₦166.95M Disbursed Across 57 Agricultural Loan Facilities
Supported by the Japan International Cooperation Agency (JICA) and ONDI, Zowasel developed and validated ACESS, converting farm data into credit scores and disbursing ₦166.95M in loans across Cocoa, Maize, and Sorghum value chains.
₦166.95M
Credit Principal Disbursed
Across 57 funded loan facilities
1,965
Farmer Profiles Surveyed
Nasarawa & Ondo core proof of concept
496
Finance-Scored Records
A-F credit risk categorization
46
FFO & Merchant Loans
Cocoa, Sorghum, Maize & Soybeans
60+
AI/ML Engineers Collaborated
13 countries global technical talent
JICA-SUPPORTED ALTERNATIVE CREDIT SCORING | NIGERIA | 2022-2024
Turning agricultural data into finance-ready intelligence
A public-review case study of the Survey on Credit Scoring System for Smallholder Farmers in Nigeria and the development of ACESS
Public review draft. The project delivered a working credit-scoring prototype and supported real lending, but reach, repayment, employment and impact claims require careful separation and reconciliation. No independent outcome or model-performance evaluation was found in the reviewed evidence.
“Collected data, scorable records, funded borrowers and improved livelihoods are four different results.”
Zowasel case-study evidence rule
| 1,965farmer records collectedcore Nasarawa and Ondo survey | 496records met scoring minimum25.2% of the core survey |
|---|---|
| NGN 166.95mprincipal documented11 farmer and 46 FFO loan entries | 60+technical contributorsAI/ML collaboration; not a jobs-created count |
1,965
farmer records collected
core Nasarawa and Ondo survey
496
records met scoring minimum
25.2% of the core survey
NGN 166.95m
principal documented
11 farmer and 46 FFO loan entries
60+
technical contributors
AI/ML collaboration; not a jobs-created count
PROJECT PROFILE
1. Project at a glance#
- Official project name
- The Survey on Credit Scoring System for Smallholder Farmers in Nigeria
- Technology / public solution name
- ACESS - Alternative Credit Evaluation Scoring System
- Lead implementer
- Zowasel.com Limited
- Core public partners
- Japan International Cooperation Agency (JICA), Office for Nigerian Digital Innovation (ONDI), formerly referenced in project records through the NITDA innovation ecosystem, and Mitsubishi Corporation Nigeria Limited
- Finance and delivery collaborators
- VFD Microfinance Bank / VBank and SimpliFi Financial Services; project records also describe technical collaboration with Omdena and Gebeya
- Target customers and beneficiaries
- Smallholder farmers; farmer-facing organisations (FFOs), cooperatives and merchants; financial institutions; agribusinesses seeking verified agricultural borrowers
- Country
- Nigeria
- State / region
- Core survey and proof of concept: Nasarawa and Ondo. Extended financing records also cover Oyo and Kwara.
- Department
- Value Chain Department, working with Product and Technology, Credit Risk, Finance, Agronomy, Sustainability and field teams
- Technology deployed
- ACESS, Alternative Credit Assessment Report (ACAR), Predictive Dynamic System (PDS), Dynamic Variable System (DVS), A-F scoring, dashboards, APIs, KYC, loan origination and management, Zowasel Marketplace and Crop Centre workflows
- Duration
- Extended project reporting period: 4 June 2022 to 29 February 2024. The original public implementation timetable ended in August 2023.
- Runtime
- Approximately 21 months including amendments and close-out
- Budget / documented expenditure
- USD 176,566.73 in the latest project expense workbook: JICA USD 87,872.00 and Zowasel USD 88,694.73. Management-reported; not presented as an independent audit.
- Program Manager
- Jerry Oche, Founder and CEO, Zowasel
- JICA project oversight
- Nao Fuwa is publicly associated with the initiative; project reports also reference JICA team members. Formal Zowasel supervisor field requires management confirmation.
