Juvo Founder & CEO Steve Polsky
Martine Paris
When Warner Brothers acquired movie discovery site, Flixster, and its reviews site, Rotten Tomatoes, in 2011, its President and COO, Steve Polsky, set out to explore how scoring consumer behavior data from mobile activity could be leveraged to solve a growing global financial crisis, one where two-thirds of the world’s population have been excluded from transacting in the digital economy because they lack formal financial identity.
To address this problem, Polsky founded Juvo to serve as a data-driven micro-financing provider that uses machine learning to extend emergency airtime loans to prepaid phone users when they run low on their balance. As customers make timely payments, they unlock access to progressively larger loans and build credit that they can use to qualify for a credit card and other financial instruments.
Inspired by the World Bank challenge to achieve Universal Financial Access for a billion people by 2020, this San Francisco-based Series B startup has raised $54 million with rounds led by NEA and Wing, and strategic partnerships with Samsung NEXT, Nokia, Cable & Wireless and other leading mobile operators around the world.
At the heart of their offerings is the data set they’re mining which boasts over 5.6 billion data points being constructed real time for over 250 million people daily across 26 countries and 4 continents. I had a chance to talk with Polsky, his VP of Data, Aristotle Socrates, and his Senior Director of Market Strategy, Josh Gosliner, about the data they’re collecting, behavior that effects scoring, and how AI-crafted identities is moving us to a more financially inclusive future. The following conversation has been edited and condensed for clarity.
Why did you start Juvo?
Polsky: After Flixster, as I was setting up an incubator for angel investing with my former founders, we were astonished to find stats that four-fifths of the world is on prepaid mobile phones and spending a trillion dollars a year on service. Amazingly, they interact with their phones in tiny increments, topping off $0.25-$0.50 cents at a time. What dawned on us was what if we stop thinking about the phone as communications and start thinking of it as payments because for most people their most frequent formal financial transactions are on their phone.
People top off a crazy number of times a week and yet they’re treated as an incredibly risky population. That’s what prepaid is all about. You can’t make your phone call until you give me the money because I don’t trust you. And that’s basically how these people live their lives. The only people they can transact with are people they can walk to and hand cash to.
We realized this just isn’t right and we can solve this problem by helping them create progressive trust through micro airtime loans so they’re never out of service, first advancing them a weekend of minutes, then a week, two weeks, a month and so on. Using advanced data science algorithms on their payment history, we knew we could score the entire population of Indonesia real-time, with the ultimate goal of scoring the 5 billion mobile phone users worldwide.
Are you creating a next generation credit score system to compete with the legacy credit reporting agencies: Equifax, Experian and TransUnion?
Polsky: Yes, 68% of the world has no financial history. They don’t show up in Equifax, they don’t show up anywhere. There’s no system for them. You could be the most responsible person in the world but if you can’t prove it, you’re treated as untrustworthy.
Do you charge interest on these loans, or fees?
Polsky: Our favorite way is to charge no interest and no fees but sometimes, depending on the territory, the mobile operator dictates the terms.
Where do you currently operate?
Polsky: Central and South America, Caribbean and Southeast Asia, parts of the European Union. We are not in North America and don’t have a presence in the United States, Canada, or Mexico.
What is your partnership with Samsung NEXT?
Polsky: They invested in the company under the idea that we would be able to help get a billion people in the world access to a smartphone. The first place we’ll be launching is in Brazil where we’ve been for five months. A good portion of the population doesn’t have smartphones because they think they can’t afford it and so once they reach a certain credit level on Juvo, they’ll be eligible for financing of a new Samsung handset.
How many subscribers are in your data set?
Socrates: We have 250 million subscribers in the superset of data coming from partnering with our mobile network operators to create loyalty and credit products for them.
Everyone who subscribes to the mobile operator network is being scored, whether they opted-in or not?
Socrates: Yes, we do that because to give a real time response within 100 milliseconds, we have to score everyone continuously, have all their machine learning profiles computed regularly, every day, every hour. For us to manipulate that data, there is an implicit opt in from users subscribing to that service.
What type of data is Juvo collecting?
Socrates: It’s quite transactional and widespread on the entire customer base. We collect data from the billing systems of the mobile network operator that gets shipped to us daily, hourly, every five minutes and real time from the mobile network operator and on the basis of those extracts, we construct profiles on individuals and join that data with high level macro economics that varies from region to region.
Every time somebody puts money into the system, there’s a recharge that could go into either buying core balance or a specific plan, like a weekend data bundle. After that, there’s a drill down where they can buy a package, like for WhatsApp or Netflix usage. You can see consumption of their balance at a more granular level from that type of data. And then we basically get metadata on all sorts of transactional information like when somebody makes a call, how long was it for, when did someone send an SMS, how many messages they send in that session. We get location information in terms of IP address and geographical latitude and longitude. What we want to understand is how much are people paying for mobile service and how frequently are they spending money on it.
How many data points do you have per subscriber and what are they?
