Frustrated with the limitations of traditional credit scoring methods, innovative start-ups are finding new ways to analyse borrower risks using social media and other online data.

The business of credit scoring is experiencing something of a revolution. Traditional scoring techniques are being challenged by a new generation of start-ups that are using unconventional information, including social media networks, to reach out to consumers previously unable to obtain a credit score.

Much of this innovation has been driven by the emergence of big data analysis, in which vast amounts of digital information is analysed using complex computer algorithms. By tapping into this wealth of information, a new generation of credit scorers is now able to provide credit access to underbanked and underserved groups in both developed and emerging markets.

Meanwhile, many established scoring and analytic companies, as well as banks, have also invested heavily in research on unconventional and big data analysis. The outcomes of this research can be seen when the metrics for creating some of the more recognised credit scores, particularly in developed markets, widen to encompass a broader range of previously unused data points. The key distinction is that very few of these more established players rely heavily on data sources, such as social media networks, to create their score.

“For a long time there were only occasional start-ups in the area of credit scoring. Now we’re seeing a lot more start-ups because of the rise of big data and the rise of analytics,” says Keri Kramers-Dove, vice-president and general manager of international scores and analytics at FICO, a pioneer in the global credit scoring industry.

In the beginning

FICO established the consumer credit scoring market in the US in the early 1960s before introducing its first generally available credit bureau score in 1989. From there, the industry quickly grew on a global scale. Typically, conventional credit scores are based on consumer information gathered by a credit bureau. From here, specialist analytics companies such as FICO may generate a score to provide to lenders and back to the credit bureaux, or the lenders themselves will generate custom scores, based on this information, to support their activities.

This conventional scoring process has traditionally made use of a number of standard information sources, including electoral roll data, court records, previous repayment information and bank account data. While this approach has generally worked well, as with any system it has some deficiencies. In developed markets, young people and recent immigrants, among others, tend to suffer due to a lack of historical data. This same principle applies in emerging markets, which typically lack extensive consumer data covering the broader populace.

With a new middle class springing up in many emerging economies, access to credit as a means of economic empowerment has become more important than ever. As such, the use of unconventional data can play a significant role in addressing the financial inclusion of these groups.

Jeff Stewart, chief executive of Lenddo, says: “We think financial services is about to have its Napster moment,” referring to the disruptive technology the online file-sharing service launched to great alarm in the music industry in the 1990s.

Borne of frustration

Lenddo, a social media-based lender and credit scoring start-up founded in 2011, was created out of a sense of frustration with the existing credit scoring. “Prior to the formation of Lenddo, we kept getting approached for loans by employees involved with our earlier business ventures in emerging markets, including China, Ukraine and the Philippines, and that seemed odd to us. These were educated and trustworthy people with rising incomes, but they were being underserved by existing financial institutions,” says Mr Stewart.

Lenddo employs social media, including Facebook, LinkedIn, Twitter and Yahoo!, to make an assessment of an individual’s creditworthiness. The company’s Lenddo Score is devised through an algorithm that analyses broader social connections, including friends, friends of friends and beyond, to assess a loan applicant’s social circles. Once a loan applicant registers, and opts-in their social media data, they nominate character referees from among their social networks to vouch for their application.

As such, having social connections with strong repayment histories can help to improve a first-time loan application’s score, meaning that both peer reference and character assessment play a vital role in the company’s scoring process. “This is how lending has worked for hundreds of years. For most of human existence, if you wanted access to credit, it was about your reputation and who you knew. This was true up until the 1950s. It was about community banking and bankers knew the character of people in their community,” says Mr Stewart.

To date, the company has 1 million members across 130 countries. Lenddo uses its own capital, and that of investors, to lend in its key markets, including the Philippines, Mexico and Colombia, issuing loans roughly equivalent to one month’s salary of the applicant. “Our mission is to empower the emerging market middle class; it’s where the bulk of the wealth creation is. They are smartphone users and generally this demographic is young,” says Mr Stewart.

