From AI Tools to Autonomous Systems: How Modern AI Could Transform Credit Access in Thailand

Key insights from Dr. Saran Ahuja, Chief Data Scientist at ABACUS digital, at Money20/20 Thailand Summit

Artificial intelligence has long been used in financial services—from fraud detection to credit scoring. But the evolution of AI is now reshaping how lending works at a much deeper level, opening new possibilities for financial inclusion at scale.

At the Money20/20 Thailand Summit, Dr. Saran Ahuja, Chief Data Scientist at ABACUS digital, shared perspectives on how a layered approach to AI—combining modern machine learning, alternative data, and the emerging capabilities of Agentic AI—is transforming credit access for underserved borrowers.

The shift represents a meaningful progression across fintech: from AI as a tool that improves specific decisions, to AI as a system that can continuously lower costs, sharpen risk prediction, and automate entire lending workflows.

The foundation: modern ML and alternative data

Traditional AI in financial services focused on specific tasks—credit scoring, fraud detection, or customer support—based largely on conventional data inputs like credit bureau records.

The first major leap forward came from combining modern machine learning with alternative data. Instead of relying solely on formal credit history, AI-powered platforms can now incorporate transaction histories, utility payments, digital behavior, and business activity to build a far more accurate picture of a borrower's true repayment ability—especially for those with thin or no credit files.

This matters because the two core challenges in inclusive lending are tightly linked: improving the accuracy of risk assessment for non-traditional borrowers, and reducing the end-to-end cost to serve—spanning acquisition, onboarding, underwriting, collections, and customer support—enough to make small loans economical. Modern ML with alternative data directly addresses the first, and laid the groundwork for tackling the second.

Thailand's lending gap: financial access vs. credit access

Thailand has broad access to bank accounts and digital payments, but access to formal credit remains uneven—especially for self-employed workers, freelancers, and small informal businesses that lack traditional income documents or credit history.

As a result, many turn to informal lending, with 42% of Thai households having used it. Closing this gap means solving both challenges at once: improving risk models enough to confidently lend to non-traditional borrowers, and reducing the end-to-end cost to serve—spanning acquisition, onboarding, underwriting, collections, and ongoing customer support—enough to make small loans economical in the first place.

This is where the layered AI approach becomes powerful. Modern ML with alternative data tackles the risk prediction side. Agentic AI then addresses the cost side more broadly, automating touchpoints across the entire lending lifecycle that were previously manual, expensive, and difficult to scale.

Putting it into practice

At ABACUS digital, the Abacus Core Technology (ACT) platform integrates 20+ proprietary AI/ML models analyzing over 2,500 risk variables to evaluate creditworthiness in real time. By leveraging alternative data and advanced risk modeling, the platform enables lenders to reach borrowers that conventional systems would reject—without compromising on credit quality.

The impact is tangible: platforms like MoneyThunder demonstrate this, with over 30% of borrowers previously rejected by traditional banks gaining access to credit through AI-driven assessments. Automation also lowers underwriting costs, making smaller, short-term loans viable for underserved segments.

Where Agentic AI takes it further

The effort to reduce end-to-end costs through AI is not new. Across the lending lifecycle, modern AI has already made significant inroads: conversational chatbots powered by NLP have handled customer acquisition and support at scale, while reinforcement learning models have transformed collections—dynamically optimizing how and when to engage borrowers to improve recovery outcomes. These initiatives have meaningfully lowered the cost to serve and improved the borrower experience well before the current wave of generative and agentic AI.

Agentic AI represents the next evolution of that work—not a departure from it, but a powerful amplifier. Unlike earlier AI tools that operate within defined boundaries, Agentic AI can reason, plan, and execute multi-step workflows autonomously, connecting previously siloed capabilities into more seamless, end-to-end processes.

In lending, this means the gains already achieved across acquisition, onboarding, underwriting, collections, and customer support can be pushed even further—with greater automation, less manual handoff, and smarter orchestration across the entire lifecycle. Borrower journeys that once required multiple touchpoints can become more fluid and conversational. Collections strategies that relied on rule-based triggers can become more adaptive and personalized. And the cumulative cost savings make it increasingly viable to serve borrowers at lower loan sizes and higher volumes.

In practice, Agentic AI capabilities are often introduced incrementally—first in operational functions like data collection and verification—while final underwriting decisions continue to be governed by established credit models and risk frameworks.

Responsible innovation at every layer

Responsible AI governance in financial services is not a new challenge — and the industry is not starting from scratch. Over the past two decades, financial institutions have built robust frameworks to manage AI risk: Model Risk Management (MRM) disciplines that govern how predictive models are developed, validated, and monitored; Responsible AI principles that address fairness, explainability, and bias; and data governance regulations that set clear boundaries around how consumer data can be collected and used. These frameworks did not emerge overnight — they were hard-won through years of regulatory dialogue, industry practice, and real-world experience.

Crucially, these same frameworks remain fully applicable as Agentic AI enters the picture. At the underwriting level in particular, modern ML credit scoring models continue to serve as the primary decision-making engine — meaning the model governance, validation standards, and compliance requirements that institutions have long followed still apply directly. Agentic AI may automate the processes around that decision, but the credit risk judgment itself remains anchored in established, auditable models.

The goal, then, is not to reinvent governance for a new era of AI — but to extend proven frameworks thoughtfully to cover the new capabilities Agentic AI introduces: greater autonomy, multi-step reasoning, and broader system integration. Keeping human accountability clear at each layer, and ensuring fairness and transparency are preserved end-to-end, remains the guiding principle.

Building the future of inclusive finance

For ABACUS digital, the development of AI-powered lending technologies is closely tied to its broader mission: expanding access to safe and transparent financial services.

By combining proprietary AI technology, alternative data, and a fully digital lending infrastructure, the company aims to reduce barriers that have historically excluded many individuals from the formal financial system. Modern ML lays the groundwork—making better risk decisions at lower cost. Agentic AI builds on that foundation, enabling greater automation, smarter borrower journeys, and more personalized financial products.

Together, these capabilities point toward a lending ecosystem that is not only more efficient—but genuinely more inclusive. And for millions of borrowers who remain underserved today, that transformation could open the door to real economic opportunity.