Data for Good

Connecting At Least One Million People to the Formal Economy by 2020: Can Machine Learning Be Used to Solve Financial Inclusion?

MasterCard Taps the Creative Destruction Lab and MBAs from Canada’s Leading Management School to Tackle a Global and Multifaceted Problem

Connecting At Least One Million People to the Formal Economy by 2020: Can Machine Learning Be Used to Solve Financial Inclusion?

July 15, 2015

The lack of financial inclusion is a problem that plagues millions of people in the developing world. Credit assessment is expensive and cumbersome, and the default position is to simply exclude those with a higher risk profile or an insufficient banking data footprint from lending altogether. The social and economic ramifications of such an approach are wide reaching. Always on the lookout for innovative technology-based solutions, the MasterCard Center for Inclusive Growth has partnered with Rotman School of Management’s business accelerator, the Creative Destruction Lab, as well the school’s top MBA students.

Students in Professor Ajay Agrawal’s strategy class were tasked with exploring machine learning and its potential to address financial inclusion by creating innovative ways to gather and interpret data. Pleased to put textbooks and case studies aside, the students delved into this crucial social issue and presented their findings before MasterCard top brass.

The Problem:

Microfinancing, the saving grace for countless small, often family-run businesses in the developing world, is fundamentally incompatible with traditional credit-rating approaches. The cost of reviewing a loan request is high in comparison to the size of investment being made, and more importantly there is limited data available about those typically seeking micro-loans. The result is no credit, or small loans at a higher interest rate. In fact, current research suggests that the microfinance industry still has not served 90% of its potential clients (Álvaro R. Arregui). For a micro-entrepreneur, access to fair and affordable credit could mean the ability to break the cycle of poverty, with positive ripple effects throughout the community.

The Solution:

Machine-learning techniques can accelerate the lending process by discovering relationships between input and output variables. The algorithms can assess multiple factors that may impact the likelihood of loan repayment. Some, like family income, education and age might be obvious, while others could be less apparent. When deployed with appropriate human oversight, machine learning could help financial institutions glean better insights from vast amounts of data that is more immediately relevant to populations in need of microcredit.

When micro-businesses prosper, other entities do as well: NGOs and microfinancing institutions enjoy a growing customer base; agencies see increased operations; telecommunication companies witness higher bandwidth utilization due to increased number of transactions; and local governments receive increased tax revenues, enabling them to invest in infrastructure and services.

Rotman’s MBA teams suggested several strategies for using machine learning to tackle this complex problem. Here are the top three:

Build and uphold an ecosystem of partners that use digital data to reduce the cost of credit assessments of micro- entrepreneurs. Developing economies like Nigeria and Egypt, are already building multi-use ID Programs that could act as a treasure trove of data. Machine learning can provide individuals with a quick, inexpensive assessment of their creditworthiness using multiple data sources associated with their national ID cards. This dataset will include all available financial data as well as non-traditional data sources such as e-payment history and cell-phone usage. A network of partners, both existing and new, will allow rapid deployment of market-ready solutions while putting the promise of machine learning to the test.

Celine Guo, Venkat Swaminathan, Sarah Kotur, Dana Bastaldo, Sheetal Persaud, John Forsyth

IMG_8461 (1)
Deploy machine learning to facilitate micro-lending in agriculture, a sector in which 30% of the world’s population is engaged. Bangladesh’s rice farmers can act as a perfect test for the applicability of this strategy. While nearly half of the entire country’s population is engaged in rice farming, less than 10% of farmers are connected to the formal economy and almost all live under the poverty line of $1.25 per day. Moreover, the sector is expected to grow by 36% over the next five years. As machine learning is used to learn more about the farmers through existing infrastructures like telecommunications, as well as analysis of data on poverty levels, region, crop, and yield, lenders could more accurately assess farmers’ ability to repay and refinance. The credit systems/scores arrived at in Bangladesh could then be adjusted to reflect the conditions of farmers in other countries based on their own distinct attributes.

Amir Ramezanpour, Ian Mullin, Balakumar Nair, Ray Amin, Eklavya Bhambri, Ben Bernardo

Deploy machine learning to help Financial Product Providers (FPPs) better meet customer needs. Machine learning should be used to develop a recommendation engine that will identify, with a higher degree of accuracy, the financial products that are most suitable for potential customers. Data from all FPPs will be collected, including data on the historical performance of financial products, and a machine-learning algorithm will then identify patterns and relevant market insights to be shared with FPPs, resulting in better matches between customer and product. As FPPs learn more about consumers, this strategy should help identify important pockets of demand, reduce costs, and increase the reach of financial services, while also improving product quality. Through ongoing feedback from FPPs’ data inputs, the algorithm can be refined to provide improved pattern detections and product recommendations, resulting in even greater customization.

Jennifer Amaral, Meghan Gervais, Katie Pearson, Thomas Dyer, Jingqui Guo, Ahmad Ajmal

The promise of machine learning to transform the financial services industry is real even in the west, but its potential is far greater in developing economies, where traditional measures of creditworthiness do not apply as neatly. Sophisticated algorithms can draw relevant insights from new sources of data or find previously unidentified connections between existing data sets.

A word of caution is in order: the results are only as good as the underlying data, and algorithms, no matter how sophisticated, lack the intuition of humans. However, when intuition is augmented by the power of novel data-driven approaches, our ability to tackle complex economic and social challenges – like financial inclusion – is greatly enhanced. And finding a solution moves from the realm of imagination to imminent reality.


Creative Destruction Lab is a business accelerator for technology startups based at the Rotman School of Management, University of Toronto. The CDL was founded four years ago by Professor Agrawal to help transform doctoral research projects into viable and scalable businesses. In addition to its regular cohort, the Creative Destruction Lab will be adding a special stream dedicated to machine learning this fall. MasterCard has been a valuable partner of the CDL and its ventures and is excited about the promise of machine learning in the financial services industry and beyond.