Subscribe
Share
Share
Embed
Michele Tucci, MD Americas and Chief Strategy Officer at credolab, spoke to Rudolf Falat, founder of the Voice of FinTech podcast, about the importance of credit scoring and the need to improve credit assessment through alternative approaches.
Aiming to inspire entrepreneurs around the world to launch their new ventures. Connect FinTech enthusiasts with start-ups, incubators, accelerators, investors and incumbents.
[00:00:00] Voice of FinTech.
Welcome to Voice of FinTech at podcast mapping out the Swiss and global FinTech scene connecting FinTech enthusiasts with startups, incubators, accelerators, business engines, and VCs, and incumbent interested in partnerships. Voice of FinTech will help you navigate the FinTech ecosystem here. You can listen to the startup, found the stories.
What investors and incumbents are looking for when dealing with startups and find out more about resources provided by incubators and accelerators. My name is Rudi Falat, and I'll be hosting this podcast.
Hello and welcome to Voice of FinTech. Today we are going to travel to Miami, Florida, and we're going to [00:01:00] talk to an Italian base there because he runs America's business for credolab. And we are going to talk about alternative credit scoring. Why is it needed? How can it help you as an individual? How is it connected to your payments experience?
Welcome, Michele. How are you today? Thank you very much, Rudi. I'm very good. Thank you for having me. Brilliant. So Kel, tell us about your background. How did you get to do what you do today and why have you decided to join a startup or scale up? And I can also explain why in Florida. Yeah, it's been a long journey, you know, about 22 years and some of it in Italy, most of it outside of Italy, and some of it with traditional banks were traditional tech companies and.
When FinTech became a thing, then to be already in the technology space doing mobile payments, mobile POS [00:02:00] terminals, and AI applied to consumer lending, and then eventually machine learning algorithms and big data applied to credit scoring, fraud detection, and even marketing. That's what we do at Credolab.
So it's been a long way. Mostly, let's say in the started in a traditional way, business, economic science, a degree in business, economic science, then an MBA in the us and I never went back to Italy after that. So I lived in four continents. And since February in Miami, why Miami? Because the US is one of the hottest countries for OLA as a company.
And from Miami. We can also manage our, our current position in Latin America. So it, it helps with both expanding into the US while also being closer to our clients in [00:03:00] Mexico, Brazil, and across other Latin American countries. Uh, of course. So obviously there are two things here. Miami has been a big hub for latam for a long time, and then I think post pandemic, this is also over Florida, is one of the states with the biggest immigration within the United States, right?
Yeah, you're right. So revisit, revisit their life and they moved down to Florida, which contrasts to what I heard when I was a bit younger, where people used to say that you can only do business in cities with bad weather. You cannot do it in Florida. Yeah, Covid helped fairly. You can, Yeah. And Covid helped with that as well.
So a combination of fantastic weather, a combination of low tax income tax and proximity to markets where the US does business with helped Florida become what they are today. So yeah, it is, you're right, is the state, US state with the highest influx of people moving. From others. Exactly. [00:04:00] So you mentioned ai, machine learning payments, credit scoring.
In your experience, so if you just think about the customer journey or the problems and pain points that the clients of incredible lab have, what's the linkage between payments, credit scoring and lending? Because payments, you have a journey from A to Z. How, How does credit scoring relate to this?
Because it sounds to me it's a little bit more related to lending and borrowing, right? Yeah. I think about this from a consumer lender point of view. The moment somebody is approved for a loan, the lender has to disperse money. That those funds. And so that's already the very first touch point where payments come across, come into the consumer lending space.
So you need to disperse funds into a prepaid card, into a wallet, into a bank [00:05:00] account, and then through the life cycle of that loan, customers have to repay the loan. So payments, again, play a role there because you need to offer as. Pay repayment channels as possible to make sure that you don't give any reasons to customers not to pay back or to pay you back on time.
