The art & science of credit scoring for informal traders

By Kim Grimwood, Product Manager

In 2006, Muhammad Yunus was awarded the Nobel Prize for introducing microfinance to the emerging market of Bangladesh in India. He understood the importance of giving women in remote villages access to credit so that they could grow their businesses and improve life for themselves, their families and their communities. 

We know that cashflow is a challenge for retailers who need to stock their shelves before they can sell anything. Unfortunately, many of the informal retailers who have a cumulative impact on Africa’s economy do not qualify for “formal” credit.

Yunus’ work sparked global recognition that informal business owners in emerging markets can really benefit from access to fair credit. Financial service providers are keen to get involved and provide credit offerings to this new market.

But there’s a slight snag. 

How do you get an accurate credit score when there’s very limited information available on the behaviour of these informal retailers - many of whom are unbanked?

This is the type of challenge that we look at solving at Nomanini. So, in early 2020, we set out to develop a process for building a credit scoring algorithm to facilitate the extension of credit to informal retailers.

Fast forward two years, and we’ve delivered on a method and platform for providing credit to the retailer, which is expected to have an up-take rate of around 100 retailers per month. It is also expected, through the use of the credit scoring approach and data, to have fairly favourable repayment rates and ambitious plans to scale.

Before a lender will offer credit to anyone, including an informal retailer, they need to assess the reliability of that person to repay the loan granted to them. A credit score is usually the measure to achieve this, and thus it is essentially a risk assessment: more technically termed a statistical probability of non-default. 

The approach to credit scoring involves a selection of the method to use for risk assessment, together with an “art-form'' combining this with some experience and industry-knowledge.  There are three main risk assessment methods that can used during this process:

1. The simplified “Expert Opinion” method, also called the “Scorecard approach
2. The more complex statistical method, which could use various statistical models to derive the probability, including a Bayesian or some other non-linear model.
3. A hybrid of the two that combines the expert opinion with the rigour of statistical modelling.

Generally, if you want to go to market by launching a credit offering to “feel it out”, the simplified expert opinion is recommended, however it is best to include an iteration on this to continually move to a more accurate model. This is because It’s relatively easy to back-track on your more simplified algorithm as you problem-solve and iterate. It is also easier to use this method when explaining to other stakeholders what you are trying to achieve through this risk assessment.  

It can’t be used indefinitely though, and as your data grows more reliable and voluminous you’ll be in a better position to enrich your scoring methodology with statistical modelling. In this instance, a hybrid model is the next logical step.

Without data, it would be impossible to correctly model, or give output to, the credit scoring process. Data allows us to look at the pattern of trader behaviour in the past and use it to make assumptions about their reliability.  

When quality data is scarce, as it is in informal markets, the trick is to go to market with a simplified model, and to draw any reliable data you can find through the model to help make decisions. Ultimately though, you need to wait patiently for better data (richer in volume, quality and intent). As the retailers use and repay credit over time, you can collect this repayment data to help your iteration in the model. 

Data is, however, available in the informal retail market - usually from suppliers - but it is often that their desire to share volumes is thin.  

“Formal” data - such as those recorded from banks through previous loans taken out - are less applicable in the informal market since the traders are less likely to be banked and usually take out loans informally (with family and friends). As a result, a growing trend is to use other “alternative” ways of gathering a perspective of a potential customer’s behaviour - for example by looking at mobile phone call times or social media activity - but this approach is still in the early stages of usage and quality.

Credit scoring in the informal market requires a special blend of art and science, as well as the ability to learn through the experience.  

We’re able to borrow learnings from the formal credit risk assessment process but, as we have seen, there are specific nuances and challenges in the informal market that can only be overcome as we reach them.

It’s an exciting time to work on credit offerings for the informal market, but we’re careful about ensuring that our solutions are delivered responsibly in the backend, and at the “front end”, we work with our partners to ensure traders understand how loans work and the implications of repayment defaults.

All in all, it’s a rewarding challenge for our team, and one we’re proud to be a part of.

______________________________