- Last updated
- 28 August 2026
The final report uses 'Agricultural Credit Evaluation Scoring System' once in its acronym list, while the product, Zowasel website and later public sources consistently use 'Alternative Credit Evaluation Scoring System'. This case study uses the current public name.
Executive summary
The project addressed a persistent agricultural-finance constraint: many Nigerian smallholders lack the hard collateral, formal credit history and verified financial records required by conventional lenders. At the same time, banks often lack the agronomic knowledge and low-cost field infrastructure needed to distinguish a viable farming business from an unacceptable credit risk. Zowasel proposed to convert farmer, farm, production, transaction, environmental and community-validated information into a structured credit assessment that lenders could use alongside their own underwriting rules.
With JICA support under the wider Project NINJA collaboration, and with ONDI and Mitsubishi Corporation Nigeria, Zowasel designed and tested ACESS. The system combines an AI-enabled Predictive Dynamic System with a human-configurable Dynamic Variable System, produces an A-F score and an Alternative Credit Assessment Report, and connects assessment to KYC, loan origination, loan management and Marketplace workflows. The project also built dashboards and integration pathways for financial institutions and farmer-facing organisations.
The core survey collected 1,965 farmer records in Nasarawa and Ondo, but only 496 records - 25.2% - met the minimum data requirements for scoring. The final report and public communications also describe an extended pilot in which about 700 farmers and more than 100 farmer-facing organisations across Nasarawa, Ondo, Oyo and Kwara met minimum scoring and analysis criteria. These figures represent different populations and cannot be added. More importantly, meeting an assessment threshold is not the same as receiving finance.
The financing records reviewed document NGN 166,952,358 in principal across 11 individual farmer facilities and 46 FFO or merchant loan entries. Eight of the 11 farmers fully repaid, two were partially repaid or restructured, and one remained outstanding after a reported crop failure. Project records describe all 46 FFO entries as settled, but the aggregate amount-repaid figure is lower than principal and lower than principal plus expected interest. The public claim of 100% FFO repayment should therefore be withheld until the loan-level ledger is reconciled.
The project demonstrates feasibility: an agriculture-specific scoring and workflow prototype was created, field data were scored, financial institutions participated, and real credit was disbursed. It does not yet demonstrate predictive validity, unbiased decisions or farmer-level impact. The reviewed records do not provide an independently validated default model, a control group, a consistent loan-outcome ledger, sex-disaggregated lending results, or verified changes in yield, income, food security, resilience, community welfare or environmental outcomes. Employment evidence is similarly bounded: more than 60 technical contributors from 13 countries participated in an AI/ML collaboration and the inception plan listed 12 core expert roles, but neither figure proves new jobs created.
Strong evidence of problem relevance and technical delivery; moderate evidence of operational effectiveness; promising but unreconciled portfolio evidence; and insufficient evidence of model accuracy, equity or farmer-level impact. ACESS should be presented as a functioning and evolving credit infrastructure - not yet as a proven causal solution.
CONTEXT AND DESIGN
2. Context#
Agriculture contributes materially to Nigeria's economy and livelihoods, yet it receives only a small share of formal bank lending. Project and public records place the share below 5%. The gap is not only a shortage of capital. It is also an information, transaction-cost and risk-translation problem: lenders receive incomplete farm records, face seasonal and climate volatility, and often use collateral-based methods that exclude otherwise productive farmers.
Smallholders commonly operate with informal identities, irregular cash flow, limited bank histories and weak documentation of land, production and sales. Farmer information may be scattered across paper surveys, cooperatives, buyers, government systems and agritech platforms. Women face additional barriers where land and farming activity are recorded under husbands or male household heads, reducing both visibility and eligibility unless the data design actively corrects for this bias.
Zowasel's Marketplace and field network created a practical starting point. Verified crop deliveries, quality tests, transaction histories, payment records, agronomy observations and farm geolocation can act as behavioural and productive-capacity evidence. The JICA-supported survey tested whether those data could be transformed into a repeatable credit assessment and then connected to actual loan decisions.
