Enjoying the podcast? Don’t miss out on future episodes! Please hit that subscribe button on Apple, Spotify, or your favorite podcast platform to stay updated with our latest content. Thank you for your support!
Imagine serving unsecured consumer loans to a population that has no credit file and no banking activity whatsoever. This population primarily uses cash and operates outside of the traditional financial system. Is it even possible to serve this population with short-term loans profitably?
My next guest on the Fintech One-on-One podcast is Roberto Salcedo, the CEO and Co-Founder of Baubap. Roberto and his team have done exactly that, building the technology and the underwriting models themselves to serve Mexico with small dollar loans. How he has been to do that makes for a fascinating story.
In this podcast you will learn:
- How he was able to leverage his knowledge as a banker to start Baubap.
- Why Mexico has lagged the rest of Latin America in adoption of financial services.
- Why the cash economy in Mexico makes it harder for banks.
- Baubap’s core lending product and how quickly it is delivered.
- How they created their hypothesis for their unique approach to underwriting.
- The types of data they feed into their underwriting algorithms.
- How they started using AI in their initial models.
- Why they created all the fraud, identity verification and KYC features themselves.
- How they have been able to improve their approval rates over time.
- Some of the signals they use to indicate creditworthiness.
- The typical borrowers they are working with.
- Their lucrative referral program.
- How they are finding their borrowers.
- Their track record of profitability.
- How they were able to close a $120m debt facility from a US investor.
- Roberto’s expansion plans for Baubap.
Read a transcription of our conversation below.
FINTECH ONE-ON-ONE PODCAST NO. 506 – ROBERTO SALCEDO
Peter Renton: Welcome to the Fintech One-on-One Podcast. This is Peter Renton, Co-Founder of Fintech Nexus and now the CEO of the fintech consulting company Renton & Co. I’ve been doing this show since 2013, which makes this the longest-running one-on-one interview show in all of fintech. Thank you so much for joining me on this journey. Now, let’s get on with the show.
Today, on the show, we are going south of the border down to Mexico, where I am delighted to chat with Roberto Salcedo. He is the CEO and Co-Founder of Baubap. Now Baubap is a microlender and a really interesting company. They’ve been doing small loans, sub $250 loans now, for several years, but what really attracted me to them was their underwriting. They have been doing AI since before it was trendy. They have developed their underwriting models using data from an Android smartphone rather than financial data. Their default rates are extraordinary for an unsecured consumer loan. Certainly some of the subprime lenders in the US would love to have default rates this low. It really is fascinating to hear more about how they’re able to do this, how they have developed underwriting models with approval rates improving dramatically over time and yet keeping their default rates the same. You will learn all about it in this fascinating discussion. Hope you enjoy the show. Welcome to the podcast, Roberto.
Roberto Salcedo: Thank you very much for having me here, Peter. I’ve been a follower of your content for a long time. So I’m very much looking forward to this discussion we’re going to have.
PR: Okay. Well, that’s good to hear. I’m looking forward to it as well. So let’s get started by giving the listeners some background about yourself. Can you just hit on some of the highlights of your career to date?
RS: Of course. Well, I was a banker, for better or worse, for 12 years. I was actually one of HSBC’s top 1% credit underwriters for large corporations. I used to service traditional lenders, like auto loan companies, mortgage companies, personal loans companies, credit card companies, all sorts of financial service providers. And so I had a lot of fun. I learned a lot. You know, I used to go really deep into their business models and into their underwriting models in order for me to underwrite loans, and create facilities and capital structures for them. And then, I moved on from HSBC and joined another Mexican bank called INVEX Bank. And I had a fantastic time there, you know, leading their corporate banking arm. I led INVEX Bank’s largest finance deal in Mexico to date. And after all this experience as a banker, I got invited to join as the chief financial officer of a media company here in Mexico. And this was eight or nine years ago. Back then, it was a very exciting opportunity for me. You know, I’d never been an operator before; I’d always been a banker. So, becoming an operator myself was extremely entertaining, you know, very challenging, lots to learn. And that’s where I met my co-founder. He was the CTO of the company. And we started working together. We implemented the ERP from Oracle at that company. And that was a terrible experience. But that’s how we became a team. That’s how we met each other. That’s how we became friends. And we started talking about building something together.
I have lived through the problem Baubap is trying to solve with my own family, with my own parents. So I was very close to the problem and I, and I thought I knew how to solve it. And we’ve been building together since 2018 now, and it’s been quite a journey and happy to share more about it in the podcast.
PR: Sure, you had the idea because you said this is a problem that your family and lots of families have. It’s not just in Mexico but around the world. You didn’t mention any consumer lending experience in your background. Was that a major learning curve you had to go through?
