Unlock Smallholder Lending Without the Default Risk
Field-level credit risk scoring for banks, microfinance institutions, and agricultural finance providers serving Zimbabwe and Southern Africa.
You're mandated to lend to smallholders, but you can't price the risk.
Agricultural lending to smallholder farmers in Zimbabwe has historically operated with 30-50% default rates — economically unviable for any institution operating at scale. The result: a billion-dollar lending gap, regulatory pressure on banks to address it, and no credible way to underwrite individual borrowers.
Traditional risk scoring fails because it depends on credit history that smallholders don't have, formal income documentation that doesn't exist, and collateral structures that don't apply to land tenure realities in Zimbabwe.
The information that does predict smallholder repayment — what they grow, how their fields perform, their yield history, their resilience to weather shocks — has historically been impossible to gather at scale.
What We Provide
A field-level credit risk score, integrated into your loan origination workflow
KurimaSense's risk scoring layer transforms agricultural lending from instinct-driven to data-driven, while remaining transparent and explainable — critical for credit committee approval and regulatory compliance.
How we work together
Engagements scale with the depth of modelling you need — from standard reporting through to bespoke prediction models — not by seat count. We scope and price each one on a short briefing call.
Core
Risk Scoring API
On-demand field-level risk scoring via API, with primary risk factors and confidence indicators.
- REST API for loan origination integration
- Default probability per field
- Top 3 contributing risk factors
- Confidence indicator per score
- Audit trail for compliance
Standard
White-Label Dashboard
A branded interface for your loan officers with field-level drill-down and portfolio analytics, on top of the Risk Scoring API.
- Everything in Core
- Branded dashboard with your visual identity
- Loan officer field-level drill-down
- Portfolio risk distribution & concentration analysis
- Borrower risk reports (PDF)
Enterprise
Custom Risk Models
Default-prediction models tuned to your specific loan products, repayment terms, and regional concentration.
- Everything in Standard
- Custom default-prediction model tuning
- Calibration to your loan products & terms
- Historical portfolio backtesting
- Quarterly model performance reviews
- Dedicated solution architect
How it works
Borrower data ingestion
Your loan applicants provide their field location (lat/lng or a polygon drawn on a map) and crop type. That's the input. Everything else is derived.
Multi-year satellite history
We pull 5+ years of satellite indices for each field — NDVI, EVI, NDRE, NDMI from Sentinel-2 optical; VV/VH backscatter from Sentinel-1 SAR. This is the field's credit history in agricultural terms.
Ground truth comparison
We compare field performance against benchmarks for the same crop in the same Natural Region. A field consistently underperforming its peers carries higher risk.
Weather and climate context
Drought frequency, rainfall reliability, and seasonal anomalies in the borrower's geography are factored in.
Default probability output
A 0–1 score representing default probability under standard agricultural loan terms, alongside the top three contributing risk factors.
What you receive
- Risk scoring API integrated with your loan origination system
- White-label dashboard for loan officers and underwriters
- Portfolio-level risk distribution and concentration analysis
- Per-borrower risk report (PDF) with explainable factors
- Anomaly alerts on existing portfolio
- Compliance documentation supporting credit committee approval
- Quarterly model performance reports
We don't ship black-box models
Bank credit decisions require explainability. Every risk score KurimaSense produces is decomposable into its contributing factors. Our underlying model is logistic regression at the production layer — calibrated, interpretable, defensible in credit committee, and compliant with regulatory expectations for transparent decision-making.
We do not use neural networks or other opaque models in the production risk scoring pipeline. We can, and do, use more sophisticated models in research — but only their outputs as features, never as the deciding layer.
Why this works where others haven't
Generic credit scoring systems fail in African agriculture because they're calibrated on developed-market borrower data. African mobile money lending models (the M-Shwari approach) work for short-term consumer credit but don't address agricultural cycle risk.
KurimaSense's risk model is purpose-built for the specific reality of smallholder agriculture in Southern Africa: seasonal cash flows, weather-dependent income, no formal credit history, but a wealth of measurable agricultural signal in satellite data that nobody has previously aggregated and modeled at this resolution.
Read our methodology
An 18-page technical document covering the data sources, modeling approach, validation methodology, and compliance considerations.
Download PDFFrequently asked
Schedule a technical demo
For your agribusiness head and risk officers. We'll walk through real field-level risk scoring on borrowers from your geography.
Book a Demo