SERASA's Crediscore and comparable Brazilian bureau products mainly assess people through their use of formal financial products: consumer credit repayments, utility payment histories, and bank account activity. A smallholder coffee grower in Nepomuceno or Tres Pontas may transact in cash through the cooperative and may never have held a consumer card or bank product. The result is a thin file. The bureau does not identify a "productive farmer with consistent yields." It identifies "insufficient data."
The analyst examining that thin file is left with limited options. Institutional risk procedures commonly treat a sparse bureau record as evidence of greater risk. The analyst may reject the application or request collateral and paperwork the grower cannot supply. The farm is denied financing not because it lacks productivity, but because its financial activity cannot be seen by the scoring system.
That is the central problem with using general credit scores for smallholder agricultural lending. The scores are not necessarily inaccurate for the borrowers on whom they were trained. They are being used for borrowers outside that original population.
What Generic Credit Scores Really Measure
Bureau scores in Brazil, including widely used SERASA Crediscore and Boa Vista SCPC products, rely mainly on payment conduct within formal financial markets. Their inputs include whether a CPF number is tied to an overdue registered debt, whether installments were paid on time, whether a checking account receives regular salary deposits, and whether earlier credit was repaid according to schedule.
Those inputs make sense for the groups these systems were built to assess: urban salaried employees and small business owners with formal accounts. For such borrowers, past payment behavior is useful because it shows a consistent link among earnings, financial commitments, and repayment habits.
Smallholder farmers in rural areas operate within another economic pattern. Income arrives seasonally rather than monthly. Their main "transaction" may be delivering coffee bags to a cooperative from June through September, rather than depositing cash at an ATM. Working capital often comes from informal sources because formal options were unavailable. As a result, missing formal repayment history may reflect exclusion from the system, not a separate indication of risk.
The Blank-File Problem: What Missing Data Really Means
A rural smallholder's minimal bureau history can point to three materially different circumstances. The grower may have managed money responsibly but never gained access to formal financial products. The grower may have avoided formal borrowing because the terms were unsuitable or the process was out of reach. Or the grower may have genuine credit management issues that never entered the formal system.
A generic score cannot separate these three situations using the bureau record alone. Each produces the same result: a thin file, limited score confidence, and a heightened risk flag. Rejecting the request or requiring substantial collateral is suitable for one case, but could be badly mistaken in the other two.
In Sul de Minas, many productive growers have never maintained a bank account because the nearest branch is a 45-minute drive and the cooperative manages their financial dealings. In this setting, the blank-file group is weighted strongly toward the first and second situations. The bureau system therefore labels many productive, creditworthy growers as unknowns with high apparent risk.
How the Problem Appears in Agricultural Credit
Using general scoring with agricultural borrowers creates a distinct failure pattern. It is worth examining closely because it can shape lender decisions without becoming obvious in portfolio results.
Lenders that make bureau scores their main rural lending screen create a self-selected book. Those who pass are the growers with formal financial histories. In rural Minas Gerais, that profile often aligns with larger, more commercial growers who have used cooperative credit programs or government subsidized lines. These borrowers may already present lower risk, making the bureau-screened portfolio appear to perform well.
The growers screened out, including many who could also repay successfully, remain unseen. They turn to expensive informal lenders, which deepens financial strain and can eventually raise their actual credit risk. In rural smallholder lending, the bureau process partly creates the risk it says it is preventing.
We saw this pattern in reviews of early-access pilot applications. Growers unable to obtain PRONAF or cooperative credit because of documentation gaps often brought us stronger production histories than growers with formal credit records but smaller or less consistent farms. The bureau and production signals diverge, and for a harvest-aligned loan, production provides the better indication of repayment capacity.
What Best Indicates Repayment Capacity for Smallholder Coffee
The issue is not whether general bureau scores have value overall. They do for the borrower groups on which they were calibrated. The more relevant question is which information predicts repayment capacity for a smallholder coffee farmer using a harvest-aligned loan.
Consistency in production is the leading signal. A grower who has delivered above 10 bags per hectare to a cooperative for four consecutive seasons, including two low-cycle years in the biennial bearing pattern, has shown stable output across normal conditions. That history says more about the ability to repay a harvest-aligned loan than a bureau score based on formal products the grower never could access.
Characteristics at the plot level provide the next signal. Altitude zone, soil classification, and satellite canopy health indicate the productive potential of the asset that produces the cash flow. A loan tied to harvest from a 1,000-meter arabica plot on latosol vermelho is backed by a more predictable income stream than one tied to a lower-altitude, less productive plot, regardless of the grower's consumer credit record.
The timing of cash flow is another structural factor that general scoring overlooks. Coffee harvest in Sul de Minas is concentrated from June through September. A borrower expected to make monthly payments throughout the year and one able to repay the same nominal debt according to the harvest cycle represent different credit risks, even when their debt is equal. Generic models miss this distinction because they were calibrated around monthly installment products.
Building a Different Model: Domain-Specific Inputs
The Culttivo model was designed for coffee smallholder credit in the altitude-growing areas of Sul de Minas. Its main inputs are altitude, soil classification, cooperative yield records, and Sentinel-2 NDVI data. Bureau scores are added when available, but when files are thin or missing, production information carries greater weight.
This method is not intended for every credit market. A production-data model for coffee smallholders in Minas Gerais is not a model for urban small business lending or rural livestock operations. The focus on one domain is deliberate: the model works because it reflects the borrower population and cash flow pattern of the financed product.
We are not claiming that SERASA scores have no place in agricultural lending. When borrowers have formal financial histories, bureau information can add value even within a production-data model. The narrower point is that for a large share of Brazilian coffee smallholders with thin or blank bureau files, using bureau thresholds as the main screen consistently leaves out productive, creditworthy growers. The alternative is to score the information that actually matters for this population, which is the approach we have taken.
The model is calibrated against our early-access cohort in Sul de Minas. As the cohort expands and repayment results accumulate, its weights will keep changing. That is how domain-specific model development works: calibration requires a real portfolio, and early-stage estimates carry more uncertainty than those from a mature model. We state this clearly to partner lenders who assess our scoring results.