External context reviewed includes JICA's Project NINJA materials, BusinessDay's 2022 launch coverage, FarmFit Insights Hub's 2025 ACESS case study and IFPRI's 2026 discussion of the promise and governance risks of AI tools in smallholder finance. These sources support relevance and learning context; they do not independently verify project results.
3. Theory of change#
Digitise
Collect consented demographic, farm, production, transaction, agronomy, KYC and environmental data through field, cooperative, Marketplace and digital channels.
Verify
Triangulate records through field revalidation, community or cooperative endorsement, identity checks, transaction evidence and geospatial information.
Score
Use PDS and lender-configurable DVS variables to estimate risk, assign an A-F grade and calculate a capacity-aligned loan recommendation.
Decide and disburse
Provide ACAR reports, dashboards and APIs so regulated financial institutions can complete underwriting, KYC, approval, disbursement and portfolio management.
Learn
Link repayment and production outcomes back to the borrower record, retrain or recalibrate the model, and adjust loan structure to crop and market realities.
Improve inclusion
Over time, reduce reliance on hard collateral, expand responsible agricultural credit, strengthen productive investment and improve farmer livelihoods and resilience.
| CRITICAL ASSUMPTIONS Data must be accurate, consented and sufficiently complete; the scoring model must be calibrated to default outcomes; lenders must retain independent credit governance; loan terms must fit crop cycles; markets or repayment sources must be credible; and excluded groups must not be penalised because their activity is less visible in the data. |
|---|
4. Objectives#
Develop a prototype agricultural credit scoring system that uses alternative farmer, farm, production and transaction data.
Create a digital scorecard, loan recommendation and assessment report that financial institutions can use in underwriting.
Build the supporting database, application, dashboards, APIs, loan-origination and loan-management functionality.
Collect and validate farmer data in Nasarawa and Ondo and identify potential borrowers for a proof of concept.
Secure at least USD 10,000 for the second-phase lending test and disburse real loans through financial partners.
Observe repayment, refine the score and lending workflow, and document operational learning for scale.
Promote collaboration among JICA, Nigerian innovation institutions, a Japanese company, Zowasel, lenders and farmer organisations.
The proposal envisaged about 200 smallholders and a financing mix for input and mechanisation credit. The TOR initially described 130 farmers in Kawo and Doma, while a later beneficiary workbook estimated 160 potential input-credit recipients. These were planning figures, not achieved reach.
DELIVERY MODEL
5. Our approach#
| Workstream | What the project did | Intended value |
|---|---|---|
| Farmer and farm data | Collected demographics, farm size and location, crops, productivity, yield and income information, input and mechanisation needs, agronomic practices, training, identity and bank-readiness data. | Create an agriculture-specific borrower profile beyond conventional bureau history. |
| Transaction intelligence | Used or proposed Marketplace data such as delivery date, crop, quantity, quality, value, buyer and payment history. | Treat verified commercial behaviour as evidence of capacity and willingness to perform. |
| Environmental intelligence | Explored weather, satellite, NDVI/EVI, crop-yield and disease information as risk and production inputs. | Improve crop and climate context around a credit decision. |
| Field and community verification | Used revalidation, cooperative or community checks, crop-centre support and alternative identity pathways where BVN coverage was weak. | Reduce data error, impersonation and missing formal records while building trust. |
| PDS and DVS scoring | Combined a predictive model with configurable variables and weights; assigned A-F grades and produced loan recommendations. | Offer consistency while allowing lenders to align scoring with risk appetite and product rules. |
| ACAR and lender workflow | Generated an Alternative Credit Assessment Report and connected it to KYC, applications, approval, wallets, disbursement and portfolio monitoring. | Reduce underwriting friction and provide a traceable decision record. |
| Pilot lending | Worked with regulated finance partners and Zowasel-linked crop programmes to test individual and FFO or merchant credit. | Observe repayment behaviour and improve the score, loan structure and operating model. |
