Data analytics is good for our economy, consumers, and good for society. Today 4.5 billion people worldwide remain financially excluded because they have little or no credit history (Source: Singapore Financial Inclusion Institute http://financialinclusioninstitute.com/projects/the-better-business-of-clean-energy/).
In an interview with CredoLab’s chief data scientist, Dymtro Kurov shares how banks and consumer lenders can leverage data analytics for greater business and social impact.
How is CredoLab different from the industry?
We want to offer a seamless, transparent experience for our customers. We make sure we get their permission before accessing carefully chosen data sources on their mobile phones.
There are some consumer lenders or even third party apps that require full access to a customer’s online social networks and email accounts. This is not something we practise as it seems quite intrusive to ask for 100% access to customer data.
We are also pretty unique in our end-to-end integrated application process, where we now have an almost instantaneous time-to-yes. In just a matter of seconds, an applicant can know if their loan has been approved or declined. In comparison, some other alternative scoring providers require you to fill in long, detailed application forms that demand too much time and information.
We are able to offer this almost instant time-to-yes using proprietary machine learning algorithms which support the entire credit scoring process.
What alternative data sources do you use to underwrite your customers?
We source from over 50, 000 data points on a customer’s mobile phone. This ranges from call behaviours, messages, phone contacts, downloaded apps and image files, to emails, and websites and locations visited.
Can you share a hypothetical example of a customer who fits the profile of a “reliable borrower” based on their smartphone behaviour?
There is no “template” profile of a reliable borrower. It differs among borrowers. However, we can share some common behaviours of a reliable borrower:
· low number of long-duration incoming calls during working hours
· high ratio of missed incoming calls after successful outgoing calls
· minimal amount of time spent on internet surfing at night
The above list is just a very simple example of what a good customer profile looks like. There are of course more complex behavioural combinations which we analyse.
What challenges do you face when collecting and analysing the data?
We initially faced some difficulty in our data collection as we were just starting out with a very small client base. Now we have over 44 signed clients and more than 2 million data sets uploaded. Having access to all this data has definitely helped in our analysis and further improvement of CredoApp.
Another challenge we faced was in the designing of digital features to capture data. At the beginning it was a challenge to design these features, but now we have developed an almost fully automated process to obtain features from raw data, which then allows us to develop over tens and thousands of digital features and perform a 360 analysis of our clients.
How does CredoLab ensure the protection/privacy of customer data?
We apply a 100% anonymous approach to data collection. We are collecting only anonymous data and that is very important because our clients (banks and lenders) take data privacy and customer confidentiality very seriously. This means that CredoLab does not know the customer’s name, address, email address, or phone number. What we are collecting is raw data available on the customer’s mobile device.
This data could be related to calls, SMS, applications installed, calendar events or mails. In general, customers understand why we collect the data and are comfortable so long as it is done anonymously.
How do you see your role as a data scientist in contributing to financial inclusion in emerging markets?
As a data scientist, I build efficient algorithm models to benefit both the banks/consumer lenders and their borrowers. It’s a win-win for both. Good customers get their loans, and lenders are protected from fraudsters and defaulters. This allows lenders to capture the potential of previously unbanked populations and incur lower costs of risk from late payments or defaulted loans. Eventually, this more accurate customer profile leads to lower interest rates and fees for good borrowers.
Also I do believe this helps to break the cycle of a "thin credit file” or vicious cycle of poverty, where a borrower has no credit history so he can’t get a loan, and at the same time, he can’t build his credit history because he can’t get a loan or access other financial instruments.
Ultimately we want to transform the way people in the developing world secure access to financing and the way we are doing this at CredoLab is through collecting and analysing a wide array of alternative data to accurately determine a customer’s credit score.
About Dymtro Kurov (“Dima”)
Dima earned his Master of Science in Applied Systems Analysis from the National Technical University of Ukraine (Kyiv Polytechnic Institute). He has strong experience in data science for consumer finance risk management, including working with a retail bank in Ukraine and Zoral Labs, one of the largest AI and machine learning labs in Europe. He serves as the chief data scientist for CredoLab.