JUMO’s Chief Information Officer, Paul Whelpton, had the opportunity to speak at a virtual AI in Africa conference. The topic he presented on is of critical importance. Paul explored how credit modelling based on machine learning — a branch of artificial intelligence — can reduce bias to make much better decisions.
“So far traditional financial players have not been able to find a viable way to get credit into the hands of many financially excluded groups. Billions of people around the world still lack access to financial services. To fix this, we need to connect entrepreneurs and ordinary people to the financial tools everyone needs to grow and prosper. It’s a supply issue, not a demand issue.
AI credit modelling can play an important role in building a more inclusive economy, as it makes it possible for anyone with a phone and mobile wallet to make transactions and develop a financial identity.
Through AI, the relative machine learning credit models can be tailored to enhance and eventually replace outdated credit thinking that has left ‘others’ out over the years.
In my presentation at the conference, I explored one of the biggest issues when it comes to financial inclusivity: biased credit scoring. To illustrate the bias that can exist with credit scoring, we looked at the example of a small-time farmer in Pakistan. With a drought predicted in the country, there is an expectation that her crop yield will be significantly smaller this year. For a credit model, the risk goes up because the probability of default is higher.
Under traditional credit thinking, you have a higher probability of default and there is a price for risk (e.g., in this case, the price increases from 10 percent to 17 percent). Also, under traditional credit thinking, facility size is reduced and the annual facility fee is increased. Under this outcome, when it doesn’t rain and the crop yields less, interest expenses are high and there is reduced income, leading to default. A self fulfilled prophecy that further validates a flawed model and credit philosophy.
However, consider an alternative approach that reduces bias and creates a better result. Under this thinking, price is reduced from 10 percent to 5 percent, credit is extended to get to the next harvest season, and there are no fees this year. With this type of thinking, the farmer has lower expenses, can make payments through the season, and does not default. By taking a new approach to offering credit, everyone can benefit. In this example we use the same data and modelling techniques with a different strategy to assist a “higher” risk customer through a difficult time.
In order to improve financial inclusivity, we need to work to eliminate biases. There are a number of ways in which we can do this. For instance, there is transactional bias. Is a thin file (a banking term for customers with limited or no credit record) really a measure of risk? On top of that there is gender bias, as well as behavioural bias. We need to understand that not all customers are the same, and to help with this issue personalisation is key. These are just some of the biases that have kept people out of the economy, as traditional credit thinking has created a less-than-welcoming atmosphere.
Ultimately, bias plays a sizeable role in creating a less inclusive economy. By using AI credit modelling, we can work toward eliminating bias and creating an economy that includes far more people who have been previously left out.”
More about machine learning
View the recording of Paul’s presentation at the AI in Africa Conference