So there is quite a deep intersection between payments and consumer lending also. If you think about mobile wallets and the way they have grown into day to day payment methods, a wallet is useless until there are funds in the wallet. So there has been a growing trend of wallets offering lines of credit.
Uh, available within the wallet itself. So you don't need to, uh, link your credit card or debit card or bank account to the wallet, but you can just open up a revolving line of credit within the wallet itself. [00:06:00] I see. All right. Now you also provide alternative credit scoring, so, Let's level set for people from outside the us.
In the US, people pay attention to their credit score, right? Because it really impacts the interest rates and the affordability of the credit for everything you have. In some countries, like in Switzerland, it's a bit binary. Either on you are on the list of bad debters or you're not. But if you are, then your life is finished, but there is nothing in between.
Score six, Sunday, six 50 or what have you, or whatever is a good score these days in the us I don't know. Why do we need something different? Because you already have that kind of system, a lot of credit scoring companies as well in the us. Why would someone go for a credolab solution? I. Oh, the Life of Americans revolves around, around having a good FICO score, a good credit score.
It is used not just for credit loan applications or credit cards, but [00:07:00] also for insurance purposes. Even when I rented my very first apartment in Miami, I had to give my credit score to the landlord, and I didn't have. No, I'm recently a white collar immigrant in the. So by definition I'm a theme file, so I don't have enough credit history within the US to prove my credit worthiness to a landlord, to an insurance company, even to T-Mobile, to have a post pay plan.
I had to give them a deposit rather than because I didn't have a credit score or a high enough credit score. Yeah, this is a problem for. About 25% of all Americans, people that there are about 54 million people in America without a thick file. And, and the example you mentioned of Switzerland is the same in France as well and a few other markets.
Brazil used to have a negative bureau only. So [00:08:00] you are only reported to a bank if you have missed a payment, for instance. Yeah. Now things are changing in some countries in, so why do banks or insurance companies, lenders at large need a solution like credit? because we compliment the information they have.
From a traditional point of view, any risk assessment is done. On a combination of data points, the most obvious one is the credit bureau score. The other one is social demographic data that people share with the banks through the application. And there are a few knockout rules that always apply. Now, if you are below 18 years old, then you cannot even apply for a loan.
So these are, and in general, a woman tends to be a better repayer than a man, or from a bank point of view, a full-time [00:09:00] employee tends to be less risky than a self-employed. And so these are all the traditional data points that are available to any applicant. What banks and lenders in the US have increased, uh, increasingly done is to leverage additional data points such as data or utility bill payments kind of data.
Or a movement that started in the us sorry. In Europe with the PSD tool regulation open banking, there is not such regulation in the us but there are a few companies. That are providing access to bank account data. But guess what? Even in my case, E, even if I connected my bank account to any of such providers, my US Bank account doesn't have a solid history or long enough history to generate.
Meaningful information for that particular bank. [00:10:00] So we are back at square one. Lack of data still is a problem for a larger percentage of the population. So Credolab comes in, we are able to provide a behavioral assessment of 100% of all applicants. So there are multiple benefits including the we. We help decrease the data asymetry between Scoreable and Unscorable populations, which from a data science point of view, already brings a benefit because you are able to assess people with the same amount of data.
Now, I don't know if I'm getting too technical. No, that's perfect. But basically you are focusing on the first time applicants, it sounds, or people excluded by the traditional grid scoring from getting access to credit, correct? That is correct, and that is our sweet spot. However, our solution works [00:11:00] across the entire spectrum of people applying for credit.
So we have case studies even done by one of the largest credit bureaus in the world that prove with numbers that our scores are predictive, not just for new to credit customers or theme files. But also for thick files, so people that have enough credit history to qualify for a loan without needing the bank to process alternative data.
And yet, because of the local relation between our behavioral source of data and. Traditional sources like social demographic data or credit bureau data, and also transactional data, we're able to uplift the predictive power of the models of these banks and lenders. So from a, if you are a chief risk officer, your goal is to [00:12:00] make better decisions.
