Alternative Credit Scoring Models For Smallholder Farmers

Sharifa Juma digs terraces to stop soil erosion in Tanzania. Photo: Georgina Smith / CIAT

According to the latest estimates by the World Bank, two billion people still lack access to basic financial services. The main barriers of lending to the unbanked have always been high costs of outreach anddelivery, and lack of traditional data for risk assessment. These barriers exacerbate when dealing with smallholder farmers (SHFs), making them the most likely to be underserved. In fact, despite agriculture remains the main economic activity and employs the majority of the people in most low income countries, only a smaller share of financial institutions portfolio is invested in it, leaving a $ 200 billion in unmet financing, as estimated by the Initiative for Smallholder Finance. 

Indeed, digital innovation has gone a long way in lowering such barriers, driving financial inclusion: from mobile money enabling faster and cheaper transactions, to microfinance institutions (MFIs) developing applications powered by tablets, to reduce their operational costs, thus increasing efficiency.

In today’s blog post, I want to focus in particular on the challenges that financial institutions experience in assessing risk of SHFs and how data analytics companies can address them.

Agricultural production cash flows are inherently more difficult to estimate accurately: a myriad of factors have to be taken into account to quantify cost of inputs, revenues from selling the produce, and, most importantly, harvest yields. One must consider land size, crop seasonality, use of fertilizer, type of seeds, quality of soil, etc. Now imagine a MFI loan officer going out into the field asking this endless list of question to the potential SHF borrower, possibly taking notes on paper, in an attempt to gather all the necessary information. Not surprisingly, the SHF might start to feel overwhelmed and intimidated. At the same time, such activity is extremely time consuming for both. Once back into the office, the loan officer would have to perform complex calculations, based on the information collected, to generate the actual predicted cash flow. Indeed, such task, besides being time consuming requires an in depth sector expertise and complex analytical capacity that MFIs often lack in-house. Moreover, they cannot rely on any other piece of information traditionally used to assess risk, such as collateral availability and credit history, because the clients they are trying to serve lack assets to pledge and are often interacting for the first time with a financial institution. No wonder, there is still a huge unmet demand for credit by SHFs: serving them on the basis oftraditional methodologies entails transactions costs and risks that are bound to be too high. 

This is where and why data analytics providers come into play. They have the capacity to employ AI and machine learning techniques to process alternative data, collated from different sources to bridge this information gap: from mobile phone usage and social media data, to build the farmer social profile, to agronomic and satellite data to assess the farm’s productivity, as well as satellite imagery to predict poverty levels. In order to make better informed and data driven lending decisions, financial service providers must leverage on such expertise. However, while there are many initiatives using this data to improve agricultural production and provide farmers with useful information such as market prices and weather forecast, there are not yet many financial service providers linking geodata and ICTs with their lending activity.

But what does it take to build reliable algorithms and make this work for financial institutions? The process is indeed complex, long and requires coordinated efforts by different stakeholders. Data originating from different streams and feeding into the algorithm must be cross-checked and validated: data “from the sky” (i.e. weather data, satellite data and imagery) must be checked against data “from the ground”, collected on farm premises, to verify the ability of predicting its productivity. Once again, estimated farm productivity and farmers social profiles must be checked against actual loan repayment performance. And such process must be a continuum, self-adjusting and updating through machine learning. In such a scenario, financial institutions must be willing to take a good load of risk upfront: one could run the risk of being deceived by ready to use scoring models developed by mushrooming start-ups, whereas they are still work in progress that need a joint effort to be refined and finally accurately predict risk.

Such an exercise is not easy and surely not cheap: from the perspective of financial institutions, increasedrisk capital availability or loan guarantee funds to pilot and develop alternative credit scoring models would certainly go a long way. Mitigating financial institutions’ risks would in turn magnify the pace and scale at which these new models can be tested. SHFs are crucial in supplying the world with sufficient food: there will be 9 billion people to feed by 2050 and we must move fast to unlock SHFs access to financial services, to ensure they have the necessary capital to improve their productivity.