SixPoint

‍

Integration of Credit Scoring and Origination: Automation, Human Involvement, and Alternative Data

‍Taimur Sajid – (SixPoint Capital)

To begin, let’s discuss the relationship between credit scoring methodologies and origination. How do these processes interact in your companies? Are they managed separately, or is there human involvement? How do you integrate these aspects in your businesses?

Andrew (Fairplay):

For us, it’s more of a gradient rather than a clear-cut process. On one end, there are cases that pass quickly through an automated process with minimal human interaction. On the other end, there are cases we actively discard. The majority of cases fall into a large middle ground where significant human evaluation is still necessary. Full automation remains challenging at this stage.

Roger (R2):

At R2, we leverage transaction-level data from the platforms we work with. Monthly, we analyze between 1.2 and 1.4 million distinct merchants. Machine learning is crucial for automating this analysis. However, there is often a conflict between origination goals and what credit and risk models suggest. We focus on ensuring that our models have adjustable levers, allowing the risk team to make strategic changes without overhauling the models entirely. While the core process is highly automated, human intervention is necessary for strategic adjustments.

Joe (Albo):

When we launched Delta AI in 2019, our initial focus was on artificial intelligence, but we quickly realized that the real challenge lay in the fundamentals and infrastructure. Initially, AI was more of a branding element. We discovered that understanding customer behavior and transaction data was crucial. Instead of using a third-party bank, we created our own platform where customers could manage all their transactions. This approach provided us with valuable insights and allowed us to develop an internal credit scoring system.

While traditional credit scores are useful, our main advantage comes from the data we collect directly, enabling us to assess risk based on real-time customer activity.

Conrad (Baubap):

In consumer lending, where we operate, we receive over 10,000 requests per day. Automation is essential to manage this scale. We built our underwriting technology using 100% alternative data, which helps us predict payment behavior and generate demand for our product. By not relying on traditional credit bureau scores, we appeal to consumers who have been excluded by conventional financial systems. Using transactional data from their phones and other alternative sources, we provide a more inclusive evaluation.

‍

Addressing the Cold Start Problem with Alternative Data Solutions

Taimur Sajid – (SixPoint Capital)

The variety in responses highlights that there is no one-size-fits-all solution. It depends on the product. Joe (Albo) mentioned overcoming the cold start problem through customer transaction data. Can you elaborate on how you addressed this challenge?

Roger (R2):

At R2, one advantage is receiving at least a year’s worth of transaction-level data once we partner with a platform. Initially, we didn’t have a credit score. To develop a model, we deployed capital, we had to experience some losses, and used manual labeling. Over two to three weeks, a team of us manually reviewed 10,000 merchants, evaluating various attributes and making financing decisions based on expert criteria. This manual process helped us establish a foundation for more scalable models. We then averaged the scores from different reviewers to build models with varied risk appetites. This approach, though manual, provided us with a starting point for model development.

Andrew (Fairplay):

Access to transactional data is crucial for differentiation from traditional banks. We focus on integrating multiple data sources to gain a comprehensive view. Combining data from various sources helps us avoid duplication and improves loan accuracy. Our approach involves creating a series of scores based on data acquisition and adjusting our models to assess risk more effectively.

Joe (Albo):

In Mexico, electronic accounting plays a significant role. The SAT, Mexico’s tax authority, mandates electronic invoicing, which creates a digital record of all transactions. This system provides valuable data, allowing us to assess profitability and cash flow. We also consider tax payment behavior as an indicator of creditworthiness. This alternative data, combined with our internal systems, helps us evaluate risk more accurately.

Conrad (Baubap):

In both B2B and B2C sectors, overcoming the cold start problem requires a broad range of data and a rapid feedback loop. For instance, our 30-day microloans allow us to iterate quickly and refine our underwriting skills. Collecting as much data as possible and using iterative feedback helps us improve our technology and adapt to changes effectively.

Andrew (Fairplay):

Emerging markets, including Latin America, are increasingly standardizing digital tax returns and transactions. This trend presents opportunities for improved credit scoring.

However, challenges remain, such as the quality of services provided by tax authorities and the need for comprehensive data collection. Companies must decide whether to rely on third-party data or develop their own data acquisition capabilities.

‍

Exploring the Role of Alternative Data in Credit Scoring

Taimur Sajid – (SixPoint Capital)

Our discussion has touched on the use of alternative data extensively. To delve deeper, can we explore the types of alternative data you use and whether it serves as a full replacement for traditional credit metrics or functions more as a complement?

