Machine learning isn’t just a niche tool anymore. It drives decisions that affect billions of dollars and millions of lives. No matter whether you’re approving a loan, forecasting global demand, or suggesting the right seller strategy, the models behind those choices need to be accurate, fair and explainable.
That’s where Hatim Kagalwala comes in. Besides being a data scientist, he’s an expert at building machine learning solutions where margins for error are razor-thin. As the first data scientist at Credibility Capital, he developed credit models from scratch. At American Express, his forecasts shaped the regulatory strategy. And at Amazon, his work on causal inference and credit scoring helped generate over $500M in incremental revenue.
In this interview, Hatim shares how he approaches high-stakes modeling, what it takes to build trust in data-scarce environments, and why transparency matters more than ever when your model is business.
Hatim, you’ve contributed to building Amazon’s credit risk capabilities for customers in emerging markets. What motivated the company to expand in this direction, and how did you get involved in the work?
In today’s consumer economy, flexible and innovative payment options aren’t a luxury—they’re an expectation. Major retailers like Best Buy, Macy’s, and Target offer co-branded credit cards to build loyalty and boost purchasing power. Even small purchases—like ordering food—can now be made using Buy Now, Pay Later services from companies like Klarna. These changes reflect a broader shift: to stay competitive and accessible, especially in fast-growing economies, retailers need financial products that lower friction and build customer trust.
At Amazon, there was a clear focus on expanding digital commerce in emerging markets. But in many of these regions, traditional credit systems are either limited or unreliable. That created an opportunity to design custom credit risk models that could safely extend purchasing power,despite the lack of conventional financial data.
I joined the effort based on my previous experience in credit and fraud risk at American Express. Working with cross-functional teams, I helped build machine learning models tailored to the specific challenges of these markets. It was a chance to create something from the ground up, combining applied science with direct business impact. For example, in one of the emerging markets, we used alternative data sources like mobile top-up behavior and delivery reliability to estimate creditworthiness.
What were some of the technical and operational challenges in developing credit risk models for markets with limited access to traditional financial data?
One of the biggest challenges in emerging markets is the lack of reliable data. Traditional credit bureaus either do not exist or have very limited coverage, which makes it hard to assess creditworthiness using conventional methods. We had to find creative solutions—looking for alternative signals that could help us make responsible lending decisions while managing risk.
Another major challenge was behavioural. Consumers in these markets often respond very differently to credit products than consumers in more developed economies. Financial literacy may vary, and there can be cultural nuances in how credit is perceived and used. For example, we explored using behavioral shopping data and mobile device metadata as proxies for credit behavior, because standard financial histories often don’t reflect how consumers actually interact with credit.
In addition, there is often a legacy of predatory lending practices in some regions, which leads to deep mistrust of any new credit offering. So beyond the modelling and data work, there was a significant emphasis on building trust and designing products that were transparent, fair, and aligned with local needs. For some of these projects, especially where we had to estimate long-term business outcomes, we started using causal machine learning. As a result, we could see what actually moved the needle, not just what might happen.
You leveraged advanced machine learning techniques to rank customers by credit risk. What guided your modeling approach, and how did you measure its effectiveness?
When building credit models in data-scarce environments, we needed a solution that went beyond binary classification to provide a relative sense of creditworthiness across customers. Ranking approaches offered a more flexible and nuanced way to prioritize decisions, especially in cases where ground truth labels were limited or noisy.
We explored various machine learning techniques that could effectively learn patterns from alternative data while maintaining interpretability and fairness. The focus was on building models that would generalize well across regions and customer segments, without relying heavily on traditional credit indicators.
To evaluate performance, we used a combination of ranking-specific metrics and business-aligned outcomes. This included how well the model distinguished higher-risk from lower-risk customers, as well as how its predictions translated into repayment behavior and default rates. We also tracked fairness and explainability metrics to ensure the models aligned with broader principles of responsible AI and fair lending.
You’ve mentioned the application of causal machine learning. Could you tell us more about that?
Causal machine learning is fundamentally different from traditional predictive modeling. In traditional machine learning, we typically have ground truth outcomes and evaluate model performance based on how accurately it predicts those outcomes. But in causal inference, we estimate what would have happened if a certain action hadn’t taken place—working with counterfactuals.
For example, in a medical setting, this could mean estimating how a patient would have responded if they hadn’t received treatment. In a business context, it often involves measuring the true impact of a program or intervention—like a marketing campaign or policy change—by comparing actual outcomes to estimated outcomes in a counterfactual scenario.
