• Justin Seek, Credit Analyst
  • March 2, 2020
  • LinkedIn

Innovating with Purpose: Designing Systems for the Bottom of the Pyramid


When we think of basic financial services in many developing countries, we think of how the poor often rely on informal services, such as payday lenders, loan sharks, or even family and friends to finance their needs. Unfortunately, these services are often exceedingly costly or unreliable, or both. The digital revolution over the last 10 years has, to a good degree, alleviated this problem by providing more affordable and dependable financial services in developing regions. From initially providing cash transfers systems, bank accounts and credit, we are now in the era of using Artificial Intelligence (AI) and machine learning (ML) to refine and improve these services. However, even with the advancements in technology, there are potentially large segments of the poor that are still excluded. One option is to continue to delve deeper into technology, however, there may be gaps in the market that technology alone cannot penetrate. Another option is to take a step back and look at alternate systems that are already used by the poor. By innovating around the core purpose of financial services, we may be able to find interventions to address the shortcomings of technology.

If we distill the purpose of saving, we see that it is simply the process of accumulating capital that can be used for investing or for security in the face of shocks. The purpose of loans can be said to be for similar ends, however, unlike savings, there are an additional 2 critical components to be factored in; trust and enforcement. Trust in terms of the borrower having the ability and intention to repay the loan, and enforcement on how a third-party makes sure both lender and borrower adhere to the terms of the deal.

In developed markets, rating agencies, such as Moody’s, S&P and Fair Isaac Corporation (FICO), create a credit rating, which is a standardized method for assigning trust levels among companies and individuals based on their repayment capacity. Risk assessments are made without having to personally know the borrower, thereby enabling money to be distributed at greater scale. However, in developing countries, many of the traditional inputs necessary to create a credit score for individuals and smaller enterprises are simply not available.

The rapid adoption of technology has sought to overcome these barriers and increase the pace of access to financial services for these groups. M-PESA, Africa’s leading mobile finance service provider, has proven that digital innovations can transform the entire financial landscape in developing countries. From originally being a SIM-card based cash transfer product, for urban dwellers to send money home to rural areas, the service has evolved to offer “loans and savings in conjunction with local banks, plus merchant payment services”1. In Kenya alone, over the last decade, the banked population has increased from just 14% in 2006 to 83% in 2018,2 resulting in an “estimated 194,000 households, or 2% of Kenyan households, being lifted out of poverty”.3 The power of this tech innovation is that it has driven people to have increased financial resilience and savings.4

The digital revolution has also democratized credit scoring by combining AI and ML methods with alternative data. This has been a boon for people in developing countries, as very often traditional credit histories are non-existent, but data on social media use, bill payments, e-commerce transactions, and even psychometric data can be used as substitutes for traditional data, thereby opening a whole range of opportunities for accessing financial services, and specifically for loans.

Figure 1: Standard Credit Model


We see established and budding fintechs using aforementioned non-traditional data, to assess the creditworthiness of their borrowers.5 They use models that rely on AI and ML to identify a borrower’s likelihood of defaulting by recognizing behavioral patterns of other borrowers’ with similar profiles. However, those familiar with these tools will know that any model is only as good as its data, and for this to work effectively, one needs a lot of reliable data.

So what about those communities where even this information is unavailable? As of 2018, a Pew Research Center study found that smartphone penetration in places like Indonesia and the Philippines is only around 40% to 50%.6 These stats highlight that assessing creditworthiness remains unfeasible for large portions of the population. In fact, there may be an argument that those who need access to financing the most may be the ones that are not captured using this system.

Before the advancement of AI and ML, there existed, in many places like Cambodia, Vietnam, and India, an age-old system known as Savings Groups. Whether they are called tontines, susus or tandas, the principle of the system is the same,7 every member of the group will put in an equal amount of money to create a lump sum that is distributed to a single member at a set period. For example (figure 2), a group of 10 decide to contribute $20 per month, then each member will receive $200 every 10 months, with the first member receiving a $190 interest free loan.

Figure 2: Saving Group System


For this system to be successful, trust among members is paramount; will each member make their monthly contributions, or cut and run as soon as they have received their payout? Trust, as well as enforcement, is mutually understood between members, who are often well known and familiar with each other. Participants recognize that “[members are] not just random people off the street, but people that you know and trust. It starts in workplaces, among families and friends… they know [the] families, it doesn't just end at the money if there are issues"8.

Grameen Bank was the first to recognize the power of community trust and utilize it with extreme effectiveness. Grameen still lends directly to the individual and earns interest on these payments, but forms ‘customer groups’ of borrowers to “cross-guarantee each other’s loans”9. The incentive for each member to repay is that they retain the right to borrow and borrow more, so long as their customer groups meet the schedules. Comparing Figure 1 to Figure 3 (below), both trust and enforcement are taken on by the groups, as opposed to the law and credit scores.

Figure 3: Grameen Bank Model


Like any system, there are also well-documented flaws with the Grameen Model. For example, the lack of flexibility in repayment and borrowing. Individuals can only borrow once the entire loan has been repaid. In times of unpredictable life-cycle events – such as funerals, weddings, new business opportunities – individuals do not have access to the necessary funds. Although, the bank has changed over time with the addition of new products and services to cope with the shortcomings of the system, there are still questions regarding efficiency and coverage. How widespread can loans be if every borrower must find and form customer groups? Ultimately, this becomes an issue of scale.

There is no doubt that technology has greatly impacted the lives of the world’s poor. M-PESA’s remarkable success for the Kenyan people is just one example of technology’s positive effect on the financial landscape. As AI and ML are now taking center stage and pushing the boundaries of inclusivity, it is timely to take stock of how far we have come in this field. We must recognize that, although we have been successful, there are still areas for us to further innovate. Just as how Grameen Bank developed a solution around a tried and tested practice in traditional communities, it may be possible to leverage technology with traditional methods to address the shortcomings of each system. Most importantly, by focusing on the core intention and purpose of providing financial freedom and security, we hope to collectively find a solution that leaves no one behind.