer at FinAccel. They get into the importance of machine learning and data science in accomplishing business goals and delivering a better user experience, the hassle of finding data science specialists, FinTech and e-commerce development in Southeast Asia, and the essence of starting small.
Data science and machine learning are complex yet important concepts. So, what role do data science and machine learning play in the development of FinTech systems, especially in countries like Indonesia? The lack of credit cards combined with the substantial use of mobile phones in Indonesia represent a sweet spot for FinTech companies to deliver advanced user-friendly financial solutions in this country.
In this episode of Data Point of View, Laurie Hood welcomes Joel Samuel, the VP, Head of Machine Learning Engineer at FinAccel. They get into the importance of machine learning and data science in accomplishing business goals and delivering a better user experience, the hassle of finding data science specialists, FinTech and e-commerce development in Southeast Asia, and the essence of starting small.
Mobilewalla - Data Point of View - Joel Samuel - Transcript
[00:00:00] Laurie Hood: And, again, it's counting down. We're ready. Okay. Thank you for listening today. I'm Laurie Hood, CMO at Mobilewalla, and this is Data Point of View. Data Point of View is a podcast for anyone interested in using machine learning and consumer data to achieve their business objectives. Joining me for this episode is Joel Samuel.
[00:00:23] Joel is VP, Head of Machine Learning Engineer at FinAccel. FinAccel is a financial technology company creating disruptive and meaningful products in retail credit in Southeast Asia. FinAccel is building the future of financial services, and I love this, Joel, fast, cheap, and widely accessible, wrapped up in a beautiful user experience.
[00:00:46] So, Joel, welcome to the program and thank you so much for joining me today.
[00:00:50] Joel Samuel: Hi, Laurie. Thank you also for having me. Thank you for inviting me for this podcast.
[00:00:57] Laurie Hood: Well, it's my pleasure. And, you know, [00:01:00] I want to start talking, this is how we'll sort of break down the program today. Let's start talking a bit about your background and what led you to your current role and then focus on how you and your team are driving change at FinAccel, and then we'll close with your thoughts for 2022.
[00:01:18] So, if that's okay with you, we're going to jump right in. Yeah, okay. Fantastic. So, let's start, you know, your background is very deep in machine learning. Can you share a bit about your career so far and how you landed in your role at FinAccel?
[00:01:37] Joel Samuel: Yeah, so it started actually since my days in college, in my undergraduate degree, so I drove my interest about data there, and then, after I, from my bitter, actually the data science scene or data science, sort of, it's not high picture in Indonesia, so I joined the consultant [00:02:00] company, but still working about data. So, I worked at what, the first product that I actually work is about data transformation tools. I took my master, specialized in process mining. Process mining basically is blacks, subs, subsidiary or specialty, more specialization in from that. And then, yeah, starting after I graduated with my master's, I decided to pursue more on the data career actually, so after I graduated with my master, I took a job as a data scientist in public sector, so I worked for the Jakarta government. So, in there movement called, it's like the initiative from the Jakarta government to introduce about the technology in, especially in that big data, to solve a problem in the public area.
[00:02:52] So, after that, I joined Jakarta Data Scientist around one and a half years, and then the [00:03:00] opportunity came from the FinAccel Kredivo. At that time, I don't know, I didn't know actually about the FinAccel because it, it was still a small company, at, I think in 2017 or 2018. One thing that I know that I want to continue my career in the stock option, so at the time the startups company, actually, already become more and more in terms of number in Indonesia, so I want to join that with victory. So, when, when the came to me, give this off, gave this opportunity, I drip it, and then starting join it, from the maybe, like, less than 50 people in the company. Currently, come FinAccel already, I think more than 300.
[00:03:43] Laurie Hood: Wow. Yeah, so.
[00:03:48] Since you've kind of teed it up, let's talk about, you know, FinTech, especially in emerging markets like Southeast Asia and Latin America is, is incredibly hot, [00:04:00] so can you sort of tell us a little bit about FinAccel, the company and the products, the different products that you have in the market?
