Credolab, a global leader in behavioral and device metadata analytics, today announced the launch of its Income Prediction Model, a new machine-learning solution that enables lenders to estimate applicants’ income levels using privacy-consented smartphone metadata.
The model helps financial institutions assess creditworthiness even when traditional proof-of-income or credit history data is unavailable, a common challenge in emerging and digital-first markets.
Open banking was expected to make income and financial data more accessible, but the reality is proving far more complex. Income data access remains fragmented and inconsistent, with many institutions still relying on screen scraping and voluntary bank-Fintech agreements that weaken reliability. Across markets, banks and lenders now face a widening income verification gap. A growing share of applicants either cannot provide documented income or rely on data sources that are incomplete, unverified, or susceptible to manipulation.
Credolab’s new Income Prediction Model addresses this gap directly, giving lenders a privacy-preserving and statistically robust way to estimate income levels even when traditional data sources fall short. The model analyzes thousands of anonymized behavioral signals that collectively correlate with income levels, such as app ownership patterns, device model and age, paid app usage, and interaction habits. The model is trained on datasets provided by each client institution and fully customized to reflect local population characteristics and risk profiles. By design, the system never accesses personally identifiable information (PII) or demographic variables like age, gender, or education.
Each model is built from metadata collected with explicit user consent via Credolab’s lightweight SDK. Proprietary feature engineering transforms raw metadata into over 11 million behavioral features, narrowed down through rigorous selection methods (information value, correlation filtering, and gradient boosting) to a few dozen highly predictive indicators. The models employ elastic-net logistic regression and tree-based ensemble techniques, validated with out-of-time and out-of-sample testing to ensure robustness and explainability. At launch, the Income Prediction Model is available for Android devices, with additional platform support planned for future releases.
“In many markets, a lack of verified income data is the biggest barrier to financial inclusion,” said Peter Barcak, CEO and co-founder of Credolab. “Our new model gives lenders a privacy-safe and statistically sound way to infer income levels using only device behavior. It’s a powerful step toward fairer, faster, and more inclusive credit decisions, especially among populations for whom traditional data simply doesn’t exist.”
The Income Prediction Model strengthens Credolab’s mission to expand access to fair credit while reducing risk and fraud. Together with the company’s alternative credit, fraud, and marketing scoring products, the new model provides an added dimension of intelligence for banks and lenders seeking to better understand applicants with thin or no credit files.