Philippine digital banks have a lending problem that is also their biggest opportunity. The borrowers they were designed to serve—Filipinos with limited credit histories, irregular incomes, and no collateral—are precisely the borrowers traditional banks have long avoided. Serving them is the industry’s inclusion mandate and its profit engine. But the same segment produces non-performing loan ratios that are roughly twice the banking system average. In 2026, the question is no longer whether digital banks should lend to this segment. It is whether they can price, underwrite, and monitor that credit well enough to make the economics work.
The risk gap is structural, not accidental
Digital banks’ NPL ratio stood at 6.16 percent in January 2026, compared with 3.31 percent for the entire Philippine banking system. By April 2026, the sector’s NPL ratio was 5.7 percent against a system-wide 3.4 percent. The gap is not evidence of recklessness. It reflects the customer segment: borrowers with thin files, informal income streams, and limited experience with formal credit. Digital banks are lending where legacy underwriting models have historically refused to go.
The trajectory, however, is more encouraging than the level. The sector’s NPL ratio peaked at 25.3 percent in March 2024 and declined to 7.0 percent by July 2025, suggesting that underwriting models are maturing as data accumulates. The question for 2026 is whether that improvement can continue as loan books scale and as macroeconomic conditions tighten.
Maya’s AI underwriting as a risk-management template
Maya’s experience offers one benchmark. Its gross NPL ratio stood at 4.9 percent as of March 2026, supported by an AI-driven underwriting system designed to extend credit access to unbanked and underbanked Filipinos while keeping losses contained. The system does not eliminate risk; it prices and segments it more granularly than traditional credit scoring can. For a bank operating at Maya’s scale—₱68 billion in deposits and 10.7 million customers—small improvements in underwriting accuracy translate into large differences in provisioning costs.
Tonik’s model pushes the same logic further. Its lending is concentrated in salary-deduction loans through employers, merchant installment financing, and digital personal loans, all underwritten by an AI-driven system that enables profitable lending to thin-file borrowers. The 51 percent net interest margin and 25 percent risk-adjusted return on capital reported in the first quarter of 2026 suggest that, at least for its current portfolio, the yield is compensating for the risk. The test is whether those metrics hold as the loan book grows beyond its current US$110 million.
GoTyme’s caution: credit expansion as a dial, not a switch
GoTyme has taken a more conservative posture. Its NPL ratio stood at 5.6 percent at the end of 2025, and management expects it to remain in the 5 to 7 percent range through 2026. The bank has said it may slow lending expansion if borrower repayment trends weaken, tracking monthly loan cohorts for days past due and repayment rates. Its credit book remains short-dated, and a substantial portion of its balance sheet is invested in low-risk treasuries, limiting exposure to a sharp deterioration in asset quality.
That caution is not an absence of ambition. It is a recognition that in a segment where credit histories are thin, the cost of learning through losses can be high. GoTyme’s approach—test, learn, and throttle back when indicators deteriorate—treats credit expansion as a dial rather than a switch.
The inclusion tension that will not disappear
The credit paradox cannot be fully resolved. Digital banks exist in part to serve borrowers whom the formal system has excluded, and that service necessarily carries higher risk than lending to established corporate clients. The goal is not to eliminate the NPL gap but to ensure that pricing, provisioning, and capital allocation reflect it accurately. The banks that master that balance will not only profit; they will demonstrate that financial inclusion and credit discipline can coexist.