[00:04:08] Joel Samuel: Yeah. Yeah. So, FinAccel actually, first nine no better solution in Indonesia is, I can say. Why we go to that, or why we choose to provide them a better solution in Indonesia, there are two main reasons. The first one is the low penetration of creditcard in Indonesia. There's only 17 millions of credit card compared to our population around 250 million current, nowadays,
[00:04:36] so it makes, like, only 0.07 credit cards per capita, so it's really low. And, then the second one is the hype inefficient of mobile phone. So, currently Indonesia has, like, more than 119 million mobile phones, so almost 0.8 mobile phones per capita. So, it makes [00:05:00] a sweet spot. It, it makes the sweet spot. So, you have a mobile phone, but you don't have the credit card,
[00:05:06] right? So, and also at that time, the e-commerce wave really, really pick in Indonesia, so we have, like, currently we have three or four unique course come start a company based on e-commerce. So, and then one of the, one of the problem that e-commerce has, not in Indonesia, I think mostly, but also in the, all the, all around the world is about the cart abandonment.
[00:05:27] So, one of these, one of the issue is more in the payment option or payment channel. So, most of the people that abandoned the cart is because they have some cash out to do the payment, so that's why we have a sweet-spot to provide our solution, buy-an-operator solution. That's our go-to-market at first time.
[00:05:48] But currently FinAccel, it's more about, more then buy-an-operator solution because we also provide many financial solution, example like simple biller. You can, you can top [00:06:00] up your electricity or your mobile phone credit and then you can take personal loan. And, then we are also, soon we will launch our smart and maybe an enterprise loan. And, just recently we launched our motorcycle loan and then yeah, the, the, the goal is FinAccel want to provide the financial services, not only the better solution in Indonesia or in the Southeast Asia, in the, like you said, in the low cost and then in, in the really great user experience.
[00:06:31] So yeah, we just recently also opened our second country after Indonesia. We opened in Vietnam, and then I think in the next year or in the coming years, we all, we also will open in the another two or three countries. Yeah, that's FinAccel right now.
[00:06:47] Laurie Hood: Well, that's fantastic, I mean, incredible growth at FinAccel and with Kredivo, and then certainly, kind of, you've been on the fast track, you know, there with your career, so can you talk about some [00:07:00] of the key responsibilities of your role and kind of what you're dealing with on a day-to-day basis?
[00:07:09] Joel Samuel: Yeah. We have quite, I don't know, maybe, maybe in some companies this is small, in some companies it's already a big, we have a quite number of people in Data Deficient, led by one of data officer. So, there are three groups in, in the Data Deficient. There are Data Scientists, Data Science group and then ML Engineer group,
[00:07:33] and then Data Engineer group. So, we defined our organization of that, into that three group. So, I'm, I'm, I'm lead, I lead the ML Engineer group and then there are also yet another leader for the Data Scientists and the Data Engineer. Basically the deficient of four for these three, three groups is a data scientist, is applied data science method in, into many business problems that we take. [00:08:00] The end product of the, the data science group, you'll be an ML model, machine learning model, or for example rules, some set of hard rules, or maybe just an, some analysis about some problems. And, then the ML engineer will tackle to develop that into a surface or the solution at data solution, so the machine learning model that comes from data scientist, we will wrap it up into a site that can be accessed either real time or we ran it on a scheduled basis, so the, the result can be used by the product, product engine, product and, or any other, any other group or any other deficient, like customer service or operation deficient. And, then at data engineer, I think this is the most stable
[00:08:52] group or, I mean, like, we already know about data engineer, I think from, like, 10 years ago, so, and also [00:09:00] the, the main job roles actually also already, like,
[00:09:04] established before the machine learning or the data scientists, so basically developing our data warehouse and also building our analytics dashboard for our business stakeholder. I believe there are total, in total there are 30 people in a data deficient, around 30 people, and also we have already more than 14 machine learning models for the front on the production.
[00:09:29] We take from the risk model up to the checkpoint vectorial. So, yeah, all these surfaces stop the, stop from the credit scoring and then a checkbook engine, in the production it will be my thing responsible to keep it up and running.
[00:09:48] Laurie Hood: So, you're in a business clearly that's highly driven by data and artificial engineering or artificial intelligence. What are some of the challenges that your [00:10:00] organization faced as it's grown?
[00:10:05] Joel Samuel: So, yeah, there's a couple of challenges here. Of course, at first, the, the main challenge as a lending business, lending business company is risk side. So, how we can tackle that using the data science methodology? So, kind of, at the first time, when we built the data thing, we only focused on that part, so on the risk management part. So, we started with only three people actually, working on that together, then the focus only to build the what we call ABC score or maybe more people knows about the credit scoring score. And also, the problem that we have as an organization at the very first starting the company is we cannot attract good talent.