Big Data Scoring

Other players are pushing into more developed markets. Big Data Scoring (BDS), an Estonian start-up that began operations in 2012, has found success in Scandinavia and central and eastern Europe. BDS relies heavily on Facebook, as well as supplementary online data sources, to offer a predictive scoring model. Its founder, Erki Kert, was formerly employed by a traditional lender and found the process of loan approvals too onerous.

“I often took my laptop to committee meetings when we discussed loan applications. A simple Google search on individuals gave me more information than their formal application,” says Mr Kert.

BDS typically analyses 7000 to 10,000 data points in its scoring process, relying most heavily on Facebook profiles. Typically, basic information, including an individual’s education, workplace and number of friends, is analysed along with their status updates, the Facebook groups they belong to and locations where they ‘check-in’. Moreover, the level of education and occupation of an applicant’s friends are also considered in this process.

“What we have seen is that the data on Facebook sometimes behaves differently than in real life. For example, usually people who are married or are in a relationship prove to be better clients for banks. With respect to Facebook, it’s the opposite. If you have stated that you are married or engaged, statistically it is a bad sign,” says Mr Kert.

Why certain data points, such as marriage, count as a statistical negative is unclear in most big data algorithms. The volume of data used during the scoring process diminishes the meaning of single pieces of information. While this approach ensures a certain blind statistical fairness, it also makes it difficult to pinpoint why certain online activity scores negatively.

Getting to know you

In conjunction with social media, BDS also analyses the type of device being used to complete a loan application, the version of the software and operating system, as well as an individual's location. “In our latest model, we can also conduct a public search for the loan applicant. This includes the location of the individual through their [internet protocol] address. Using various sources to support this information, we can work out how many supermarkets are in that applicant’s vicinity, and whether they live near schools or cinemas, [for example], to get a sense of their local neighbourhood,” says Mr Kert.

Similar frustrations with conventional credit scoring have led to the development of a unique approach from Hamburg-based credit rating service Kreditech. With the initial intention of developing a scoring engine to sell, the founders eventually decided to further develop their initiative by moving into the business of lending. “We don’t see a bank that is able to serve consumers globally, on a fully automated basis, independent of existing institutions,” says Rene Griemens, Kreditech’s chief financial officer.

The company’s algorithm analyses about 15,000 data points, including social media accounts, behavioural analysis, e-commerce transactions, bank account information, as well as more basic information points, to assess each application. The average decision time for a loan is 35 seconds, with a typical loan disbursement taking 15 minutes.

“Our scoring model uses a broad array of information. It starts with personal details, such as a person’s occupation, income and residency. Then we are able to measure behaviour during the input process for the loan, including whether an individual is copying and pasting information, their spelling accuracy and whether someone might be intoxicated based on typing behaviour,” says Mr Griemens.

The level of scrutiny offered by this type of big data analysis has given companies such as Kreditech a unique insight into a rapidly changing digital world. “We have identified that a certain gambling software regularly changes fonts on computers and our algorithm discovered that customers with this font typically had installed this programme, making them a greater credit risk,” says Mr Griemens.

Constantly evolving

The emergence of unconventional scoring techniques has profound implications for the existing scoring architecture, as well as the broader financial inclusion agenda. Yet, innovative work is also being conducted by established players in the market. FICO, for instance, has launched its Expansion Score in a number of emerging markets. The scoring process considers a variety of non-traditional data sources, including utility bills and property records, to develop a score. To date, FICO’s Expansion Score is in use in several emerging markets, including Russia, Mexico and South Africa.

Similarly, FICO has adjusted the weight of medical repayments under the ninth version of its score in the US. “Our scoring process is constantly evolving. In August, we announced that we are revising the way we treat paid and unpaid medical costs. With FICO Score 9, we will give less weight to unpaid medical bills and eliminate any negative scoring from accounts that have ultimately been settled,” says Ms Kramers-Dove.

Meanwhile, UK credit score firm Experian has recently introduced rent payment data to its scoring process with the creation of the Rental Exchange, which acts as a secure database that houses rental payment information, enabling those with no credit history to establish a credit score based on their rental payments.