RS: No, the interesting thing is that as an underwriter at HSBC, I had to go into the models and understand how other people were originating consumer loans. So, I worked on everything from mortgages to auto loans to personal loans to payroll loans. I loved the business because it provided me with a lot of understanding of why there was such a huge underserved population in Mexico and most emerging markets. And so I was able to connect the dots. So, okay, this is the piece that is missing here. This is why these traditional lenders are not able to serve this underserved population. And that’s where the expertise came from.
PR: Okay. That makes sense. Then, let’s dive into that. Why is there such a large underserved segment of the population in Mexico? Latin America, in general, has come a long way over the last five to 10 years, but it seems like Mexico hasn’t as much. Why is that?
RS: I think it all has to do with the economic challenges a large part of the population has. So, 95 % of Mexican workers make less than $750 per month. So that means that most people are low-income people in Mexico, right? And then we have 60 % of workers operating in the informal economy, so that represents another layer of challenges. So, these people are in desperate need of various small amounts of money. So, it’s very difficult for traditional lenders to reach out to these customers profitably in a cost-efficient manner because they have thousands of employees, thousands of branches, and such large infrastructures. It becomes very challenging to adapt data into the digital world rapidly and start servicing these consumers, right? And the other main challenge is that people have very little banking activity. So, there is not much data to underwrite loans for people who work in the informal sector because they don’t have banking activity. They don’t have a credit history so there is not much to go on there for traditional banks and lenders.
PR: Right, right. Yeah, that makes sense. One of the things that strikes me is that I was in Mexico City on a Friday afternoon and saw lines at the ATMs there. The average Mexican seems to be very much cash-oriented. And does that make it harder for companies like you to really get scale?
RS: I think it makes it easier for us, harder for banks and traditional players. The thing is that even people with bank accounts are very used to going to the ATM and withdrawing the entire bank account balance. What that means is that a bank has no data points to underwrite to these customers because there is not much data there. There’s only a deposit and a withdrawal from an ATM, which is also a very unprofitable transaction for any bank. That’s where banks are starting with this population.
PR: Gotcha.
RS: And for us, we built everything based on using only alternative data. No banking activity, no financial history, no credit reports. So we built this on the entire idea that there were other data patterns that we could leverage in order for us to underwrite this huge amount of people cost-efficiently and with accuracy.
PR: Right. Right. Okay. So let’s dive into Baubap and your core product. Obviously, you’re a consumer lender. What is your core product?
RS: Of course. We service all of our customers through an Android app. And it is through the app that we extract all device data from a user’s mobile phone. So, for example, SMS, contacts, calls, calendar events, applications, geolocations, the device’s detailed footprint, and the phone line’s data footprint. We then combine all these data with the user-provided info in order for us to create thousands of features that we then feed into 15 different in-house machine learning models and LLMs. All this core technology has contributed to Baubap getting the best default rates in the consumer market in Mexico. Our core product is a 30-day term loan that starts at $25 and can go up to $250. It’s a loan that is always available and delivered to the end customer within 30 seconds of the underwriting decision. So that’s the core product.
PR: There’s quite a lot to unpack there. I’m really interested in the underwriting because, as you said, people might be taking out their money from the bank, and that’s not in any database you can use for underwriting. I’m curious about the underwriting model you created that used this alternative data. Did you create that from scratch? How did you know what data sources to use and how did you know how to improve it? Tell us a little bit about the journey you’ve taken with your underwriting.
RS: Of course. A lot of creativity was required for us to make this technology work, especially at the beginning, because it was pure imagination. We came up with the hypothesis that the data that we were building ourselves in our mobile devices by interacting with them throughout any given day had a correlation with our economic activity, our payment capacity, and our willingness to repay a loan. So that was the entire premise of everything. We took a leap of faith and decided to start extracting this data first with our own devices. So we extracted the data from our own devices and started imagining if I were a banker or the CFO of this company, I should have emails about this activity. I should have text messages about this activity. I should have a different way to communicate or to write down or to take notes or my calendar should look like this, right? So that was in essence, how we started finding insights, powerful insights, around this data. And we started with our own data. And then we launched the app. We launched the first beta of the app on the Play Store. And it didn’t lend any money because we were not capable of lending out any money yet. But it was a loan request. They had to fill out a loan request. And we could extract, with their permission, of course, all this device data. So we started doing this and we got like 20,000 downloads organically with just people willing to share their data to get a loan. And with those initial 20,000 users, we built the first features, the first models, and we started then testing out different models and looking for different patterns. And we were very early in AI. We started using foundational models as early as 2019. We started using BERT, Google’s first LLM, to classify all text transactions and to get insights from those text transactions. So I guess that’s how we got started. And it started to evolve extremely well, right? So now it’s a large set of different models for different purposes because we also did everything ourselves. Not only do we do underwriting, but we also do fraud, identity verification, and KYC ourselves. So, the entire platform was built on this idea of only using models, only using predictive AI, and generative AI to deliver the service as cost-efficiently as possible.
