AI-powered lending for Colombian businesses, with Viviana Siless

Nearly two out of every three Colombian jobs are in the informal economy! That's a huge number that needs to come down, and no doubt it will, but until then, someone really needs to bring the benefits of formal financing to the small business owners in the informal sector. Enter Quipu with whose AI-powered loan decisions "growing is easier, fairer and more alternative".

You can find Viviana on LinkedIn at https://www.linkedin.com/in/viviana-siless/

There's a Quipu page there, too, https://www.linkedin.com/company/quipulatam/ but there's even more information on their homepage over at https://quipu.com.co/

I am on LinkedIn, too, so feel free to find me and send me a connection request there: https://www.linkedin.com/in/brendanlegrange

My action-adventure novels are on Amazon, some versions even for free, and my work with ConfirmU and our gamified psychometric scores is at https://confirmu.com/ and on episode 24 of this very show https://www.howtolendmoneytostrangers.show/episodes/episode-24

I met Viviana while in Edinburgh at the 18th meeting of the Credit Scoring and Credit Control Conference - the next one is in two years time and it’s well-wroth planning for. In the meantime, follow the CRC’s work at https://www.crc.business-school.ed.ac.uk/home

If you have any feedback or questions, or if you would like to participate in the show, please feel free to reach out to me via the contact page on this site.

Keep well, Brendan

The full written transcript, with timestamps, is below:

Viviana Siless 0:00

More than 50% of the economic activity is informal in Latin America, and so, because of that, they don't have access to capital for growing their businesses.

And what we are trying to do is to help out to Yeah, to make it a little bit more fair for economic growth for everybody.

Obviously, you can do it with a pen and paper, but you know, I can say from experience, or at least the experience that we have, that doing it manually really doesn't work! And so really our scoring, what we are building, is to try to analyse the informal business.

Brendan Le Grange 0:42

When I was in my third year of university, four friends and I did a tour of the British Isles: I used my birthday money for the plane ticket, my waiter's wages for beer money, my uncle towed caravan into London for us to use as a base, we stayed at a friend's sister in the Cotswolds, my aunt lent us a car (with its boot full of goodies and its tank full of petrol), so we could explore the Lake District... you get the picture. I'd been to Namibia once as a kid and to England once as a teenager, but this was my first real experience of travelling.

And it was incredible.

But when I think back to it now, probably the evening I remember more than any was spent in Edinburgh. We had seen the castle and we had started a hop on hop off tour of the city, I think we jumped off when it got too cold. So it wasn't because of the city's charms, as undeniable as they are. And it wasn't because of anything extravagant we did later. I think we spent our whole night sitting in a railway station bar drinking Red Stripes, because they want a two for one special. But we'd been joined for that excursion by two other school friends who were living in England at the time, and something in the stars just aligned. And then we returned to our youth hostel for the funniest telling off I've ever had in my life. And maybe we would have had as much fun in any other city. But I'll always love the Scottish capitol for that night.

All of which is really just a very long way of explaining why I jumped at the chance to take the sleeper train north for the 18th meeting of the credit scoring and credit control conference. And while I was there, I picked some of the most interesting papers and asked the authors to tell me more. The next one is still two years away, but I suggest you mark it in your calendar now. And until I see you there, welcome to How to Lend Money to Strangers with Brendan le Grange.

Viviana Siless, welcome to the show. You're the Chief Technology Officer and co-founder of Quipu, but also a lecturer in AI at UTDT. Now I did actually scroll down to see what UTDT stood for and I was going to use that, but I see my Spanish probably doesn't stand up to the name pronounciation.

So maybe we can start there with your academic side, and then we can talk about the lending and the banking side of things.

Viviana Siless 3:08

Yes, I'm a professor at the Universidad Torcuato Di Tella in Argentina. And I mainly focus on teaching AI. And then I also lectured on blockchain. Yeah. And I have a really strong background in academia because I did a PhD - well, it was computer science and AI but it was targeted to neuro imaging, so I was analysing MRI brain scans. I did that in France at in reality, my PhD and then I went to Harvard Medical School, and I did a postdoc there.

When I was in Boston, I met my co founders at MIT, and this journey of microlending started from there. And

Brendan Le Grange 3:48

I think that's, I guess, the power of AI where data is everywhere now and we can use some of the same techniques. But obviously, I don't know anything about medical, so we'll talk about Quipu Bank.