Despite their perceived weaknesses in the areas of financial inclusion, conventional credit scoring techniques, and the bureaus that support them, are characterised by high degrees of transparency. For the consumer, this means they are able to see the criteria used to determine their credit score. It also means they are given the opportunity to improve their score against these objective criteria.

“The public perception of credit scoring has changed significantly in the past few years. In the late 1980s and early 1990s it was considered a mysterious activity but now there is so much more transparency; consumers want to know what their credit score is. There is a whole business around actually obtaining credit scores and improving credit scores,” says Chris Curtis, head of analytical solutions at Experian.

Consumer challenge

The same cannot be said for unconventional data scoring methods. Given that thousands of data points are typically used, it is very difficult to determine why a particular constellation of these points may equate to a negative score. Communicating this to the consumer is even more challenging.

“New scoring technologies are relatively under-regulated, which allows for greater freedom in terms of the levels of transparency. In addition, consumers are probably not yet ready to understand credit scores that are derived from big data. In the world of big data, it is sometimes very difficult to find the exact causality for a decision,” says BDS’s Mr Kert.

Moreover, as the algorithms used to evaluate this data are typically the prized intellectual property of these firms, it seems unlikely they will reveal their processes. For now, this pits the newer, unconventional scoring systems against the older, established architecture in terms that loosely equate to financial inclusion versus transparency. While these big data algorithms substantially accelerate the financial inclusion agenda, they do so in a way that is less transparent than traditional scoring methods.

“It’s not just about predictive power. Does everyone want their social media data to play a part in credit decisions? Are you talking about a small segment of the population? Is that fair? Do enough people have social media accounts with scorable data, or is your sample biased toward people who are young and tech savvy? Can a consumer figure out how their data is scored and can they change it?” says FICO’s Ms Kramers-Dove.

Power to transform

The issue of transparency and privacy versus the benefits of financial inclusion is pressing, particularly in developed markets. Nevertheless, for the emerging middle class in markets that lack comprehensive consumer data, the use of unconventional data could be transformative.

“The idea that there is this privacy versus non-privacy world view is completely the wrong narrative," says Lenddo's Mr Stewart. "The real narrative is that most people in the world can't get credit. In the developed world, where credit is easily available, we tend to forget this. So it is very easy to lose track of the real issue: empowering the global emerging middle class through access to credit.” 

This point is particularly compelling as, to some extent, the economic and social development of groups of people who are difficult to score through conventional means is being stifled. “Credit bureaus cover less than 25% of the world population. Of this number, probably about 60% to 70% of people are covered with outdated information. That leaves a large number of people who are both unserved and unservable for banks and most other lenders,” says Mr Griemens from Kreditech.

If this gap in both emerging and developed markets can be filled by innovative start-ups, making use of unconventional data sources, it seems the benefits substantially outweigh questions over transparency. Crucially, regulators in these emerging markets are keen to engage with these new players.

“Regulators, particularly in emerging markets, have welcomed our approach as an effective way of reaching out to a populace which is largely underbanked. You’re probably going to see access to financial services in China and Indonesia surpass places such as the US or Europe because they’re innovating faster.” says Mr Stewart.

While the use of unconventional data to develop credit scores will surely grow, it will be unlikely to replace or diminish the role of conventional scores. Rather, the use of unconventional data seems to be emerging as a complementary trend and one that will vastly accelerate the financial inclusion agenda in the long term. With conventional and unconventional scores operating together, consumers in both developed and emerging markets stand to benefit. 

PLEASE ENTER YOUR DETAILS TO WATCH THIS VIDEO

All fields are mandatory

The Banker is a service from the Financial Times. The Financial Times Ltd takes your privacy seriously.

Choose how you want us to contact you.

Invites and Offers from The Banker

Receive exclusive personalised event invitations, carefully curated offers and promotions from The Banker



For more information about how we use your data, please refer to our privacy and cookie policies.

Terms and conditions

Join our community

The Banker on Twitter