Gosliner: To date we’ve constructed 120 million profiles, each with 50 data points or attributes. We expect to be at 250 million profiles by February or March. The attributes are the consistency of consumption in their wallet and can be looked at in terms of four blocks of data: a social block, a consumption block, a location block and a regularity block, and within each block there’s a drill down. What’s their propensity to be out of balance? Do they display normal social patterns? Do they have thee active people they talk to, or 10 per month? Do they send messages to 100 people or just one person? What city is this person located in? From location information, you can even infer if this person has a full time job.
Is the credit score affected by engagement with ads?
Polsky: No.
Are you collecting data on browser history? The videos watched on YouTube? Physical items being purchased?
Polsky: We’re not looking at information on your phone, what you’re watching, where you’re surfing, clothes you’re buying, or who you’re calling.
We look at people’s consumption patterns of mobile usage. So the average consumer has transacted 31 times, that behavior is predictive to the 32nd payment. We’re looking at how much people use their phone, how frequently they top off, how much they’re spending. Like if they spend 25 cents at a time, do they do it once every other week or are they spending more erratically like 75 cents then $3 then $2 then $5.
Gosliner: We’re not like an ad tech company where we’re trying to get a detailed picture of a person relevant for advertising. What we’re trying to build is a credit system. If you think of the credit system in the U.S., it’s a data platform keyed off of our social security number, address etc. and there’s this profile of information and a proprietary score. There’s this regular kind of transactional information that historically goes back to when your credit was starting to be collected on you. And from that information banks can make pretty good decisions. That’s sort of our philosophy, we just want to know if this person is volatile or not volatile, if they have normal patterns or not. We’re going from zero to 60% efficiency, rather than the ad tech industry in their approach to data which is more like 98% efficiency.
What drives the credit score up or down, other than timely repayment?
Gosliner: Credit favors loyalty and consistency. There’s a relatively steady population you want to lean into and a volatile population you want to avoid. For example, SIM hoppers chasing deals show up as inconsistent in their behavior. They may not be bad actors but there’s not enough history on them to know otherwise.
What is in your AI toolset?
Socrates: Amazon Web Services, Redshift, DynamoDB, S3, MapReduce. The toolset is so rich these days, we’re able to operate with a just a small team of 15 machine learning engineers.
How long does it take for someone to qualify for a credit card?
Polsky: With our progressive lending model, there are four levels – bronze, silver, gold and diamond. Diamond (or platinum in some countries) is the level you can access credit outside of phone usage. It usually takes 90 to 120 days to achieved diamond status.
At what point do Anti-Money Laundering/Know Your Customer (AML/KYC) laws require that you connect the financial identity you’re creating with the actual identity of the user?
Polsky: What’s nice is that we don’t have to ask for the social security number at the beginning because it’s not a bank account or credit card. Once users graduate to a checking account or apply for a credit card, they can no longer be anonymous. You need their name, government identification like a social security number, a physical address and other personally identifiable information.
What percent of your users are at that level?
Polsky: I think 25% are at gold and platinum in our more mature markets.
Are you planning to port Juvo credit scores into other areas?
Gosliner: Yes. With our Financial Identity as a Service (FiDaaS) platform, we’re leveraging our understanding of consumer behavior into other services, like helping ride-sharing companies like Lyft evaluate whether drivers are volatile or stable. Our scoring system can identify the pool of drivers who are good bets.
In the prepaid space, 70-80% of individuals display very consistent behavioral patterns, and then there is the 20-30% which could be characterized as volatile. It’s not that they’re nefarious, they’re just not consistent on the mobile network.
Socrates: I’m excited about FiDaaS because it will bring greater transparency to data monetization where the user can have more control over the data they’re providing for what benefit. We’re in countries that are data deserts and we’re surfacing that data so people can be approved for more transactions. For example, as Josh was mentioning ride-sharing, in Argentina, 60,000 drivers have been declined by Uber not because they’re bad but because there isn’t enough information on them to say yes. We’re trying to fill in the gaps and leverage the trust earned in the Juvo ecosystem to infer driving behavior.
You operate in the European Union where the customers right to data privacy and right to be forgotten is protected by GDPR, how do you comply?
Polsky: We architect for it.
Could Juvo create an alternative ID than anonymizes personally identifiable information but can serve as a thumbprint unique to each user? For example, drivers’ licenses get scanned to enter office buildings, ride scooters, check-in at hotels and go to bars, but it gives away more information than is required to ID the person.
Gosliner: We have the ability to do this and might as we gain traction in certain markets.
Data privacy concerns have been growing, and to some extent the desire for anonymity, particularly with the issues raised in films like Snowden and The Great Hack. How does that factor into your credit scoring algorithm?
Gosliner: Fundamentally somebody who is choosing to remain anonymous is opting out of the formal credit system. You cannot create a digital identity for someone who wants to remain anonymous. Remaining off the grid is a luxury that is afforded to people in the first world who’ve already had a significant amount of their data exposed in markets where people are able to get access to everyday products and services. It’s just not a concern of the unbanked where credit means essential access to critical financial services.