AI di Perbankan: Efisiensi vs. Risiko Kredit yang Mendesak

The rapid evolution of technology and data is fundamentally reshaping the landscape of credit, particularly in the realm of underwriting. Artificial Intelligence (AI) stands as a pivotal force in this transformation, poised to continue its profound impact on financial services.

AI's Strategic Shift: From Operational Backends to Core Balance Sheets

During the recent third-quarter earnings season, leading U.S. banks reported robust balance sheets, a normalization of credit cycles, and significant strides in digital efficiency. Beneath these headline figures, particularly within the operations of the banks' partners and FinTechs they back, AI is progressively redefining the nexus between risk assessment and credit opportunity. This shift, however, is being approached with a notable degree of circumspection by executives, who seem careful to temper expectations.

As earnings season gained momentum, Citigroup emerged as one of the most vocal proponents of AI's strategic integration. Its Q3 2025 earnings report highlighted that "investments in new products, digital assets, and AI are driving innovation and improved capabilities across the franchise." The bank further detailed a compelling efficiency gain: "~1 million automated code reviews completed by our Gen AI tools year-to-date saved approximately 100,000 hours per week across our developer population." Such efficiency is not confined to technology operations alone. Citi's CEO, Jane Fraser, conveyed to analysts that the underlying infrastructure supports a broader initiative: "we launched a firmwide effort to systematically embed AI in our processes end-to-end to drive further efficiencies, reduce risk, and improve client experience."

Elsewhere, JPMorgan Chase reported another strong quarter, though CFO Jeremy Barnum acknowledged "slightly elevated charge-offs as a result of a couple of instances of apparent fraud in certain secured lending facilities." These remarks serve as a pertinent reminder of tightening risk governance and the corollary that expanding automation and data utilization, particularly within the extended "financial supply chain," necessitate increased oversight costs. Similarly, at Goldman Sachs, executives articulated AI as a critical lever for achieving "productivity gains, process automation, and client service enhancement," integral to its overarching "One GS 3.0" strategy aimed at streamlining operations. Bank of America CEO Brian Moynihan, reflecting on Q3 results that surpassed expectations—reporting $8.47 billion in profit on $28.09 billion in revenue—attributed these successes, in part, to strategic investments in digital engagement as key drivers of efficiency.

From Customer Engagement to Sophisticated Credit Modeling

It has become increasingly evident that AI's influence within banking now spans the entire spectrum, from basic customer interactions at the teller window to intricate calculations affecting the balance sheet. Customer-facing AI tools are proving instrumental in educating financial institutions on how to personalize services effectively and, by extension, how to gain deeper insights into borrower behavior.

PYMNTS Intelligence research indicates that a significant 72% of customers would either remain loyal or return to a bank if they experienced personalized services delivered through embedded conversational AI. When banks deploy digital assistants to answer queries or identify spending patterns, the data signals generated from these interactions can be highly informative for credit decisions. Another PYMNTS analysis highlighted that AI-powered fraud prevention systems are revolutionizing the battle against financial crime, capable of analyzing thousands of data points within milliseconds. This same formidable analytical capability, when re-directed towards underwriting processes, empowers banks and other lenders to assess repayment ability with unprecedented dynamism and precision compared to traditional methods.

Alternative Data: Broadening the Credit Assessment Spectrum

At the core of this transformative shift lies the utilization of alternative data—an ever-expanding array of non-traditional signals that banks employ to evaluate risk. As Concora Credit executive Kyle Becker explained to PYMNTS, "We use a wide variety of credit bureau data as well as alternative data in making our credit decisions. Alternative data is super useful because it allows you to maintain or reduce risk while also providing access to credit to more people."

This expansive dataset encompasses a wide range of information, from rental and utility payment histories to mobile-bill records and real-time transaction data. When these diverse signals are processed through advanced machine-learning models, they possess the remarkable potential to unlock credit access for borrower segments traditionally overlooked by conventional credit bureau scoring systems. However, the deployment of such data and models mandates rigorous validation processes and vigilant bias monitoring to ensure fairness and accuracy.

The Cautionary Tale of Tricolor: Emphasizing Data Discipline and Governance

Recent events have unequivocally reinforced the paramount importance of caution in AI's application within lending. The bankruptcy of Tricolor Motor, a lender specializing in financing used-car purchases for thin-file borrowers through AI-powered underwriting, has sent ripples throughout the financial sector and was notably addressed during Q3 earnings calls. JPMorgan CEO Jamie Dimon candidly admitted that the bank’s exposure in this instance was "not our finest moment." Tricolor’s unfortunate collapse serves as a stark illustration of the inherent dangers associated with rapidly scaling credit portfolios that are underpinned by complex data pipelines. An expansion in the number of data points and sophisticated models inevitably amplifies the potential for model drift, introduces data gaps, and can expose vulnerabilities arising from weak control mechanisms.

For financial institutions actively experimenting with AI-driven underwriting, this lesson resonates deeply. The pursuit of efficiency and financial inclusion must never outpace the foundational pillars of validation and robust governance. As PYMNTS recently observed, investments in AI ought to be treated akin to capital expenditure: requiring measurable returns, traceable risks, and a tight alignment with overarching business outcomes. This disciplined approach ensures that the transformative power of AI is harnessed responsibly, mitigating potential pitfalls while maximizing its strategic benefits.