Case

Big financial institution

Financials

Scoring credit risk with industry context

Tabular MLBayesian modelingRisk reductionProcess efficiency

Modulai developed a Bayesian hierarchical logistic regression model to create better industry-specific payment delinquency estimates for SME lending. The scoring system gives a quantitative assessment of the probability that a loan goes 'bad', including default, insolvency, or bankruptcy, while accounting for how financial KPIs affect companies differently across industries.

Challenge

The client is an American software company that provides compliance, credit risk, and lending solutions to financial institutions. How financial KPIs affect a company's performance and ability to repay loans differs depending on its industry. For example, a current ratio below 1.0 might be acceptable for a retail company with a short cash conversion cycle, while manufacturing companies typically need a significantly higher ratio.

Solution

Rather than a pooled model, which assumes all industries share the same regression coefficients, or an unpooled model, which treats each industry independently, Modulai used a partial pooling approach. This hierarchical, multilevel model assumes different regression coefficients per industry while drawing them from a common prior, capturing industry-specific effects while sharing statistical strength across groups.

Tools

The Bayesian model was developed using the probabilistic programming library NumPyro, which has a JAX-based backend for fast compiled execution and supports GPU acceleration. This addresses the common challenge of slow sampling when building Bayesian models.

Value created

By modeling each industry separately while still letting them inform one another, the system produces credit risk estimates that reflect how the same financial signal can mean different things in different sectors, where a current ratio that is fine for a retailer would be a warning sign for a manufacturer. For a lender, that means a more accurate read on the probability that an SME loan goes bad, and a credit decision grounded in how a given industry actually behaves rather than a one-size-fits-all benchmark.

Learn more

Related content

Financials

Forecasting balances to catch high-risk customers early

Modulai developed an interpretable, robust time-series forecasting solution for a major e-commerce checkout provider. By decomposing a complex credit-risk challenge into distinct ML subproblems, the system produced forecasts that enabled earlier detection of high-risk customers and reduced credit losses.

Case

Financials

Predicting a company's future financial performance

Modulai built a deep learning system that predicts a company's annual report one year into the future. By using a single transformer-based model that learns the interdependencies between financial features, it produces more consistent and accurate predictions than separate models for each KPI.

Case

Financials

Building deep research AI assistants for an investment firm

Modulai built an agentic research system for a leading European investment firm that enriches reports by performing iterative research. It autonomously analyzes documents, reasons about potential gaps, and conducts structured investigations to surface overlooked insights and recent developments, providing a more complete picture for high-stakes decision-making.

Case

Financials

Transaction risk model for a major online payment platform

Modulai built a transaction risk prediction model that processes millions of payments annually across thousands of e-commerce sites. The project covered feature engineering, model development, decision engine specification, and a full deployment plan, enabling the client to proactively contain financial risk across their payment network.

Case