So if you have access to data that can help you. The way you discriminate between false positives and false negatives, for instance, you will adopt the new source of data. Then the question becomes only about price. How much should you pay for that additional source of data given the marginal uplift you're getting into your general model.
Sure. But let's dive into the examples of this edition data that you mentioned. You mentioned rent and things like this, but if you just come to the US today, you just sign the lease, you open a new account and there is no information there. What can, what else can you use? So in that case, what is the safest choice of a lender?
It's just to reject your applic. Now the, We are a complimentary tool to [00:13:00] existing the models, So the applicant will still be assessed based on whatever information is available to him, either provided by the applicant herself or through credit bureaus. The what we add is a behavioral assessment that predicts the probability of you, for instance, missing a.
Or the probability of as an applicant being similar to a confirmed fraudulent applicant. So we add a layer of assessment that is based on behavioral data, and in particular we have two solutions, one for web and one for mobile. On web, we analyze the way, uh, people type or interact with the user interface.
Or gestures, how fast somebody scrolls, how fast somebody types patterns. And then we find [00:14:00] through machine learning and through our proprietary data modeling, we find correlations with similar behaviors of confirmed risky and fraudulent applicant. On mobile, we add on top of this, we add also data from the digital footprint.
And please bear mind, this is always privacy consented data. It's always data that we collect after we receive the privacy consent, uh, after we have anonymized the data sets and after we have de personalized data. So, We don't know if Rudy or Michaela is applying for a loan. We just know that one data set is coming through the door onto our cloud, and we calculate the probability of that data set to default or to be fraudulent.
So the it's privacy consented, [00:15:00] permissioned, anonymized, and de personalized data. Now, having said this, what data do we. On a mobile, we look at the way people use the mobile phone, what kind of apps they download, what kind of categories of apps. For instance, we know that for one client, if a, an applicant has more than 13 apps in the finance category.
They could be 2.6 times riskier than somebody with less than three apps in the finance category. Now, and this is just one example, we most lenders apps have access to medium. The media permission, which allows applicants to either take a selfie for KYC purposes or to upload a picture of their national id.
Now we mirror this permission and we access [00:16:00] metadata Underlying metadata. Definition of metadata is data about other data, so we don't actually process. The content of the image, but we look at how many images were created, how many images were created during the day versus night, or weekdays versus weekends, and this is very behavioral in nature.
What if all delinquent customers tend to take the same amount of pictures? In the same time span, so that could become a predictive feature into the model. We look at address book, if we have the permission from the app and from the user. We look at how contacts are created, how many contacts, again, are created with what is the percentage of context with image associated or a personalized male.
Again, this is all behavior. We are looking at the behavior of the user and in the absence of credit [00:17:00] history worth of data, this can be used to determine your probability to default a payment. We are not absolutely that. That's great. Will you help me to get a credit card without going through a prepaid card?
When I come to the. And I signed a lease and I have a new job. It's full time. I got the Visa. It's all legit. And I only need the credit card because I need to buy the furniture for my first month until I get the paycheck. Yes, Credit is a FinTech company, so we provide. Insights for risk management, for fraud detection and for marketing segmentation, for instance.
But we are not a lender ourselves. So the decision to, of course, but do you work with the credit card companies or with the banks? Yeah, we do. We do, and we work with a number of different type of players. Banks, neobank, challenger banks by now pay late traditional point of sale lenders and all sorts of digital lenders.
Even a right daily app offer. [00:18:00] Working capital solutions to their drivers. Uh, that's an example of client or even a fly now, pay later, uh, type of client. So they all use our technology to improve the way they make decisions. All right. Understood. I think obviously that should help. Now, maybe one basic question though, because if you ever had an anti fraud training, if you were audit or compliance, then the traders would say, Look at the people, their role and their lifestyle.