So Quipu is a an inclusive FinTech that's developing fair and easy alternative finance. What did you and your co founders see that's said, actually, this is a problem we should be solving.

Viviana Siless 4:08

Yes, my co founders really come from my background from informal economies. And that's what they were doing at MIT.

And so the original idea really originated from them that they were already working in the ground for years. What they were seeing was that people in informal businesses such as like slums or favelas, the economy there is big, everybody has a business, there are many transactions happening there, but everything is offline.

So people are unbanked, the businesses are unregistered. And so because of that they don't have access to capital for growing their businesses or for improving you know, their way of living.

Most of the financial services generally use great Budo data, and 90% of our users are like listed there. So our users really access pay the lender or in Colombia is called the OTA OTA. And that is me 200% interest rate. And it's aggressive, like they come with a gun, and they threaten you to give the money back. And I know lots of really we wanted to help out to Yeah, to make it a little bit more fair for economic growth, or everybody.

Brendan Le Grange 5:27

How important is the informal economy to the economies in Latin America? But I guess more importantly, to the people of Latin America, how much economic activity does that account for?

Viviana Siless 5:37

Yeah, so surely it is more than 50% of economic activity is informal in Latin America. And what happened is that with COVID, it got worse, but it's not only that some things from COVID were good, actually, I don't want to say this, because obviously it sounds bad, but really COVID helped digitalization. Before COVID, or users were really not having any digital, well, not bank accounts, but not also digital wallets, they didn't have that either.

And really with COVID, yeah, now most of the people actually have a digital account. So it's easier to work with that, you know, like to give credit and to be paid back. Because before we were really doing it through through cash.

But yeah, the numbers are huge, is really the majority. And I think it's really unseen, this market, because it's not only for people living in slums on favelas, you might also have, I know a mother with the family that is also having like a side show on internet. And that's an informal business as well.

And if they get a loan, it's really for the person but not for the business and what we are trying to do is to give loans that are really productive loans. And so really our scoring, what we are building, is to try to analyse the business, the informal business.

Brendan Le Grange 7:03

You know, it wasn't so long ago that if we're talking about the informal economy and talking about micro loans to micro businesses, it would have been done in a very manual basis, it would have used agents on the ground who personal relationships or group lending or some way of trying to get the money there. But without any of the analytics that's now available to us.

So talk to you about how you are using AI to solve and to calculate credit worthiness in a way that's more applicable, more useful, more accurate for this sort of customer base?

Viviana Siless 7:37

What you are saying about doing it manually, we obviously started there, right, because at the beginning we didn't have any data to train an algorithm, right.

So at the beginning you need to like give loans, not know if they will pay you back or not, and based on those results, you can start training the model. So really, at the beginning, we were doing a lot of manual work, I can say from experience, or at least the experience that we have doing it manually, that really doesn't work!

We are going through a journey. We also change a lot the application at the beginning, we were in marketplace and we were hoping to use transactions that will happen inside the marketplace, we also have a crypto token. And we are hoping that transactions will be with that crypto token, we pass that we are not building a marketplace anymore, we are really focusing on having an application where you sort of ask for a loan, they have some data, we analyse the data and based on that we decide we give them the loan or not.

We still have a crypto part in that we are willing lending protocol. And they are we allow anybody to be an investor, so of these informal businesses, and we disburse the money through this lending protocol on blockchain, but it's not a currency anymore. And I'm saying this because the data that we have been acquiring Are you seeing or thinking Have you seen in each stage have changed greatly right at the beginning, we were hoping to use transaction from our marketplace. And now we are really thinking like, well, what information I can ask to a user that equally useful for me to understand the business and if it's profitable, or if we can give the loan as well with what information can I get without asking the user right?

What can I go outside on the internet and get for example, if you use open banking manually, you're going to do very basic computations. They are very short finger in a month or a week. But like if you are able to put many variables in one model, then the model will find the weight for you. I'm not really more accurate but not only more accurate, it also will have more power because he's so they're like all the steps that can help you work with massive amounts of data with different dimensions of data. That if you do it manually is really it's not possible really.

