A RECENT report showed that over 70 percent of consumers are unbanked or underbanked across Southeast Asia. Figures from Public First revealed that 77 percent of under-35s in the region are using gen-AI weekly at work, with a mobile-first user base that has exceeded 7 million in the past year.
Dr. Otgonbayar Uuye, head of AI and co-founder of leading fintech company AND Global, firmly believes that the impacts of artificial intelligence (AI) go beyond consumer models, as they are significantly altering the financial landscape, especially for SMEs. In a recent interview with The Manila Times, Dr. Uuye, who earned his Ph.D. in mathematics from Pennsylvania State University, discussed the role of AI and machine learning (ML) models in underwriting microloans using alternative B2B data. He said these models are also helping reduce the digital divide by lessening dependence on loan sharks.
The Manila Times (TMT): What are the capabilities of artificial intelligence (AI) and machine learning (ML) in financial transactions?
Dr. Otgonbayar Uuye (Dr. Uuye): At its core, AI and ML allow us to do something traditional finance has always struggled with: make fast, accurate decisions at scale using data that would overwhelm any human team. In financial transactions, this spans three broad areas: risk assessment, document intelligence and process automation.
On the risk assessment side, ML models can evaluate a borrower’s creditworthiness in minutes by analyzing hundreds of data points simultaneously, such as transaction histories, behavioral patterns and business activity indicators, and weighting them against known outcomes. This is fundamentally a pattern-recognition problem, and the math behind it involves techniques suited to different types of data structures.
On the document intelligence side, AI-powered OCR and natural language processing can extract and verify data from financial documents, such as bank statements, invoices and tax filings, at speeds up to 300 times faster than manual entry, with up to 80 percent fewer errors. This is not just about reading text. The system must understand context, validate consistency and flag anomalies.
The third capability, process automation, ties these together. A complete loan origination process — from application intake to credit decisioning to disbursement — can now be executed end-to-end with minimal human intervention. At AND Global, through our loan origination platform Looms, we have demonstrated that 100-percent automated credit decisioning can be achieved within three minutes of customer onboarding.
TMT: How do AI and ML reduce credit risks, especially for lending to small and micro-enterprises with no collateral?
Dr. Uuye: Traditional credit risk assessment is built on a simple premise: past behavior predicts future behavior. If you have a long credit history with a formal bank, the model has plenty of data to work with.
But for micro and small enterprises — the sari-sari store owner, the market vendor and the small logistics operator — that history often does not exist in any formal system.
This is where the learning process becomes critical, and I mean learning in a very practical sense.
When we first extend credit to an underserved borrower, we start small: a micro-loan, a modest amount with limited risk exposure. As that borrower repays and builds a track record within our system, the model learns their behavior, repayment consistency, cash flow patterns and business cycles. Over time, loan amounts can increase because the model now has real performance data to work with.
This gradual approach means we are not making a single high-stakes bet on an unknown borrower. We are building a data relationship, one transaction at a time, and the algorithm refines its confidence with each cycle.
For uncollateralized lending to MSMEs more broadly, the key innovation is alternative data. Instead of requiring a credit bureau score, we look at what the business actually does. Bank transaction data, for instance, reveal cash flow regularity, average balances and seasonal patterns. Sales logs, even handwritten ones, can reveal revenue trends when digitized through document processing. These alternative data signals feed into models that combine rule-based systems with machine learning, an approach we call hybrid scoring.
TMT: Can AI/ML overcome the challenge of borrowers with no official record or digital footprint? How do you distinguish legitimate applicants from bad actors?
Dr. Uuye: This is one of the most meaningful problems in financial inclusion, and it is where our work becomes both technically interesting and socially impactful.
The challenge breaks down into two distinct problems. First, identity verification — confirming that this person or business is who they claim to be. Second, credit assessment — determining whether they are likely to repay.
For identity verification with limited digital footprints, we have developed what we call an e-KYB (electronic Know Your Business) framework. Our e-KYB approach uses a multilayered verification strategy. Instead of relying on a single authoritative database, the system triangulates across available data points, such as a business license (if one exists), physical location verification, supplier relationships and customer transaction patterns.
No single data point is definitive. But the convergence of multiple weak signals can produce strong confidence.
For distinguishing legitimate applicants from fraudulent ones, the model learns from patterns. Legitimate micro-businesses tend to show consistent, predictable transaction behaviors — regular small purchases from suppliers, seasonal fluctuations that match their industry and gradual growth patterns. Fraudulent applications tend to show different signatures — synthetic identities, inconsistent documentation and unusual application patterns.
The critical philosophical point is this: if AI is going to replace human loan officers in the decision process, it has a responsibility to recognize legitimate borrowers without bias. Our AI systems need to be free of that bias and recognize applicants as fairly as possible, based on data, not appearance. That is both the challenge and the promise of what we are building.
TMT: Is this risk-reduction capability near 100-percent assurance? Or is it one validation point in a lending officer’s checklist?
Dr. Uuye: This is a question I appreciate because it gets at a common misconception. No AI system, regardless of sophistication, provides 100-percent assurance. That is not how probability works. What AI does is dramatically improve the accuracy and consistency of risk assessment compared to manual processes.
In our deployments, we have seen AI-based credit scoring reduce credit-evaluation time from one hour per person to approximately five minutes, while simultaneously improving prediction accuracy. However, the role of AI in the decision process varies by product type and regulatory environment.
For smaller-ticket loans, such as micro-loans and small cash advances, the AI system can handle 100-percent automated decisioning. The risk exposure per loan is small, the volume is high, and the cost of manual review would make the product economically unviable.
For larger loans or more complex products, AI serves as a powerful decision-support tool within a broader workflow. The model generates a score, flags risk factors and provides an explanation of its assessment. A human underwriter then reviews it, applies judgment on factors the model may not capture and makes the final call.
The key insight is that AI and human judgment are not competing alternatives. They are complementary.
TMT: Is there still a need for government guarantee facilities to cover microloans?
Dr. Uuye: Absolutely. AI reduces risk, but it does not eliminate it. Government guarantee facilities serve a fundamentally different function than credit scoring, as they distribute systemic risk, provide a safety net that encourages lenders to extend credit to higher-risk segments and signal government commitment to financial inclusion. Both AI and guarantees are necessary protection measures.