So you did mention some banks have access to the media, et cetera. What about to social media? I'll give you an example. These fantastic life on Instagram, then you check their job on Linked. And you can see they cannot afford it. Yeah. So that should red flag. That should be easily picked up. And you don't even need an algorithm for this, but of course if you wanna do it at scale, you might need technology.
True. What do you about using this [00:19:00] sort of insights historically, the way companies, they look at social media activity, number of posts, number of likes, but they ended up not being predictive for. Now at Ello, we do not use any social media profiles or activity. We don't process any personal data whatsoever.
We don't even know the email or phone number of the applicant, so we cannot check what is their lifestyle on Instagram. And frankly, we don't need to do that. But to your point, if the bank finds a way to use social media data, For to verify the lifestyle. Then it's in that information becomes predictive for them and uplifts their ability to make good decisions.
Then they should, but of course it has to be all in compli privacy, compliance, not the user has to know what's. Being accessed by the, Of course, especially in Europe as you mentioned, But so obviously Credolab doesn't spy [00:20:00] on you. You need to get content from people and then you can do some digging on their phone activity and come up with some predictions that help them to improve their life because they, that helps them to get access to finance.
But how is your perform? If of predicting bad debts versus traditional methods. If you compare, you mentioned the the FCO score in the us. If you compare your method and what that score would tell you about me defaulting on a credit card debt by this end of this year or not, how would you Fair? So we measure predicted power of our models in by using the gen coefficient and how, although technical.
Give you a brief explanation why Genie. The Genie Coefficient is widely known to measure as a measure of income inequality in risk management and modeling. Genie is a way to measure the predictive power of a model and is a coefficient that goes between zero and one. Zero [00:21:00] is like flipping a. You, your model has a random outcome, so you cannot predict the outcome.
A gen one is when you can always predict the right outcome. And, excuse me, so the A model built on social demographic data plus credit bureau data. In a developed market may have a genie of 0.6 to 0.7, so again, the coefficient goes from zero to one, so it's very predictive. Now the A starting model of Cradle A built only on digital footprint or key patterns, behavioral data that is proprietary to credolab.
Has a gene coefficient of 0.2. This coefficient, the model, the digital model is used as input into the general model, so [00:22:00] it's addictive to what the bank is already doing and has very local relation with existing traditional data or social demographic data. Correlation is below 10%, so whenever you add the cred score into a model, You are able to uplift the overall predictive power of the model.
So we are not here to replace what the banks are currently doing, but to give them tools to make their own models more predictive. Now we have cases uplift. Uplift by how much? It could be 10%, it could be 30%. So we have cases where the digital score in isolation has a gene of zero point 50. So almost a part with the genie of a model built with credit bureau scores.
Now, the additional benefit is that we can provide a score for the entire incoming population, a hundred percent [00:23:00] penetration heat rate. That is how technically credit bureaus refer to. So the percentage of customers for which they can return a score is defined as heat. Our heat rate is 100%. So you mentioned who your clients are.
Who are your key clients, though? The neobank it sounds, or BPL companies. By the way, do they actually do any credit assessment? Because some of them just talk about smooth checkout experience as if they are not checking anything at all when it comes to critic. So who are your key clients? Yeah, we have some of the largest, uh, newer banks in the world that we.
Uh, tonic in the Philippines. We had time bank in South Africa. We have some of the largest also by now, pay later, but we cannot mention the names publicly. And for Neobanks is important to lend money because it's important to their survival. A neobank that only collects deposits [00:24:00] will lay interests out of, uh, venture capital money.
So it is not a sustainable business model. The solar new banks realized they need to go into lending, the faster, uh, they will become profitable. And so some of the clients we have already realized it and they are already lending to their customers by pay later. Consumer finance companies, the, we are an embedded scoring technology that can be embedded in any online and mobile front end.
So regardless of how these banks or, and we worked with 20. Five banks or 26 banks today in 37 countries. So regardless of where the bank operates, where the lender operates, we can provide the same data points to everybody. If you are a multinational company, if you're a Bino operating a multiple countries, we can provide you with the same [00:25:00] amount of data standardized every.
