Vietnam's State Bank Credit Information Center (CIC) has grown steadily over the past decade, but bureau coverage still reaches fewer than one-third of the adult population. For a lender trying to originate personal loans or working capital credit in Ho Chi Minh City or Da Nang, that number means roughly two out of three potential borrowers come in with either no bureau record or a file thin enough to be statistically unusable. The standard response, declining or heavily pricing-up this segment, is rational at the individual loan level. At the portfolio level it leaves most of the addressable market on the table.
Mobile wallet adoption in Vietnam tells a different story. MoMo has tens of millions of registered users. ZaloPay is embedded in the country's most popular messaging platform. VNPay QR acceptance is widespread across grocery stores, pharmacies, and street food vendors in every urban district. The urban borrower who has no CIC record often has 18 to 36 months of mobile wallet transaction history sitting in a database that a lender has never asked to look at.
We started building Vietnam-specific cash-flow scoring in late 2024. What we found pushed us to restructure how we weight data sources at origination.
The CIC Coverage Floor and What It Actually Tells You
To understand where bureau scores help and where they don't, you need to understand what CIC actually contains. CIC stores data on formal credit facilities reported by licensed credit institutions: bank loans, consumer finance products from licensed companies, credit cards. The reporting cycle is monthly, and there's a lag between origination and first reporting. A borrower who took out a small personal loan six months ago may have four months of on-time payment history that hasn't yet propagated fully to CIC depending on reporting latency at their lender.
For the thin-file population, CIC gives you one of three things: a blank file, a single inquiry with no associated credit product, or a history covering one product over a short period. None of those inputs are sufficient to price a 12-month personal loan with any precision. Lenders who use bureau score alone typically set a file thickness threshold, declining anyone below it, and then apply score bands above the threshold. That logic works fine for salaried employees at formal companies who have built credit histories through bank products. It fails systematically for the informal trader, the freelance designer, the small restaurant owner, and the gig platform worker who may have significant financial capacity but no CIC history to show for it.
Which Cash-Flow Features Actually Carry Weight
When we analyzed transaction histories from a cohort of borrowers in Ho Chi Minh City applying for personal loans between 2024 Q3 and 2025 Q2, several feature categories separated well-performing from defaulting borrowers within the thin-file population.
Inflow timing regularity. The coefficient of variation on the time gaps between income deposits matters more than the deposit amounts. A borrower who receives income every 13 to 15 days consistently is showing something about employment or contract stability that a self-declared employment status field in an application doesn't capture. We compute this over rolling 90-day windows and look for consistency across multiple windows, not just the most recent period.
Outflow-to-inflow ratio stability. The ratio of total spending to total income in any given month is less interesting than how much that ratio varies across months. A borrower who consistently spends 70 to 80 percent of inflows, month after month, is demonstrating budget discipline even if the absolute amounts are modest. A borrower whose ratio swings from 50 percent to 130 percent in adjacent months is showing volatility that's worth investigating even if the average looks acceptable.
Minimum balance behavior. How a borrower manages the floor of their wallet balance between income cycles encodes financial buffer preference. We look at the rolling minimum balance as a fraction of average monthly inflow. A borrower who consistently maintains even a small positive buffer behaves differently from one who depletes to near-zero before the next inflow arrives.
Small-obligation clearing speed. Phone top-ups, micro-installment payments on consumer purchases, peer transfers with repayment characteristics. These are small informal obligations with no formal credit consequence, but the speed and consistency with which a borrower clears them tells you something about how they prioritize payment behavior when there's no formal enforcement mechanism.
Inflow Amount vs. Inflow Pattern: Getting the Feature Right
A persistent mistake when working with mobile wallet data is over-indexing on income amounts rather than income patterns. In Vietnam's informal economy, nominal inflow amounts vary substantially by season, market conditions, and gig volume. A food vendor in Binh Thanh District might see inflows fall by 30 to 40 percent during a slow week and recover the following week. That amplitude variation is normal commercial behavior, not financial distress.
The predictive feature is the structural pattern, not the level. Ratio-based features, rolling pattern metrics, and timing gap features consistently outperform raw amount features for this population in our models. A borrower with lower nominal income but a highly regular and stable pattern scores better on cash-flow signals than a borrower with higher but erratic inflows, and that difference in score maps reasonably well to realized default rates in the cohort we analyzed.
We want to be precise about what we're not claiming here: income level still matters for repayment capacity. The capacity-to-repay calculation requires knowing what actually flows through the account. What we're saying is that capacity and behavioral predictability are separate dimensions of creditworthiness, and standard income verification methods collapse them into a single variable. Cash-flow modeling separates them.
Where Cash-Flow Features Fall Short
Cash-flow scoring from mobile wallets isn't a universal solution for Vietnam's thin-file population, and treating it as one is a path to model failure.
The signal degrades significantly for borrowers who split transactions across multiple platforms. If a borrower uses MoMo for peer transfers and ZaloPay for merchant payments and VNPay for top-ups, and we're ingesting data from only one source, the partial observation looks like lower-activity behavior than the borrower actually exhibits. We've seen this produce artificially low scores for borrowers who are financially active but payment-fragmented. Building a robust multi-wallet reconciliation layer is a prerequisite for accurate scoring, not an optional enhancement.
Recent platform adopters create a different problem. A borrower who created a wallet account three months ago because a lender's onboarding flow required it has a thin transaction history that resembles genuine financial inactivity, even if they've been financially active in cash or through different channels for years. We apply a minimum history threshold of 9 to 12 months before relying primarily on cash-flow features, and fall back to supplementary signals or conservative pricing below that threshold.
Rural and peri-urban borrowers in provinces with lower digital payment acceptance present a structural gap. In areas where QR acceptance is sparse and cash is the primary transaction medium, wallet data simply doesn't capture most financial activity. The cash-flow approach is, in its current form, an urban and peri-urban tool. Applying it to rural populations without adjusting for data coverage risk generates misleading scores.
Practical Architecture for Lenders
For a digital lender in Vietnam trying to serve the thin-file market, the right architecture isn't replacing bureau scoring with cash-flow scoring. It's building a routing layer that sends borrowers to the appropriate model based on what data is available for them.
Borrowers with adequate CIC history (we typically use a minimum of 12 months with at least one credit product as the threshold) go through a bureau-primary model that may also incorporate cash-flow features as supplementary inputs. Borrowers below that threshold get routed to a cash-flow-primary model, provided they meet the wallet data sufficiency requirements: minimum 9 months of wallet history, a transaction frequency floor of around 10 to 15 transactions per month, and data from a single primary wallet rather than highly fragmented multi-platform data.
Borrowers who meet neither threshold shouldn't be automatically declined. That population often contains creditworthy individuals who simply haven't been in the digital financial system long enough. Building a manual review track or a small-first-loan product for this segment is a customer acquisition decision, not strictly a credit decision, but it's worth separating it from the broader scoring question rather than conflating "unscorable" with "uncreditworthy."
Vietnam's market is at an inflection point. CIC coverage is growing as more financial products move through licensed digital channels. Wallet adoption is extending into age cohorts and income brackets that were previously cash-only. Lenders who build cash-flow infrastructure now will have an advantage as both data sources thicken, rather than starting that buildout when the market has already moved.
