By Jim Savage, Data Science Lead at Lendable Inc.
Securing debt financing to grow a medium-sized lending business in East Africa is not easy or cheap. Interest rates are high—often north of 25% a year in local currency terms—deals take a long time, and pre-financing requirements are onerous.
A few factors contribute to these problems. East Africa’s credit bureaus are a recent phenomenon and still underutilised and underdeveloped (though this is improving) and so lenders often don’t have a good idea of how likely a borrower is to repay on its debt. Resolving this uncertainty requires thorough investigation into the firm borrowing (so-called due diligence) by lenders, and these due diligence costs get baked into higher interest rates. Similarly, banks—who have the best information on their customers’ ability to repay on debt—face low deposit rates, necessitating costly external financing.
New technologies are helping to bring these costs down. One such solution, which I have helped build, is the Lendable Risk Engine.
One of the many beautiful aspects of East Africa as a business destination has been the phenomenal uptake of mobile payments. M-Pesa and the like have resulted in high-quality, externally verifiable data on the actual cashflows flowing into mid-sized ‘alternative lender’ businesses (the sorts of non-bank lending businesses we mostly work with). When an end-customer borrows from an alternative lender, they typically repay their loan with mobile payments. To handle these payments, alternative lenders have had to invest in sophisticated management information systems. These systems contain information that can result in swifter, fairer financing—all it takes is the right approach to analyse the information they contain.
The Lendable Risk Engine is a set of systems that can plug into alternative lenders’ databases to help make sense of all the data. The system first converts the alternative lenders’ data into Lendable’s proprietary format. Because all our alternative lenders’ data is then in the same format, it makes it straightforward to deploy our analysis libraries and models, illuminating risks and opportunities in a particular alternative lender’s portfolio.
The first set of analyses answer “how does the business currently work?” sort of questions. These are the questions that we are very commonly asked by clients and investors, and which used to take a long time to answer. Today this analysis is all automated; what once took weeks now takes minutes.
The second set of analyses ask questions like “how do we think the business will perform over the next 2 years?” To answer these questions, we’ve built a prediction platform that makes use of cutting-edge Bayesian machine learning techniques. It generates hundreds of predictions for every loan on our clients’ books for every month into the relevant future. This allows us to gauge the likely future loss rates (and uncertainty about loss rates) in a debt portfolio. Again, this is almost completely automatic. What used to take a month or two now takes a day.
The Lendable Risk Engine helps highlight risks in a deal, and minimizes the information gap between lenders and borrowers—reducing borrowing costs. But we’re not done yet. We want to make it as easy as possible for alternative lenders to make use of our tech. A big part of that is to minimize the amount of data that is required to make high-quality financing decisions, reducing the amount of time it will take to “plug in” the risk engine to alternative lenders’ systems.
In many ways, mobile money (and the information it generates) is allowing finance in East Africa to leapfrog the west. Firms like Lendable, unencumbered by legacy tech, are able to make use of this information. That means fairer, faster debt financing for borrowers, and ultimately a more mature financial system in East Africa.
Jim Savage
Data Science Lead
Lendable Inc.
Jim is a data scientist, econometrician, and empirical finance geek. Most recently, he worked at Australia’s Grattan Institute, where he worked on policy design to help improve efficiency in Australia’s retirement income savings system. Before that, he worked in the Australian Treasury’s Macroeconometric Modelling Unit, working extensively on carbon pricing in Australia. In 2014, he was a fellow at the Eric and Wendy Schmidt Data Science for Social Good summer fellowship, at the University of Chicago.