Joe (Albo):

The role of alternative data often varies depending on the context. In our case, alternative data acts as a complementary tool that validates our internal models rather than completely replacing traditional credit metrics. Early on, in 2019, it was more efficient for us to develop our own systems for data collection rather than integrating with existing systems. We focused on creating a robust infrastructure and digital models to quickly access and act on data. Traditional credit scores were initially less relevant, but as we evolved, incorporating traditional metrics helped us refine our approach. The emphasis on digital data collection and real-time processing was crucial for our operational efficiency and model accuracy.

Conrad (Baubap):

In the consumer lending sector, there is a high demand for rapid decision-making. To meet this demand at scale, we minimize reliance on third-party integrations and prefer handling as much data in-house as possible. While traditional financial data from credit bureaus can complement our alternative data, we have found that alternative data provides a richer and more comprehensive view of an individual's financial behavior. This includes data from calendar events, app installations, and social interactions, which offer deeper insights into an individual's habits and preferences, giving us a competitive edge in underwriting.

Roger (R2):

At R2, we rely entirely on alternative data, such as transaction-level data from platforms. This data allows us to predict future income streams and assess business performance with high accuracy. Additionally, we analyze digital footprints, such as social media activity, to gauge business activity and credibility. However, structuring and scaling this data can be challenging. Traditional credit scores, used more recently in our processes, serve primarily as a validation tool rather than a primary data source. We consider indicators like frequent credit score checks or excessive outstanding debt as red flags, rather than relying solely on credit scores.

Andrew (Fairplay):

Alternative data can sometimes present gaps, and it's essential to address these gaps effectively. Banks, despite their traditional methods, use useful data that should not be overlooked. We focus on understanding both the business's operational data and the individuals behind it. For example, analyzing seasonal business performance can reveal potential cash flow issues. Our approach involves using alternative data to enhance our understanding of both the business and its operators, with ongoing iterations to improve our models. ‍

‍

Enhancing Fairness and Inclusivity in Credit Modeling

Taimur Sajid – (SixPoint Capital)

This discussion highlights the balance between alternative and traditional data. Moving forward, how do you see your credit modeling efforts contributing to fairness and inclusivity, as opposed to traditional methods that might be more restrictive?

Conrad (Baubap):

We are committed to excluding demographic data from our models, such as place of birth or gender, which are often used by traditional banks but do not reflect an individual's true financial behavior.

By focusing on transactional data that individuals generate through their behavior, we aim to provide a fairer assessment. This approach not only aligns with our business needs but also promotes inclusivity by avoiding biases inherent in demographic data.

Roger (R2):

At R2, our analysis of partner ecosystems is pseudonymous, focusing on business performance rather than personal data. This approach helps eliminate biases and allows us to tailor offers based on actual business capabilities. While demographic data could introduce biases, our priority is to develop unbiased underwriting models. We use demographic information cautiously, aware that once incorporated, it is difficult to remove without disrupting the model.

Joe (Albo):

Our initial challenge was accessing credit history for new businesses, which led us to create our own credit card product. This allowed us to gather performance data and identify promising businesses that lacked traditional credit histories. We focused on leveraging alternative data to provide financial services to those who were previously excluded, thereby promoting fairness and supporting small businesses in Mexico. ‍

‍

Best Practices and Strategic Insights in Risk Management and Model Development

Taimur Sajid – (SixPoint Capital)

This discussion underscores the importance of overcoming traditional limitations to foster inclusivity. As we wrap up, what advice would you offer to platforms based on your experiences with risk management and model development?

Andrew (Fairplay):

It's crucial to define what your risk model aims to achieve. Whether it's assessing a borrower’s ability to repay or their intent to repay, understanding these aspects will guide your model development. Risk management requires ongoing refinement and iteration to address various factors, including unexpected events and fraud.

Roger (R2):

Consider using your product's application funnel to identify suspicious or fraudulent behavior. For instance, offering different conditions and monitoring responses can help detect fraud.

This approach allows you to leverage your system to enhance risk modeling and minimize adverse behavior.

Conrad (Baubap):

Startups should approach risk management with a mindset of taking calculated risks. Experiment with small, controlled bets to test assumptions and refine models. This approach not only improves your risk management but also enhances overall business performance.

Joe (Albo):

We opted not to innovate in risk management but rather to adopt traditional methods with modern enhancements. Speed of execution became a key differentiator for us. By providing faster responses than traditional banks, we gained a competitive edge and built a strong customer base. Fast, reliable service combined with a traditional risk model allowed us to succeed in a challenging market.

Taimur Sajid – (SixPoint Capital)