Agricultural lenders serving smallholder markets face a clear issue: the information required for sound credit decisions is available, but it has not been organized in a format that credit models can use. Cooperative yield books, EMBRAPA soil maps, satellite canopy measures, and INMET rainfall records can all indicate whether a particular farm will produce enough to repay a loan. The problem is not a lack of data. It is putting the data together.
Why Rural Lending Default Rates Exceed What They Should
Weather uncertainty, commodity price changes, and the concentration of income in one season are the usual explanations for high agricultural default rates in Brazil. Those factors matter. Yet another cause receives less focus: selecting unsuitable borrowers when loans are originated.
When a bank or cooperative bases lending mainly on what an applicant reports about their farm, the information is necessarily incomplete. A grower may overestimate expected yields. Without records from third parties, past informal income is difficult to confirm. A self-reported application does not show the farm's physical condition, soil health, true canopy density, or exposure to altitude-related microclimates.
This leaves lenders making probability-based decisions from weak inputs. Some approved loans will perform well, while others will not. Default rates therefore reflect both actual farm risk and the information gap present at origination. Better origination data narrows the range of outcomes, producing fewer unexpected defaults and more precise portfolio pricing.
Three Data Layers That Indicate Repayment Capacity
Our work with early-access pilot growers in Sul de Minas supports what agronomists and cooperative managers have long observed: when combined, three variables account for most differences in coffee smallholder repayment capacity. These are production consistency, meaning yield per hectare over multiple seasons adjusted for biennial bearing, site quality, including altitude, soil organic content, and frost exposure, and financial load, or existing debt compared with expected harvest income.
The first two variables can be observed through sources already available in the public domain or through cooperative records. Measuring production consistency requires cooperative yield registers, which most large cooperatives in Minas Gerais keep electronically. Assessing site quality requires GPS plot boundaries, SRTM elevation data, EMBRAPA soil maps, and Sentinel-2 NDVI time series. Financial load can be seen partly through the Banco Central's SCR and partly through the grower's disclosure of informal credit obligations.
A lender with all three layers for an applicant is in a very different position from one using only a SERASA score and self-reported income. SERASA indicates formal credit behavior, an area where most Sul de Minas smallholders have little or no history. The three-layer approach provides an indication of the actual cash flow the farm may produce over the next twelve to eighteen months.
A Practical Example: Assessing a 15-Hectare Applicant
Take a grower seeking a R$45,000 working capital loan for a 15-hectare plot in the Alta Mantiqueira region, at approximately 1,100 metres above sea level. The application provides GPS boundaries and two seasons of cooperative delivery records, with yields of 22 to 25 bags per hectare, each bag weighing 60 kg. Those figures match what that altitude zone produces with normal management.
With no formal payroll and only a thin SERASA file, a conventional bank might reject the request or ask for collateral the grower cannot offer. Using the available plot data, a plot-level model would estimate gross revenue from a 22-bag-per-hectare yield at current arabica prices, compare it with the loan amount and monthly interest rate, calculate the debt-service coverage ratio, and test whether altitude, NDVI, and soil type support the reported historical yields.
The application could still be rejected if debt-service coverage is too narrow or satellite records indicate that canopy health has declined over the past two seasons. The difference is that the decision now assesses whether this particular farm can produce enough cash flow for repayment, rather than judging whether the grower has the paperwork profile of an urban salaried employee.
Portfolio Benefits of Detailed Scoring
Improving individual loans is only one advantage. Portfolio design is another. Scoring applicants against actual farm-quality factors lets lenders see and manage how risk is distributed across the portfolio. They can identify concentrations in higher-risk agricultural segments by altitude, soil type, or weather exposure, then change pricing or exposure limits as needed.
This level of portfolio detail is common in mature lending markets. A commercial real estate lender tracks loan-to-value and property-specific cash flow for every asset. Plot-level scoring gives an agricultural lender a comparable view of the production capacity supporting each loan. Without that view, the portfolio consists of reported incomes and credit bureau scores that may not correspond to real farm performance.
Detailed scoring also supports repayment schedules tied to the harvest cycle, which is vital to viable agricultural lending. A grower whose coffee cherry harvest comes in June and July cannot make identical monthly payments from January through May. If a lender knows the expected yield and harvest timing for every plot, it can set schedules that match cash flow to debt-service obligations and reduce technical defaults caused by timing differences rather than a true inability to repay.
What Data-Based Underwriting Cannot Promise
Plot-level information makes borrower selection more accurate. It does not remove agricultural risk. Weather shocks, sharp commodity price declines, and disease outbreaks can harm even the highest-scoring farms in a portfolio. A model calibrated with Sul de Minas production data cannot foresee a frost or coffee leaf rust outbreak that destroys canopy across an entire growing zone in one season.
Data provides a more reliable baseline for each farm's capacity under normal conditions. That allows lenders to price risk more accurately and set loan sizes the farm can realistically service. Even a well-scored portfolio still needs sensible concentration limits and macro-weather hedging strategies. The model addresses individual farm selection, while portfolio risk management calls for further discipline.
Model calibration also deserves a direct note. Our scoring model was built and calibrated using Sul de Minas production data from our early-access pilot cohort. As we move into other Minas Gerais growing regions, Cerrado Mineiro and Chapada de Minas differ in altitude profiles, rainfall patterns, and cultivar mixes. Parameters must be calibrated for each sub-region. Applying a model trained entirely on Sul de Minas altitude data to cerrado lowland farms without adjustment could misclassify risk, harming both lenders and growers.
How Lenders Can Use Plot-Level Scoring Today
We work as both a direct lender and a data partner. Credit cooperatives and rural lending institutions in Minas Gerais that want to add plot-level scoring to their underwriting can use our data API and co-origination structure. Culttivo supplies the risk score and data packet, while the cooperative or lender makes the final credit decision under its own criteria.
For cooperatives seeking to lend to members who do not meet current documentation requirements but do have documented cooperative delivery histories, this offers a practical route. Cooperative yield records enter our scoring model, the model returns a risk assessment, and the cooperative can decide with more information than before. The underlying data was already available. What was absent was a model for interpreting it.