Understood. So that leads me to another question. So where are you based as a firm altogether? You are based in Miami, but the credolab is active in which markets? So the headquarter is in Singapore. We have offices in Miami, in London, in Jakarta, and a distributed team in 17 c. So the fastest growing market for us today is Mexico.
We have about 50% of our revenue coming from Asia, not just Southeast Asia, but also India. And we see growing opportunities coming from Nigeria, from Kenya, from from Brazil for instance, or other markets that could be small, but they have all problems with lack of data. And this is a common theme across all markets developed or developing.
And so 37 countries [00:26:00] in total. And so what are your plans for scaling up, especially in the US or the Americas? You said Mexico in LA is is a key market, but in other countries you have only negative credit scoring kind of system. In the US you have established credit scoring companies. Uh, what are your thoughts on the US market when it comes to alternative credit scoring?
The US market is a tough nut to crap and the common feedback, and we've been in the US actively selling in the US only since February in 2022. So the common feedback is that. People didn't know that this was even possible. They didn't know that behavioral data was even remotely possible for credit scoring or fraud detection.
So there is a lot of interest, but there is also, there are concerns related to compliance and federal laws, how we comply with federal laws. So the sales cycle is a [00:27:00] bit longer than we expected, but we are see buying signals in other c. Where regulations are not so strict, then the, it becomes relatively easier to sell and we can sell even without having a local presence.
So today we put people in key markets where we felt, well, we prioritized that these markets based on opportunity, based on the credit pool, based on the lack of credit bureau data, based on smartphone penetration, for instance, or digital origination. We had about 26 macro variables, including in the model where that we use to prioritize countries and uh, and now we are raising serious.
So the objective is to push on the accelerator even more. As soon as we raise money, then we can invest more and build a, a bigger footprint, bigger sales network. Brilliant. Now [00:28:00] before we go, maybe one other question, which is quite essentially is how do you make money? Oh, we are a SA business, so we make money with a subscription model.
And the, we charge like credit bureaus tool for, there is a term based subscription and a consumption based subscription, but basically we charge, um, pair score, delivered pair anti fraud check delivered rich data set, delivered for marketing purpose. From the more our clients use our scores, the less they pay for as a unit economic, So it's a volume kind of game.
But how does the initial setup work? Do you have an api? Do there is any, Is there any integration work needed? And do you charge a special setup fee for that, or that's all included in the usage fee as you. Yeah, so we have an embedded scoring technology, so we need to embed our [00:29:00] mobile SDK and web SDK into the front end of the client.
That's relatively easy to do. The mobile SDK takes about five main days. The web SDK just one mandate, the, and then we can start collecting data immediately, metadata, and at the moment we, And then the second part is also the api, so we deliver. Through the API in real time without any lag. All right, brilliant.
Sounds very interesting. So good luck to you and Ola, but if you'd like to hear from people who just listen to the podcast, what kind of people would you like to hear from most and what's the best way to reach out and get in touch? Yeah, so we would like to hear from risk managers, fraud managers, and increasingly we get interest from marketing departments that are looking to.
And reach the data sets they have today. And so we are able to help them build [00:30:00] actual personas, not just segments. The best way to reach me would be through my email and tochi.com or through our website. We have the ability to book at demo, speak to an expert. So there are different contact points to reach out to us.
And Tucci is spelled as Stanley Tuchi, correct? It is. . All right, So thank you so much and good luck to credit a lab. Thank you very much. Have a good one.
Thank you for listening to Voice of FinTech podcast. If you haven't already, check out also voiceoffintech.com, where you will find all the episodes and additional resources related to the podcast. You can also subscribe to Voice of FinTech on Apple Podcast, Spotify, Google, or any other podcast app that you like.
If you have any suggestions on the topics or guests or how to make this podcast better for you, please email us at info@voiceoffintech.com. Happy to hear from you. Thank you.[00:31:00]