6. ACESS technology and decision workflow#
| Component | Documented function | Evidence boundary |
|---|---|---|
| Predictive Dynamic System (PDS) | Uses available borrower and agricultural attributes to produce a risk score and A-F grade. | The reviewed records do not provide out-of-sample default calibration, AUC, precision/recall, false approval/decline rates or stability by cohort. |
| Dynamic Variable System (DVS) | Allows approved users to add variables and adjust weights to reflect lender or product requirements. | Flexibility can improve fit but also creates governance and consistency risk unless versions, approvals and overrides are logged. |
| Alternative Credit Assessment Report (ACAR) | Provides farmer profile, score, grade, eligibility and recommended loan parameters. | An ACAR supports - but should not replace - regulated lender underwriting and documented reasons for decisions. |
| A-F grades | Categorises borrower records from strongest to weakest risk profile based on configured attributes. | A grade is not a guarantee of repayment; no A-grade records appeared in the core 496-record score distribution. |
| Loan origination and management | Supports application, KYC, approval, sub-wallets, disbursement, repayment and portfolio views. | The prototype front end and dashboard were not reported as fully complete and require production controls. |
| Crop-yield submodel | An internal Random Forest experiment reported R-squared of 0.9598 in cross-validation for crop-yield prediction. | This is not validation of the credit-default model and should not be used as a headline ACESS accuracy claim. |
| Data verification | Field, cooperative and later three-stage validation practices were used to improve accuracy. | The final report identifies the lack of a fully automated verification pipeline and the need for continuing improvement. |
PERFORMANCE
7. Results#
The project produced both infrastructure and operational evidence. The most defensible public results are the system components delivered, the core data funnel, and the documented principal disbursed. Claims about finance access, repayment and farmer impact require tighter definitions or additional verification.
| Result area | Project position | Evidence judgement |
|---|---|---|
| Prototype and workflow | ACESS scoring, ACAR, dashboards, database, APIs, KYC, loan origination and loan-management components were built and tested. | Delivered as a prototype; the final report calls for continuing improvement and notes incomplete front-end or dashboard elements. |
| Core data collection | 1,965 farmer records were collected in Nasarawa and Ondo. | Strong management evidence of collection; no independent data-quality audit found. |
| Core scorable records | 496 records met minimum data requirements: 272 in Nasarawa and 224 in Ondo. | 25.2% conversion from collected to scorable; a major data-completeness finding. |
| Extended scoring reach | About 700 farmers and more than 100 FFOs across Nasarawa, Ondo, Oyo and Kwara reportedly met minimum scoring and analysis requirements. | Management-reported extended population; not added to the 496 and not described as funded. |
| Direct individual credit | 11 farmers received input credit totalling NGN 429,800. | Documented in the project loan table; small direct-borrower sample. |
| FFO / merchant credit | 46 loan entries totalled NGN 166,522,558 across cocoa, maize/soy and sorghum programmes. | Real portfolio activity; entries may not equal unique organisations or people supported. |
| Total principal | NGN 166,952,358 across the 11 farmer and 46 FFO or merchant entries; public materials call this about USD 250,000. | Use the naira figure. The USD equivalence is project-reported because no conversion date or rate was supplied. |
| Farmer repayment | 8 of 11 fully repaid; 2 were partially repaid or restructured; 1 remained outstanding after crop failure. | Use counts. Percentage labels conflict across the final report. |
| FFO repayment | The report says all 46 entries were settled, but the aggregate amount-repaid total is below principal and expected repayment. | Settlement and 100% repayment claims require loan-level reconciliation before publication. |
| Farmer outcomes | A public media story reports improved yield and school-fee payment for one participant. | A testimonial, not an evaluated outcome. Consent, loan-record linkage and evidence should be confirmed before feature use. |
8. Reach and the finance funnel#
