Ask residents in any low-income Sao Paulo neighborhood which bill they pay first each month, and the near-universal answer is electricity. The reason is not that utilities frighten people more than banks or landlords. Losing power is immediate and tangible: no refrigerator, phone charging, or light for children studying. The behavioral economics are plain. People protect what matters most. For households carefully managing scarce cash, utility payments serve as the anchor.
That observation shapes our view at Supersymm of utility payment data as a credit signal. Our initial question was not "can we collect this data" but "does it predict repayment behavior." After early-access pilots with lender partners, the answer is a qualified yes. The qualification matters: as with every alternative signal, utility data offers information, not certainty.
What Utility Payment History Can Reveal
Bureau data gives a credit underwriter a structured record of formal obligations: missed payments, defaults, and balances due. It shows how a person acted when money was explicitly owed to a financial institution. Utility history captures another behavior, namely how someone pays obligations embedded in ordinary life, month after month, without a collections department calling.
Predictive features go beyond whether a bill was paid. The stronger information lies in the pattern: how early or late was payment relative to the due date? Did timing stay consistent across months? When cash tightened, shown by a late or partial payment, did the borrower recover the next month, or did the problem spread? Cash-flow resilience, rather than consistency alone, distinguishes good risk from marginal risk.
In a pilot with a regional digital lender serving the greater Sao Paulo area, borrowers paying electricity within three days of the due date in at least 10 of 12 months had default rates roughly one-third below the average for new small loans. This was not a controlled clinical study, so we do not claim statistical finality. Still, the direction matches findings from researchers in similar markets: regular utility payments meaningfully predict risk, especially among borrowers lacking another financial data footprint.
The Feature Engineering Problem
Utility records in raw form are not model features. They may show that a bill was paid on a certain date and for a certain amount. Several processing steps are needed before that becomes useful input.
Normalization comes first because billing cycles differ by utility and region. Some utilities bill monthly, while others bill every two months. Comparing timing across borrowers requires a common frame. We convert every record to a 30-day cycle and express the payment date as a delta from the due date.
Next come aggregation windows. In our experience, 6 to 18 months of history is most predictive. Newer months receive greater weight, since behavior can change. Someone punctual three years ago but consistently late for the past six months is not a good risk, whatever the long-term average says.
Third is consistency across signals. Utility regularity is most useful when compared with other indicators. Someone paying electricity on time but recharging a mobile erratically may require a more nuanced assessment. No individual signal acts as a standalone gate. Instead, signals enter a weighted ensemble, with each contribution tied to its predictive power on the validated training set.
Consent, Coverage, and Data Access
For lenders, the practical issue is obtaining the data. At our current scale, we have found two workable routes.
The first route is borrower-initiated sharing. During the loan application, the applicant consents to provide utility account information. That may mean uploading a PDF statement or, where utility APIs exist, retrieving structured data through OAuth-based authorization. This route has clear consent and suits applicants seeking a better credit result. Yet active participation creates selection bias: people willing to share may already display stronger credit behavior.
The other route involves a utility operator partnership, with the lender and utility entering a data-sharing arrangement. That avoids selection bias and supports full-population scoring, but utilities are not always organized around third-party data products, so building the relationship takes time. In Brazil, utilities span state-owned and privatized firms with uneven data infrastructure. The route is viable, though slow.
We cannot yet tell every lender, "plug in and get utility data for any borrower you send us." Our data access layer is still being developed, and we want prospective partners to understand that plainly.
Utility Data Within the Signal Stack
Across the alternative signals we use, utility payment history performs strongly on one dimension and more weakly on another. Its stability is a strength: utility behavior tends to remain consistent over time, unlike mobile recharge patterns that may shift within a two-week period. Coverage is weaker because not everyone has a bill in their own name. Renters whose utilities are included in rent, young adults in family homes, and people in informal housing may have no utility history.
Accordingly, we use utility data as one layer of a multi-signal method, not as the main scoring input. Where history exists, it gives the model meaningful lift. Where it does not, the model uses other available signals. Requiring utility data would unintentionally leave out the most marginalized borrowers, undermining the purpose.
What This Signal Does Not Show
A lengthy record of on-time utility payments is not nearly enough to establish creditworthiness. It is not. Across a population, regular utility payments correlate with creditworthiness, but individual outcomes differ. Someone may pay electricity on time for years and still default after a health emergency, business failure, or sudden income shock. The signal predicts a distribution, not a person's result.
More fundamentally, the record describes the past. A borrower who recently lost their main income may have an immaculate utility history and still be a poor risk today. That makes freshness as important as depth. We heavily weight recent months because a six-month-old indication of a changed financial situation is nearly noise.
For lenders establishing alternative data scoring programs, utility payment history is among the more accessible and useful places to begin. But useful and accessible do not mean sufficient. The signal performs best within architecture combining several signal types, weighting recency, and reporting confidence in the output instead of treating one input as decisive.