PR: That’s really interesting. So you’ve decided that you’re going to be the developer of everything. That’s a really interesting way to go. And I’m curious about how things have developed. You’re talking about huge amounts of data; I imagine you’ve probably got thousands of data points or more for every single user that comes in. So you’re talking about pretty large data sets. How have you improved that, and how has that evolved? I’m particularly interested in improvements in default rates and how you can use your knowledge and understanding of the past to keep getting better.
RS: That’s where getting early into this technology really paid off because we were early adopters of everything, right? So, every new release of BERT, every new release now of LaMDA, one of the models we use locally, every new release, every time a model got better, we got better. We were able to start optimizing, keep optimizing hyperparameters. And, of course, we updated the sample with which we were training all the models. Now, we’ve given out 8 million loans in these six years. So now we are training with a massive data set that has become a competitive advantage. So it is that massive data set that allows us to be very accurate and precise and to use the different ranges of models for different purposes. So I guess that’s what happened.
PR: And it’s great that your loan term is so short. You can tweak things very quickly, I imagine, and change your criteria. Maybe on that, what is the approval rate? Obviously you’re going to have some people who are a really high-risk customer, and you’re not pulling in any financial data, right? When people download the app and apply, what are the chances they will get a loan?
RS: Of course. This is where most of the improvement of the models has benefited us. Because we started with very low approval rates. I remember back in the day, we were approving about 5 % of the new applicants. So what ended up happening, and this is where it got very interesting for us, is that we massively improved the approval rates. At the same time, we kept default at bay, especially early payment default, as we think healthy pockets for our business. So now we get a 40 % approval rate for new applicants. So 40 % of new applicants will get an offer from us. Then, the first payment default right now is around 15%, which is something unseen in this market with that high of an approval rate.
PR: Right. And these are unsecured loans, right? You’re not securing against anything.
RS: That is correct. These are unsecured loans.
PR: Right, right. Okay. And if you could, I know you don’t want to give away your secret sauce, but I’d love to get a sense of the underwriting when you’re looking at the data on someone’s phone. What are some signals that are good, and what are some signals that are bad for these people?
RS: Of course. So it looks different. This is the superpower of the supercomputing power of the models because they look different from customer to customer. Some customers will get rewarded by the models for some reason, and others can get punished by the models for other reasons. So, the models go through a wide range of data patterns. So, for example, for our probability of default, we use SMS data, text data. One of the very interesting insights that we found early on was that customers who text at night behave better in payment behavior than customers who text during office hours. So it was small patterns like this that we could not only find but actually pick up the phone and start talking to customers and trying to learn why this was the case. For example, we validated that people, especially informal workers, are very engaged with their economic activity during office hours, so they are unable to text during office hours. So they text at night. This was very interesting for us, and we gained thousands of insights just like this one. And this is why it became really powerful.
PR: Who is the typical Baubap customer? Are they people in the informal economy? What type of jobs do they have and what are they using the money for?
RS? For that, I always like to share a customer story. One of our customers is Mrs. Patricia Pena. So Patricia makes quesadillas as a street food vendor in Mexico. Before discovering Baubap, she was rejected by five different banks, essentially having no banking history, and no banking activity. She was then unfortunately scammed twice while getting an informal loan, losing the little savings she had. So this is another huge risk factor for this population in emerging markets, especially in Mexico and other Latin American countries. And it was because of all this that she discovered Baubap. And with a loan from Baubap she began to chip away at these personal debts and reinvest these small amounts of money into her economic activity. So buying raw materials, buying inventory for her small, very, very small business. And around 50 % of our customers are informal workers just like Patricia, you know? And what has been amazing in Patricia’s journey is that in the last couple of years, Patricia has made $20,000 through Baubap’s referral program. So, referring customers to Baubap, which is astonishing.
PR: Wow. That is, and how much are you giving away? How much do you give away for a referral?
RS: It can go from $10 up to $25. Now, we treat the referral program as a product itself because it has so much engagement from our customers that we decided to build this as a product itself. So now we have different levels, different rewards. We have given out two and a half million dollars to our customers who have referred another customer for us. So that money couldn’t be in better hands.
PR: Right, right. That’s great. Beyond referrals, because obviously that’s a major source of new customers it sounds like, you said you’ve had roughly 2 million customers who have taken out loans. How are you reaching these people for the most part?
RS: Of course. Half of our customers come to Baubap through the referral program or organic efforts. And the other half use traditional paid media, you know, Facebook, Google, TikTok, a little bit of everything. So it’s a nice, interesting mix that provides a blended CAC that is profitable for us. So we’d use affiliate networks, DSPs, different partners for that other 50%, but we need all the traffic to go into the app. And I think that’s another key element of what we do because app traffic is a lot cheaper than web traffic. Everyone that is trying to compete for these kinds of leads in search are paying huge amounts of money per lead. And for us, it’s a lot more cost-efficient.
