The SME Credit Risk Problem
Small and medium enterprise lending in emerging markets presents a fundamental information asymmetry. The borrower knows more about their business than the lender. Financial statements — where they exist — may be incomplete, inconsistent, or prepared to present the most favourable picture. Audits, where affordable, provide point-in-time assurance that may not reflect current conditions. And traditional credit scoring models, developed for data-rich environments with deep credit bureau coverage, perform poorly in markets where formal financial data is sparse.
Our research into SME credit risk — spanning over 1,000 academic papers and a citation graph of nearly 10,000 references — reveals consistent patterns in how this problem manifests and where emerging technology can address it.
What the Research Shows
Financial statements are necessary but insufficient. Every credit assessment framework starts with financial statement analysis. But in emerging markets, the quality and reliability of financial statements varies enormously. Small enterprises may maintain multiple sets of books. Tax-minimisation strategies produce financials that understate true performance. And the lag between statement dates and credit decisions means the lender evaluates a historical snapshot, not current reality.
The academic literature is consistent on this point: financial statement-based models achieve adequate discrimination in data-rich environments (AUC values typically between 0.70 and 0.85) but deteriorate significantly when applied to thin-file borrowers — precisely the SME segment where lending opportunities are largest.
Alternative data improves prediction. Research published between 2020 and 2025 demonstrates that incorporating alternative data sources — transaction data, mobile money patterns, utility payment history, trade credit relationships, social media activity, and digital footprint signals — significantly improves credit risk prediction for SMEs. Studies using mobile money transaction data in East African markets achieved discrimination levels comparable to or exceeding traditional models, despite the absence of formal financial statements.
For banks in emerging markets, this finding has immediate practical implications. The data exists — in payment systems, in mobile platforms, in trade registries, in utility databases. What does not exist, in most institutions, is the infrastructure to ingest, normalise, and analyse this data alongside traditional financial information.
Cross-referencing catches what single-source analysis misses. The most consistent finding across the research corpus is that credit risk assessment improves when claims from one source are verified against independent sources. Revenue figures stated in a loan application can be cross-referenced against GST filings, bank transaction volumes, and trade registry data. Asset claims can be verified against property registries and equipment financing records. Customer concentration claims can be tested against available trade data.
This is not a new insight — it is the principle behind every serious due diligence process. What is new is the ability to automate the cross-referencing at scale, across multiple data sources simultaneously, and to flag discrepancies before a credit decision is made rather than discovering them during collections.
A Pattern from the Research: The Georgian Banking Context
Georgia’s banking sector illustrates the challenge clearly. The country has a well-regulated banking system supervised by the National Bank of Georgia, strong economic growth driven by tourism and agriculture, and a growing SME sector that needs credit to expand. Georgian banks have made significant investments in digital infrastructure, core banking systems, and credit risk technology.
But the SME credit challenge persists. Financial statements from small enterprises may not capture the full picture of business activity. The informal economy, while shrinking, still accounts for a meaningful share of economic activity. And the credit bureau, while functional, has limited history for newer businesses.
Research on credit risk assessment in comparable markets suggests that Georgian banks could benefit from three capabilities that traditional credit infrastructure does not provide. First, automated cross-referencing of borrower claims against public registry data, tax filings, and transaction records. Second, alternative data integration — incorporating signals from digital payment platforms, utility systems, and trade relationships into credit scoring. Third, adversarial verification — treating every claim in a credit application as unverified until independently confirmed, rather than accepting submitted information at face value and relying on the credit officer’s judgment to spot inconsistencies.
The research does not suggest replacing credit officers. It suggests augmenting them with tools that check what they cannot check manually — because the time and data access required for comprehensive cross-referencing exceeds what any individual can do at the volume a bank’s SME book demands.
Implications for Emerging Market Banks
The research points to a structural gap in how emerging market banks assess SME credit risk. The gap is not in the quality of credit officers or the rigour of credit policies — it is in the tools available to implement those policies consistently at scale.
An adversarial approach to SME credit assessment would treat every application as a claim to be verified, not a statement to be evaluated. Revenue claims would be cross-referenced. Asset claims would be checked. Relationship claims would be confirmed. And the output would not be a score — it would be a verification report showing which claims were confirmed, which could not be verified, and which contradicted available evidence.
For banks evaluating AI infrastructure for credit risk, the deployment model matters as much as the capability. Credit data — borrower financials, transaction histories, credit bureau pulls, internal ratings — is among the most regulated and sensitive data a bank holds. Processing this data through external AI platforms creates both regulatory risk (data sovereignty, cross-border transfer, third-party access) and competitive risk (credit methodology is a competitive advantage). Self-hosted, sovereign AI infrastructure that operates within the bank’s own data governance framework addresses both concerns.
This analysis draws on DiligenceWorks research into SME credit risk assessment, including a systematic literature review of over 1,000 academic papers covering machine learning approaches to credit scoring, alternative data integration, fairness in credit decisions, and regulatory frameworks across multiple jurisdictions. No confidential client data is referenced. Geographic context is drawn from publicly available information about banking sector structure and regulation.