[00:10:52] So, because the, the, the name itself, the company name itself is not [00:11:00] well-known so not many people also want to join the company, so I'm lucky enough to have this chance to direct this opportunity and put my bet on.
[00:11:14] Laurie Hood: Now it's a super hot company to work for those, so you were the who had the vision.
[00:11:21] Joel Samuel: Yeah. So, yeah, just, so that's why I said that I bet, I have a good bet. So, at the very first there's a couple, there's not many, many people that actually knows about Kredivo, FinAccel, right? So, and also knows about the FinTech industry. So yeah, that's the, the, the main challenge when we have the, when we started. Next, after we, we've done the bigger, bigger, bigger, the other challenge is a good data science utilization or data science implementation become more, more than a risk.
[00:11:56] So, maybe part of business actually [00:12:00] already starting gain the confidence about data science and then start to think creative that all the problems in the world can be solved by the data science. Then the other problem comes so we need more than a risk model or are we need more than a data science that knows only about the risk modeling.
[00:12:23] So, we need specialization. So, that becomes another challenge because find, finding data scientists that specialize in, for example, in the natural language processing, it's kind of difficult because we don't have any basic content. All the people that already joined it very well because we only focused to find the people that already know about the risk modeling. Another, but data science is really huge and also widespread,
[00:12:51] so finding the good people or the good person under specialists, specialists, the specialized field is actually become also the challenge. [00:13:00] Another piece that.
[00:13:02] Laurie Hood: What are you doing to deal with your staffing challenges? Because you hear that, that there's a tremendous need for data scientists that outpaces the available talent, and then if you're looking for someone with some specialized skills, are you working with any local universities? Have you kind of maybe broadened where you're hiring from?
[00:13:25] Like, what are some of the ways you're looking at addressing that?
[00:13:29] Joel Samuel: Yeah. So, we, we already actually tried that, that some of the approach that you mentioned. So, we, we tried to working in with the university, but, but apparently it not match, it doesn't match with our culture. So, because we really very, very fast-paced company, so sometimes we don't have a time to grow in this talent or we, we, because we don't have a big team.
[00:13:53] So, actually this is to, to, to get and then we realized that as [00:14:00] a company or as a team, we also a little bit too private, so there is no many people that actually come up to the surface, the surface to show that what we are doing. So, in the community, they don't know what actually the data scientist in Kredivo.
[00:14:16] It looks what they are working or what work they are, they are looking. So, we starting to open at them, of course, this is one of the joining, joining this podcast also one of the, one of the, the strategy that we want to implement right now so comes up to the surface a little bit more and then show what we are doing as a company and also as a data scientists, as a data thing.
[00:14:44] Hopefully it can attract more good talent to join to the Kredivo. We also have this.
[00:14:51] Laurie Hood: Well, we're, we, I'm, I'm flattered that you are using our podcast to get Kredivo more, more play in the market. That's [00:15:00] fantastic, and I think it's important to, to get the brand out. And, I mean, all of the exciting things that you guys are doing. So you, go ahead.
[00:15:10] Joel Samuel: I just wanted to add one more challenge. So, currently the, the, the incoming supply, actually, already become more and more, a bit, become better and better, but, as you know, the issue right now is the supply in, in data science, ML engineer of itself. So, since this is a new thing in Indonesia, and all the people that want to pursue this sexy new field, but the good talent,
[00:15:39] we, we lack of, we lack of the good talent, it's the supply in Indonesia. And also, we open up the, the opportunity to this region in Asia. That's also one of the strategy that, besides come to surface a little bit more, we open up to more countries to open more, more [00:16:00] opportunity to get a good talent, currently.
[00:16:01] So, yeah, that's a really big problem right now, the supply side.
[00:16:08] Laurie Hood: Well, and I would guess it grows as you increase, you had talked about bringing more products to market, and you're, you're going to continue to grow, you know, the amount of models you're producing and supporting with that. So, want to go back to something interesting that you had mentioned, and that was the utilization of the data science team across the organization and, kind of, as people realized all the things you could do, they thought you could solve any problem. But with that usually comes a lot of internal trust in what your teams are doing, so could you talk a little bit, I mean, we do hear people say they have a hard time educating people, other people within their company about data science, about what your team can do [00:17:00] and building that trust.