PR: That’s really interesting. What’s the application process like? What data are you asking for at application?
RS: We have a very simple loan request in which the customer provides personal info, work activity info, this sort of traditional loan request. And the very last step of the loan request is to provide access to the data we will extract. So, customers have to opt in. We’ve been GDPR compliant almost from day one because we understood this was our source of power. We take the data privacy extremely seriously. Customers need to consent for us to extract these device data. They need to understand how we will use this data and that we will not resell it to anyone. We are only going to use it ourselves. And the purpose is for us to make an underwriting decision. Once we have all the data, it takes us about 30 seconds to make the credit decision and to fulfill the rest of the disbursement process of the loan. So it’s a very straightforward process for our customers.
PR: And how are you disbursing the loan? Does that go into a wallet?
RS: We basically can do anything our customers want. We prefer to wire transfer the funds. Customers who don’t already have a bank account or a wallet account can open an account through our app with an API we have with a third party, if they engage, if they don’t have an account, but we can do anything our customers prefer to.
PR: Yeah. Right. Okay. Interesting. I saw in the news earlier this year that you guys landed a $120 million debt facility. I thought that was quite an achievement. Can you tell us a little bit about that process and how you were able to land it?
RS: I think one of the key factors about being able to scale this business has been profitability. We’ve been profitable for the better part of three years now. And I think this is almost unheard of in fintech lending. And the main reason was because the technology actually worked in delivering great payment results for us. So we could access debt funding very early on, as early as 2021. And we raised the first facility in Mexico from a debt fund here in Mexico. Then, we raised from a US debt fund. And then we started really scaling up. Last year, or at the beginning of this year, we raised this $120 million credit facility, essentially based on the amazing results and the amazing performance and the stability of the loan portfolio, even though we had massively scaled loan origination, right? Loan origination grew in the last three years, like 300 times; we originate 300 times more money than or more loans than we did three years ago. And we basically have the same default rate. So that’s why the funds got comfortable with us, and we’re able to size up the facility, and we’ve been deploying from that facility for all of 2024. We still have a good amount remaining there. So we can keep growing for at least the next couple of years.
PR: That’s one of the great things about your product too, is that you don’t need a big facility to do massive amounts of volume because the loans turn over so quickly. Looking at the state of fintech in Mexico, there’s obviously lots of fintech companies now, but you are in a competitive market. You may have the upper hand on most of your competitors, but I’m curious about how you’re looking at expansion in the future. Are you looking at staying in Mexico? Are you looking at going into other countries in Latin America? What are your plans?
RS: Right now, I can confidently say we will win Mexico. The problem is scaling significantly. We will start doing other very interesting things based on AI; we already have UX-based AI that we deliver to our customers. Our customers right now are engaging us through voice. The company in Mexico will keep maturing, and we’ll keep lending out money ourselves in Mexico. However, the next phase of the platform is transforming our entire platform – from underwriting, KYC, fraud, servicing, and everything – into a platform that we can lend as a service, that we can allow other brick-and-mortar players to use as a service to originate loans to the underserved in the emerging world. And for this, we look forward to expanding into other Latin American countries first. But the go-to-market strategy there is to start partnering with large brick-and-mortar lenders and large brick-and-mortar banks so they can plug into our entire platform and they can start servicing the people that they are not able to serve right now in a profitable way and in a way that they capitalize on the technology we’ve been building for six years. So that’s what the future looks like for us.
PR: Have you started doing this sort of lending as a service product, or is this something you’re planning to launch soon?
RS: We are building this now, and we’ve already started partnerships in Mexico to test out the platform and build everything to package and resell it. So, I think this will materialize within the next couple of years. And it’s one of the reasons that we are fundraising right now. We haven’t raised for three years. We were extremely capital-efficient in getting to this scale, but now we are raising essentially because we want to fund the next go-to-market phase in Mexico and other emerging markets.
PR: Right, right. Well, we’ll have to leave it there, Roberto. It’s really been great to chat with you. What an interesting story you have. Thank you for coming on the show, and best of luck to you.
RS: Of course, Peter, it was an amazing discussion. As I said, I was very much looking forward to talking to you. We’ve been great fans of your podcast and all your content for a while now. I’m so happy to share a little bit more about Baubap and myself with your audience and with you.
PR: Okay. Well, thank you. I’m blushing now. So it was great to chat with you.
Well, I hope you enjoyed the show. Thank you so much for listening. Please go ahead and give the show a review on the podcast platform of your choice and go tell your friends and colleagues about it. Anyway, on that note, I will sign off. I very much appreciate you listening and I’ll catch you next time.