Open finance opens the door to a new era of financial inclusion, where unprecedented access to data is revolutionizing credit decision-making processes. However, to realize the promise of new and more credit models through AI, there’s a critical need for ethical responsibility.
I recently had the chance to join Scott Zoldi, Chief Analytics Officer from our partner FICO at the World AI Cannes Festival, as well as during a panel at our event Open Views 24, to talk about the role of ethical artificial intelligence in leveraging open finance data to drive credit access responsibly.
After our conversation, I couldn’t shake off the feeling of witnessing a real turning point in our industry where the possibility of creating true financial inclusion through technology goes from promise to reality. Something that resonates deeply with our mission at Belvo, which is no other than to democratize access to financial services across Latin America. Witnessing firsthand how this mission intersects with the power of open finance and ethical AI principles, I felt compelled to share some insights.
Open finance presents unique opportunities to build extremely powerful credit models. It allows us to see and understand how people spend their money with unprecedented levels of detail: by looking at transaction data you can see users’ level of spending on things like travel, groceries, debt, and much more, getting a truly vivid picture of how they manage their finances.
Uri Tintoré, cofounder and co-CEO of Belvo
Potentially, this also opens the door to unprecedented ways to make better credit decisions. But as we explore these new ways, we also face big challenges. First, to build robust models we require the highest levels of data quality. And secondly, how do we ensure that the models that we create make decisions that we can explain and follow ethical principles?
Let’s explore these two big topics: how to build robust credit models, and also ensure that are ethical and explainable.
The challenge of data quality
Building robust credit models requires a comprehensive approach to data management and analytics. Firstly, we need to deal with the intricacies of open finance data, including challenges like missing, duplicated, or inaccurate information. Identifying and rectifying such pitfalls is key to ensuring the reliability of our models.
Then, we need to structure data, normalizing and labeling it. This is when cleaning and categorization come in. At Belvo, we’ve managed to move from 37 available endpoints to sort data into four main groups: owners, accounts, credit card bills, and transactions.
Third, we need to create relevant metrics that can be used to build the models, as dealing with the raw information is too complex. After all these steps are done, that’s when we can start creating models and train them.
A powerful partnership for financial inclusion
This is when the collaboration with FICO becomes relevant. FICO brings unparalleled technical expertise to the table when it comes to developing advanced features and models for credit assessment. For instance, they make it possible to detect flags that group users based on various criteria. For example, they can identify patterns like late payments on credit cards, which are crucial indicators in building accurate credit models.
Furthermore, FICO shares our commitment to equality in financial services. While financial inclusion is often talked about as a goal, for us and FICO, it’s a deep-seated commitment. By collaborating with them, we have the opportunity to develop models that prioritize fairness, transparency, and accountability.
This means not only building ethical and responsible models but also ensuring they are explainable and auditable. In essence, by combining our potential to reach underserved populations in Latin America, with Belvo’s high-quality structured open finance data, and the responsible AI principles embedded in their technology, we can achieve true financial inclusion.
Uri Tintoré, cofounder and co-CEO of Belvo
Ethical, explainable, and auditable AI
But what does it even mean to create models that are ethical and responsible? Let’s break that down.
Ensuring that our models are ethical and responsible is a top priority for both FICO and Belvo. This is no easy task, especially considering that many machine learning models operate as black boxes, making it difficult to understand their decision-making process.
To address this challenge, we can actively restrict our models’ learning processes to prevent the introduction of biased outcomes. A more accurate model that includes non-inclusive biases is not something we would accept. For this, we need to analyze the model and include constraints, such as restricting certain combinations of variables.
Explainability is another key aspect of ethical AI. Humans should be able to understand and explain why a particular decision was made, especially when it comes to something as significant as loan approval. That’s why we work with interpretable neural networks, which allow us to maintain transparency while still ensuring fairness and equality in our models.
This approach enables us to provide clear explanations to individuals about why they were approved or denied a loan, empowering them with knowledge about the decision-making process.
Uri Tintoré, cofounder and co-CEO of Belvo
So, even after we follow all these principles. How do we ensure that our models remain consistent and trustworthy over time? By leveraging blockchain technology, FICO is capable of logging every step of the model development process, from bias detection to explainability enhancements. This ensures that any changes or additions to the model are permanently recorded and cannot be altered. As a result, the development of our models is perfectly auditable.
Creating a real impact
Creating ethical and responsible models requires a greater effort than building simpler models. But the positive results that we are already seeing serve as a source of motivation, as it reminds us of the importance of our mission of democratizing access to financial services for all individuals.
A compelling example of the impact of our efforts can be seen in a recent case study involving a Brazilian lender. By using open finance data-based models, they achieved a sixfold increase in loan approvals while reducing the risk of loss by threefold.
This is just an example of the transformative potential that lies ahead of us through this collaboration to drive real financial inclusion in the region. By empowering individuals and communities with access to fair and transparent financial services, we are not only reshaping the landscape of lending but also contributing to a more equitable and prosperous society.