| Stage | Reported result | How this case study interprets it |
|---|---|---|
| Core records collected | 1,965 farmers | Survey reach in Nasarawa and Ondo; not automatically scorable or finance-eligible. |
| Core records scorable | 496 farmers | 25.2% met minimum data requirements. This is the most clearly documented scoring funnel. |
| Core score distribution | A: 0; B: 39; C: 99; D: 255; F: 103 | Most scorable records were in D or F. Score eligibility rules and lender decisions must be reported separately. |
| Extended qualified population | About 700 farmers and more than 100 FFOs | Later multi-state scoring or analysis reach. Population overlap and unique-ID rules are not supplied. |
| Individual farmer facilities | 11 | Direct funded farmer records found in the project lending table. |
| FFO / merchant loan entries | 46 | Funded organisational or merchant entries; not assumed to be 46 unique organisations or a farmer-beneficiary count. |
| Livelihood impact | Not established | No verified denominator, baseline/endline or comparison evidence for income, yield, food security or resilience. |
| RECOMMENDED PUBLIC REACH STATEMENT The core survey collected 1,965 farmer records, of which 496 - 25.2% - contained enough information to be scored. Project records separately report an extended group of about 700 farmers and more than 100 farmer-facing organisations that met minimum assessment criteria. Direct lending records reviewed cover 11 individual farmers and 46 FFO or merchant loan entries. |
|---|
9. Finance, disbursement and repayment#
| Portfolio segment | Entries | Principal | Repayment position |
|---|---|---|---|
| Individual input credit | 11 farmers | NGN 429,800 | 8 fully repaid; 2 partial or restructured; 1 outstanding after reported crop failure. |
| Cocoa FFO / merchant | 3 entries | NGN 69,083,908 | Reported settled; loan-level reconciliation required. |
| Maize and soy FFO / merchant | 2 entries | NGN 34,764,680 | Reported settled; loan-level reconciliation required. |
| Sorghum FFO / merchant | 41 entries | NGN 62,673,970 | Reported settled; loan-level reconciliation required. |
| Total | 57 funded entries | NGN 166,952,358 | Do not interpret entries as 57 unique beneficiaries or as proof of model accuracy. |
The FFO summary lists NGN 166,522,558 principal, NGN 183,174,813.80 expected including interest and NGN 165,309,911.80 repaid, while also describing all 46 entries as settled. The approximately NGN 1.21 million difference between FFO principal and amount repaid must be explained before a 100% repayment claim is approved.
The portfolio pattern is informative even before final reconciliation. The larger FFO loans were often embedded in crop programmes, buyer receivables or organised market relationships. This structure can improve repayment by linking credit to verified production, contracts and cash flows. It also means the observed performance cannot be attributed to ACESS scoring alone. Market linkage, borrower selection, programme controls, guarantees, collections and loan restructuring are plausible co-drivers.
10. Budget and co-investment#
| Budget view | Total | JICA | Zowasel | Reporting decision |
|---|---|---|---|---|
| Latest project expense workbook | USD 176,566.73 | USD 87,872.00 | USD 88,694.73 | Primary management-cost view for this draft; 49.77% JICA and 50.23% Zowasel. |
| Final report finance summary | USD 166,730.36 | USD 87,872.00 | USD 78,858.36 | Lower total; appears to omit later or excluded costs identified in the report. |
| FarmFit 2025 public case study | Not stated | 40% | 60% | Later public characterisation; does not reconcile to the latest project workbook. |
| Latest workbook cost category | Amount | Share of total |
|---|---|---|
| Software engineering | USD 93,403.61 | 52.90% |
| Field personnel | USD 52,915.66 | 29.97% |
| Logistics | USD 30,247.46 | 17.13% |
| Total | USD 176,566.73 | 100.00% |
| BUDGET PUBLICATION RULE Use USD 176,566.73 only as documented management expenditure, not as audited project cost. Before public release, Finance and JICA should confirm the approved cost basis, treatment of late or excluded expenses, and the public co-financing split. |
|---|
MONITORING, WORK AND IMPACT
11. M&E results and evidence quality#
The project maintained inception, monthly, quarterly and final reporting; score distributions; field and revalidation records; loan and repayment tables; model-development notes; and expense records. This provides a meaningful delivery history. It is not yet an evaluation system capable of supporting strong model-performance or farmer-impact claims.