[00:17:01] So, could you talk a little bit about how you've addressed that?
[00:17:07] Joel Samuel: Yes. So, luckily this is not the case in FinAccel. So, since beginning, we already have to buy in from the top-level management. Moreover, actually, the top level of management realized that this is the one thing that can break to market, this is the one thing that can be a deal breaker. If we want to disrupt the best player in the market, like the bank or the multi-finance company that already there, so one thing that we can do is we introduce the data science methodology and we solve it with better way because they believe that the data science has a big, big opportunity and then a big big capability to do it. But yes, of course, even though that we have already defined aim and also the improvement or the initiative comes from the top management, we have to prove that, [00:18:00] that initiative or the buy in at the very first feeling that we can deliver in fact.
[00:18:06] So, this, this is the actually the biggest problem will happen in the data science part in the company. They cannot pro, prove that this solution approach can, that data science approach can solve the, the real, the real problem, business problem. So, we believe that in, in, in Kredivo data science project is a never lived, never-ending loop.
[00:18:27] As there, we started with the real business problem or the real, the business question, and then we iteratively do the, we iteratively do the process so that we can have a really nice solution. But, one thing that I want to highlight, there are two things that actually can maintain the buy in and also expand our use case in the Kredivo. We started, of course, we started with a really poor problem. That's the first thing. But, we do a rigorous process to do that. I mean, we know the total process but we don't, we don't [00:19:00] do like, like, really a mess. We don't mess up with the process of when we are building the model. There are separate checks.
[00:19:06] We have a regular meeting with our COO and CEO in the first two years to present the result, and then we, we have a really good monitoring workflow and framework, so that we can, we can quickly spot that if there's something wrong with our model that we pro, push to production.
[00:19:26] We also really believe about the fail, "fail fast and learn fast." So, we always push it little by little to production, to see the effect, the impact of the model so that we don't, we don't start with big thing. So, start with the simple thing and a small thing. We, we believe that there's many low-hanging fruit that can, can we crack, and also can we solve so that the thing that actually built the confidence in the top minutes in the red. So, we start with a small problem, we [00:20:00] push it quickly to the production and we see the results fast, so the management can immediately see the impact of the effect of their data science approach in the company.
[00:20:11] That's, that's the thing that keep, keep maintaining the blind in, and also the, the trust from the top management. As long as it can, so, so, and also, one thing that, that, that data science, or data team at Kredivo actually do is we really pragmatic, pragmatic people.
[00:20:27] So, as long as it can solve the problem, it's good enough. So, we, we, we, we are not a real get, get, get out or get, get want to, want to, want to implement the really cutting-edge technology, you know? So, so we, we, we go to the bottom of the problem, the business problem, and then we see that those that we have received a set of algorithms that can be, can solve the problem but
[00:20:53] we pick the most efficient and effective way. So, we don't, we don't really get shine [00:21:00] by the really new technologies. So, I think that's what we do to maintain the buy in. But yeah, we have the opportunity, we have to affect to have to already have the buy in at the very first time we implement this data science solution.
[00:21:16] Laurie Hood: Well, that's, I mean, organizational support so important and you've actually made my job easy because my last question for you is going to be to summarize some of your key findings and, and you just did that talking about, you know, having a rigorous process. What we see too is it's not always about the complexity of the solution, it's almost finding even sometimes the simplest algorithm that's going to solve the most complex problem instead of increasing that level of complexity, and that just puts more pressure on your data science team. But I love the "Start simple." I think there are so many organizations that we work with at [00:22:00] Mobilewalla who really are just starting out, and I think "Start simple and prove yourself" is really the way to go. And, when you're so fortunate, I mean, your business, it's so based on data running models. I've got some experience in the US credit business so I understand it from a more credit-based economy, but it's just, it's just so interesting and I wish you the best of luck you're in such a great hot business.
[00:22:30] And, and I want to thank you so much for joining us today and sharing your insights. We really appreciate it. Thank you also, Laurie. You're welcome. Well, and to our listeners, thank you for your time today and please join us for another episode of Data Point of View, brought to you by Mobilewalla.