This is still an emerging field, but it’s gaining momentum across industries. Major companies like Amazon, Google, and Netflix are investing heavily in causal methods because they help drive better decision-making. Instead of just predicting what is likely to happen, causal models help prioritize what should be done to achieve the best outcomes.
At Amazon, we’ve applied these techniques to evaluate the impact—both financial and behavioral—of key programs, helping leaders focus on the most effective initiatives when faced with competing priorities. One such model is the Potential Sales Lift, which uses causal inference to quantify how seller revenue may vary under different listing actions.
At American Express, you were involved in capital stress testing as part of the Comprehensive Capital Analysis and Review (CCAR). What was your contribution to this process, and why was it important for the company’s financial stability?
At American Express, I worked on statistical models used in the Comprehensive Capital Analysis and Review (CCAR) process—a critical regulatory exercise led by the Federal Reserve. My specific contribution involved developing forecasting models for credit card spending volumes and paydown rates over a 13-quarter horizon, based on macroeconomic scenarios prescribed by the Fed. These forecasts served as foundational inputs for the company’s projected P&L and capital planning.
The primary goal of this work was to ensure that American Express maintained sufficient capital buffers to withstand severe economic downturns. This process is not only vital for internal risk management but also serves as a public signal of the company’s financial resilience. Failing to meet the Fed’s standards can result in significant regulatory penalties and restrictions on shareholder distributions.
The importance of robust capital planning has become even more evident in recent years. For example, the failure of Silicon Valley Bank in 2023 highlighted how a lack of stress-tested planning for liquidity and interest rate risk can lead to a rapid loss of confidence and eventual collapse. Institutions that take regulatory stress testing seriously are better equipped to navigate uncertainty and maintain trust with regulators, investors, and customers alike.
Your career spans both fintech and e-commerce. What differences do you see in how data science is applied across these two industries?
Both fintech and e-commerce rely heavily on data science, but the stakes and goals of those applications can be quite different.
In fintech, especially in areas like credit risk and fraud detection, the stakes are incredibly high. Decisions often have direct financial consequences for individuals, such as whether someone is approved for a loan or how much credit they receive. These decisions need to be explainable, fair, and compliant with strict regulatory frameworks. During my time at American Express, for example, I was acutely aware that even a small modeling error could trigger regulatory scrutiny or negatively impact a customer’s financial well-being. That instilled in me a deep sense of responsibility toward model governance, fairness, and transparency.
In contrast, the e-commerce space, while still data-driven and complex, tends to allow for greater experimentation. At Amazon, I’ve worked on a wide range of machine learning initiatives—from causal inference to credit models for customers with limited or no credit history. Many of these projects allow for rapid testing and iteration, enabling us to experiment, learn, and optimize for long-term outcomes. While the models still need to be robust and responsible, the tolerance for failure during early development is generally higher, especially when testing new features or recommendation strategies.
That said, my experience in both domains has shown me how transferable data science skills are across industries. While the objectives may differ—risk mitigation in fintech versus customer experience in e-commerce—the underlying principles of responsible modeling, experimentation, and impact measurement remain the same. This crossover has allowed me to apply a risk-aware mindset in fast-moving environments and bring an experimentation-driven approach to more regulated settings.
Many of your projects have involved high business stakes and significant uncertainty. How do you manage pressure when millions of dollars—or millions of users—depend on the accuracy of your models?
When working on projects that impact millions of dollars or millions of users, I manage the pressure by grounding everything in disciplined processes and clear communication. First, I make sure the model development pipeline is rigorous: from data validation to feature selection to interpretability. No matter how innovative the technique, it needs to be auditable, reproducible, and justifiable to both technical and business stakeholders.
Second, I emphasize stress testing and scenario analysis early in the process. Understanding where a model might break—or how sensitive it is to certain assumptions—is key to building trust and resilience in high-stakes environments. I also lean on causal inference frameworks when appropriate to evaluate true business impact, not just predictive performance.
Finally, I believe in transparency. When the stakes are high, it’s essential to clearly communicate trade-offs, risks, and limitations to leadership. I’ve found that being upfront about what a model can and cannot do builds credibility and leads to better decisions. Pressure is part of the job, but with the right tools, mindset, and collaboration, it becomes manageable—and even motivating.