Brendan Le Grange 9:59

So In the dangerous area of talking about things I'd know very little about, but my understanding of AI, in the credit scoring space is, you know, in the developed markets, we see it and we can add it to our modelling. And it can make modelling much faster for sure. But from an accuracy point of view, it gets a little bit better. But actually, regression models were doing a very good job and are still quite fine. But where AI comes in is this scruffy data, this data that's not neatly organised and reported every month in the same format by all the banks, which is actually the the data from an informal economy where it doesn't all look the same. It comes from multiple sources, it comes up and it disappears. So it's really well placed to help solve that problem.

But then I guess so when you think about where the AI hubs and the investment are, it's an area that probably doesn't naturally get a lot of AI effort there. So talking about that side of things, obviously, you're also training AI professionals in the region, how much is happening in that space in Latin America? How much investment and effort are people putting into using AI to solve these kinds of social issues? Yeah,

Viviana Siless 11:06

I think from efforts is a lot for really talking about informal economies, like, there are many digital lungs today in Latin that are giving micro loans for consumption, right? You use AI for that, really, but we're analysing informal businesses, I haven't seen this many people working on that I'm really trying to get the score to understand the business. Something that happened to us is that we we need to ensure that the person actually has a business, you know, because this information is not richer.

So the US load of information where they sell, we asked for photos of brokes, we also ask for a video of like the shop and the person planing. And so that's really something that you know, you couldn't do it or otherwise. And we are really working on we like this image analysis, a media analysis, voice analysis. Yeah, sometimes, by looking at the application only, we were not really sure that the person, you know, this person really has seriousness, or is he showing some clothes that he has in his house, it really opens the door for more and more information, like social media as well, like most of the people's on social media, or on our platform for new gun, a scrap that on blade into the into the your analysis.

So in that sense, yeah, it allows you to explore more options that are really targeted to your user, like, for example, our users tend to use more Instagram or Tiktok, you know, and so maybe, you know, sort of the platform's go, but maybe if you are targeting on our population, you know, maybe Amazon will be the place to go

Brendan Le Grange 12:56

when they're borrowing from you. How fast is that process? How different is it for them, compared to what it would have been a few years ago?

Viviana Siless 13:03

Yeah, it's really fast. We tried to this worth the money, like within a day or three maximum. So there were some is fast, but then we still how old operations going on that we need to make them faster. Also on the way of experience for the user. The digital literacy is not great, really. But everybody knows how to use WhatsApp. And also, generally WhatsApp is free. So you don't have to pay additionally for that data. So everybody uses WhatsApp, but maybe people don't use like real internet because they have to, they have to how, like they have to pay for that. Right. And so we started to have our application form through WhatsApp. So today, they can answer all the questions of law the most, it's the only information they can do through WhatsApp, he was he a huge change.

When we started to implement the WhatsApp application. It was a pain for us, because we had to map that data that came in a different format that maybe had some errors as well. So it was really much, much harder. Like I wouldn't recommend that to anybody to no more than on WhatsApp.

But we notice like an increase of users of loan applicants, because it's something that they are used to doing. It's very easy for them.

Brendan Le Grange 14:26

I think that when we think of financial services and borrowing, it's intimidating and it's more intimidating, the further you are removed. So when you are a small business in the informal economy, and you think of borrowing from a bank, you're often self excluding yourself, you think I'm not going to qualify, I'm not going to do it. And then you fall into parts of the loan shark and we need to get rid of some of these hurdles in the financial services make it easy.

So it's great to hear that you're meeting their customers kind of where they are, and to hear that you're really growing very fast. So let's Talk about that, as you got bigger plans, where are you on that growth journey? I think you're raising capital as well. Yeah, we

Viviana Siless 15:06

actually, luckily, we close the seed round just recently, I think last month. And so yeah, we are really preparing for series A right now, we for our blind spots, really to start in Mexico. But now I think our plan is really to be able to serve really well in Colombia and to grow later to Mexico, or maybe Ecuador, Panama. The countries are really around Colombia, because the culture is more clothes, really. And

Brendan Le Grange 15:39

if anyone listening wants to follow their journey wants to see what you're doing and look at it in more detail. where's a good place for them to go online to start a conversation or to read more about what you're doing?

