Utility bill payment history has been discussed as a potential credit signal for long enough that some people in the industry have started treating it as established fact. It isn't. The empirical quality of utility data as a credit predictor varies substantially across markets, across borrower segments, and across data access channels. This piece documents what we've measured, what context it sits in, and where utility payment history adds real predictive value versus where it doesn't.
The short answer: in Vietnam and Indonesia, utility payment history shows meaningful predictive lift for thin-file borrowers on smaller loan products when the data is recent, consistent, and clean. In the Philippines, the picture is more complicated. We'll walk through each.
What We're Measuring and How
Utility payment history as a credit feature is not one thing. There are at least four distinct signals extractable from a utility payment record, and they carry different information.
On-time rate is the most obvious: what percentage of bills did this person pay before or on the due date? This is a direction signal but not a precision signal. In Vietnam, EVN and provincial electricity co-ops send billing notices via mobile wallet and SMS; on-time payment data is generally accessible and reasonably accurate. In Indonesia, PLN's billing infrastructure produces digital records that can be accessed through the borrower consent flow. Both of these cases produce a usable on-time rate.
Average days-to-payment is more predictive than binary on-time rate. A borrower who consistently pays two days before due date, month after month, looks different from one who pays exactly on due date and sometimes a day late. The behavioral regularity in days-to-payment maps to a habit structure that correlates with loan repayment behavior.
Seasonal variation in payment timing is a feature that requires at least 12 months of history to compute. Does this borrower pay later in November and December? Does payment timing shift around Ramadan? Seasonal variation that is predictable is not the same as erratic payment behavior. A model that can't distinguish the two will misclassify borrowers who are actually lower-risk than they appear during stress periods.
Account continuity matters. A borrower with 36 uninterrupted months of utility payments at one address tells you something different from one with an 8-month record and gaps. Continuity is a partial proxy for address stability, which in turn correlates with employment stability in the populations we score.
Vietnam: High Data Quality, Clear Signal
Vietnam's EVN (Electricity of Vietnam) has achieved fairly broad digital billing coverage in urban areas, and several provincial co-ops have followed. Mobile wallet integration through MoMo and ViettelPay means that a borrower who pays their electricity bill via mobile wallet produces a clean timestamped record. The data access infrastructure is there, and when we have consent-based access to 12-plus months of electricity payment history for a borrower, the signal is useful.
In our Vietnam analysis, borrowers with 24 consecutive months of on-time electricity payments, combined with 90-day average days-to-payment below 3, showed default rates in our validation cohort that were roughly consistent with a bureau-scored borrower with a mid-range credit score. That's a meaningful finding for lenders who want to extend to borrowers the bureau can't assess.
The caveat is data coverage. A significant share of urban thin-file borrowers in Ho Chi Minh City rent rooms in buildings where the electricity account is in the landlord's name. The tenant pays the landlord, who pays EVN. There's no record connecting the borrower to the utility bill. Urban rental concentration is high enough in Vietnam that this coverage gap is a real constraint on utility data usefulness, not a fringe case.
Indonesia: Stronger Data Infrastructure, Different Context
Indonesia's PLN (Perusahaan Listrik Negara) has invested in digital billing and payment infrastructure, and GoPay, OVO, and Dana have all integrated utility bill payment into their platforms. The data trail is often better than Vietnam because mobile wallet utility payments are more prevalent in the urban segments we primarily assess.
The predictive pattern in Indonesia is similar to Vietnam for the on-time rate and average days-to-payment features. Seasonal variation is more pronounced around Lebaran, which creates a consistent pattern of slightly delayed payments in the weeks before Eid followed by normalization. A model that doesn't account for this will misflag a large share of otherwise-reliable borrowers during Ramadan.
One notable finding in the Indonesia data: water bill payment history (PDAM records) adds incremental information beyond electricity payment when both are available. The two utility types have slightly different demographic coverages and different payment cadences. Borrowers with both electricity and water records in their own name, with consistent payment across both, are a relatively rare segment but perform well in risk assessments.
Philippines: Coverage Gaps Limit Usefulness
The Philippines is where utility payment data gets complicated. Electricity distribution is handled by a fragmented set of distribution utilities across regions. In Metro Manila and nearby provinces, Meralco produces reliable billing records accessible through digital channels. In the Visayas and Mindanao, the picture is more patchy. GCash and PayMaya utility payment integrations exist but are not as universal as Indonesian mobile wallet utility payments.
More significantly, OFW remittance receiver households, which are among the borrower segments we most want to score in the Philippines, often have utility accounts in their name but receive most of their bills through paper or through family members who manage them. The digital payment trail may be incomplete or may not be linkable to the borrower identity in the consent flow.
We are not saying utility payment data is useless in the Philippines. We're saying it's a weaker primary signal than in Vietnam or Indonesia, and it works better as a supplement to other alternative data (GCash transaction patterns, remittance receiving cadence) than as a standalone predictor. Relying on it as a primary thin-file signal for Philippine borrowers would be a mistake.
Feature Engineering Considerations
Raw utility payment records are rarely usable directly in a credit model. The feature engineering layer matters almost as much as whether you have the data at all.
Time-window normalization is important: a 6-month utility payment record should not be treated as equivalent to a 36-month record. We weight recent months more heavily and apply a coverage discount to short histories. A borrower with 8 months of on-time payments gets a different utility feature value than one with 36 months, even if the on-time rate is identical.
Seasonality adjustment requires knowing the local festive and agricultural calendar. Raw days-to-payment in month 11 in Indonesia is not the same signal as raw days-to-payment in month 7. Failure to adjust produces a systematic misclassification of otherwise-reliable borrowers during predictable stress periods.
Missing data handling is a genuine risk. When utility payment data isn't available for a borrower, the temptation is to impute a neutral value. This is wrong for thin-file populations where data absence has a different meaning than it does for banked populations. A borrower with no utility payment record may be a renter (data inaccessible, not indicative of behavior), or may genuinely have no stable housing, or may have only recently moved to a covered area. These are different situations and the model should reflect that uncertainty rather than imputing a middle value.
What This Means for Lenders Evaluating Alternative Data Sources
Utility payment history is worth incorporating for thin-file lending in Vietnam and Indonesia when the borrower consent flow can produce at least 12 months of clean records. It should be treated as a supplementary signal in the Philippines, not a primary one. In all three markets, the feature engineering choices around time weighting, seasonality, and missing data handling have a larger effect on signal quality than the raw data access decision.
The broader point: individual alternative data sources rarely transform a thin-file model on their own. The value comes from combining three or four signals that capture different dimensions of borrower behavior. Utility payment history, when it's available and clean, is a genuinely useful piece of that combination. What it is not is a substitute for cash-flow analysis or repayment velocity measurement, both of which carry more predictive weight in our market experience across these three markets.
