When an app on a smartphone decides whether a person can borrow, the decision rests on patterns found in data rather than on paper forms or long bank histories. World Bank Group 2018 documents how uneven access to formal credit and rising mobile connectivity have created both demand for and the raw material of algorithmic underwriting, and McKinsey Global Institute 2021 describes how machine learning can reduce processing time and operating costs that have long kept small loans uneconomic.
New data, new decisions
Fintech companies and alternative lenders now draw on nontraditional signals such as payment flows, phone metadata and retail interactions to score risk. Bank for International Settlements 2019 cautions that this shift is driven by three systemic changes: abundant computing power, widespread digital footprints and advances in predictive models. The result is deeper segmentation of applicants and faster credit decisions, which can open borrowing to people previously excluded because they lacked formal employment records or lengthy credit histories.
Why this matters
Expanding access reshapes communities. In regions where informal work predominates, algorithmic underwriting can translate cash transfer histories or mobile airtime purchases into a bridge loan for a small trader. World Bank Group 2018 shows that such digital pathways matter for inclusion. At the same time, automated decisions reach into cultural practices and local markets, where seasonal earnings or family remittances alter risk profiles in ways that models must learn to respect rather than erase.
Winners, risks and regulation
The benefits are tangible: faster approvals, tailored credit sizes and lower transaction costs that can make microloans viable. McKinsey Global Institute 2021 finds that lenders deploying AI can improve loss prediction and scale services more cheaply. However, reliance on proxies for creditworthiness brings risks. The European Commission 2020 White Paper on Artificial Intelligence warns that opaque systems can reproduce social biases and entrench exclusion if training data reflect historical inequalities. Bank for International Settlements 2019 highlights concentration risks when a few platforms control both data flows and lending markets.
Human consequences surface in distinct ways. A gig worker whose earnings vary week to week may be offered small, high-cost credit because a model interprets variability as volatility rather than a predictable seasonal pattern. In rural territories where digital footprints are thin, algorithms tuned on urban users may underwrite poorly or deny access, amplifying territorial divides instead of narrowing them.
What will change
Practical responses are emerging. Lenders are investing in explainable models and human-in-the-loop reviews to catch errors and contextualize decisions. Regulators and standards bodies are experimenting with validation frameworks that demand transparency and data governance, as recommended in European Commission 2020 and reflected in supervisory discussions at international banks. If implemented carefully, AI-driven underwriting can democratize credit while requiring new institutional safeguards to prevent new forms of exclusion. The choice will determine whether algorithms become instruments of inclusion or of deeper, data-driven stratification.