Viviana Siless 15:51

Yeah, so our website is keepo.com.co. And you can find us in LinkedIn and in Twitter and Instagram. I'm Jana, Phil, is the CEO Merces VR this as well. Great collection if you look further, and yeah, we are in every social we are in Tik Tok as well. Great.

Brendan Le Grange 16:13

Thank you so much. I'll put those links as well in the show notes for anyone who wants to go. But thank you for making the time to speak to me. I think it's a project that can inspire a lot of people in other areas as well.

Viviana Siless 16:25

Thank you for the invitation.

Brendan Le Grange 16:29

Galina Andreeva, welcome back to the show. We're recording here in a room at the credit scoring credit control conference, which is just coming towards a close. It's been a fantastic event, I think, for anyone in the industry that gets their hands dirty in the data in particular. So I've really enjoyed myself. In fact, as I was saying earlier, one of the biggest challenges I've had in organiser interviews is all the attendees want to be at the talk.

So it's very hard to pull people away. So congratulations on a great event. I would say most of the people I bumped into around Yeah, hell coming back year after year after year or two years. After two years. After two years. I suppose. We've got another one coming up in two years, of course. But before we talk about that, what, what are some of your thoughts or highlights from what's happened these last three days

Prof Galina Andreeva 17:13

busy, nerve wracking and very satisfying. We reached almost the maximum capacity.

Brendan Le Grange 17:19

And from around the world as well. Lots of different accents and languages heard.

Prof Galina Andreeva 17:24

We have delegates from 34 countries, 80% of delegates come from the industry, more than 100 papers split into five parallel streams.

Brendan Le Grange 17:36

It's a nice interchange there. And I've sat in a range of different talks from some being more familiar to me to some really technical ones getting into generative AI and how those models will have been built at some interviews where I've got no to all sorts of words, I don't understand.

But it's really been refreshing on that front people getting into the technicalities of machine learning. When is it useful when it's not? How do you stop overfitting? How do you make sure that your policies are unbiased, but also then going into more human aspects, there have been talks on the psychological impact of financial distress some more mundane stuff on just you know, scorecard monitoring and scorecard building lots of climate change. Set on a talk about cyclones in Mozambique, and the impact that's had and how the lender responded in very practical terms with payment holidays and such to some talks that are really much more about the modelling of climate change. Yeah, the

Prof Galina Andreeva 18:33

variety of talks and a variety of different topics touched upon is truly mind blowing. This is the modelling conference. And I would say the majority of the talks adhere to this main topic, main objective, but I was increasingly pleased to see how many divergences are they're into mostly societal aspects of credit, and how credit models and credit decisions affect different aspects of general life. And I'm always amazed how responsive the industry is, to any new challenges. You have a new problem, and almost immediately, there are some solutions, maybe not perfect, but there are solutions and this conference is really the place to bring them to receive positive or negative but useful in any case, feedback.

Brendan Le Grange 19:34

Every talk I've been to, regardless of the stream, regardless of it was a roomful of people, or a smaller audience, regardless if it was been presented by someone very comfortable speaking or somebody who is a little bit more reserved, great questions and feedback. As you said, from the audience, there wasn't a single one where there were no questions. I think every speaker had some really good feedback and thoughts added. Many, many people I spoke to there have been here a dozen times I've been here five times to seems to be once people come, they come back again. So, in two years time, I guess we've got another event coming up if they want to make a little pencil mark in their calendars, when are we looking at getting back into Edinburgh again to do this all over?

Prof Galina Andreeva 20:13

Traditionally, this will be the last week of August 2025. So the exact date will be announced slightly later. But people interested can already put this into their diaries. And just one last point, life does not stop after the conference credit Research Centre will be still ear operating until the next conference and if people have projects do get in touch with us, we are eager to discuss them.

Brendan Le Grange 20:48

And thank you all for listening.

Please do look for and follow the show on your favourite podcast platform and share the updates widely on LinkedIn where lending nerds are found in our largest concentration. Plus, send me a connection request while you're there.

This show is written and recorded by myself Brendan le Grange in Brighton, England and edited by Fina Charleson of FC Productions.

Show music is by Iam_wake, and you can find show notes and written transcripts at www.HowtoLendMoneytoStrangers.show and I'll see you again next Thursday.