| M&E area | Available evidence | Priority improvement |
|---|---|---|
| Implementation | TOR, RFP, inception report, monthly and quarterly reporting, final report and activity records. | Create one approved results framework with version control, target definitions, owners and dated source files. |
| Data funnel | 1,965 collected and 496 scorable in the core survey; score distributions by state. | Use unique farmer IDs, missing-data reasons and a reproducible collection-to-score-to-decision-to-loan funnel. |
| Portfolio | Loan tables, principal, expected repayment, amount repaid and status narratives. | Reconcile every loan to a unique borrower, disbursement date, product, score version, maturity, restructuring, amount repaid and final status. |
| Model performance | Prototype descriptions and a crop-yield experiment are available. | Report credit-specific out-of-sample validation, probability-of-default calibration, rank ordering, overrides, drift, false approval and false decline. |
| Equity and inclusion | Qualitative notes identify women's land and data-visibility barriers. | Report sex, age, disability and location through every funnel stage; test performance and approval rates for disparate impact. |
| Farmer outcomes | A small number of narrative stories and reported repayment outcomes. | Measure yield, gross margin, net income, productive assets, food security, resilience and borrower protection at baseline and endline. |
| Employment | Technical collaboration, staff plan and cost categories are visible. | Maintain a unique employment register: existing/new, employee/contractor, dates, full-time equivalent, location, sex, age and continuing status. |
| Data protection | KYC, community validation, bureau and alternative data are part of the design; later public material describes consent and data minimisation. | Document lawful basis, purpose limitation, consent, access, corrections, retention, security, model explanations and adverse-action processes. |
| Independent assurance | No independent impact, model, portfolio or fairness evaluation found. | Commission technical validation and an outcome evaluation before using 'proven', 'low risk' or causal impact language. |
Recommended ACESS performance dashboard
Funnel: records collected, consented, KYC-complete, verified, scorable, A-C, lender-reviewed, approved, disbursed, repaid on time, restructured and written off.
Portfolio: unique borrowers, unique FFOs, principal, interest, maturity, arrears, portfolio at risk at 30 and 90 days, recovery, restructuring and loss rate.
Model: score version, probability-of-default calibration, rank ordering, stability, override rate, decision reason and performance by crop, state, sex and borrower type.
Inclusion: sex, age, disability, remoteness, identification status, land-control status, loan size, approval rate and outcome at each stage.
Impact: input use, production, yield, verified sales, gross margin, net income, assets, food security, climate shock, coping strategy and satisfaction.
Economics: cost per collected, verified, scored, approved and repaid borrower; lender processing time; acquisition cost; service revenue and support cost.
12. Employment opportunities and employment created#
ACESS generated technical and field work in AI/ML, full-stack and front-end development, UI/UX, data engineering, agronomy, sustainability, data capture and cleaning, verification, field operations, KYC, credit analysis, loan administration, sales, transaction support and maintenance. The evidence supports participation and financed work, but not a single jobs-created total.
| Employment measure | What the records show | Public-reporting decision |
|---|---|---|
| Technical collaboration | The final report says more than 60 senior and junior AI/ML engineers from 13 countries collaborated for about three months. | Report as technical contributors or collaborators. Do not call these 60 new jobs without contracts, paid status and unique-worker records. |
| Core implementation team | The inception staffing plan lists 12 core expert roles across management, credit, technology, agronomy, data and field delivery. | Report as planned or staffed project roles, not automatically new employment. |
| Paid work categories | Expense records include engineering, field management, agronomy, data capture, agents and external developer support. | Describe the work categories financed. A unique headcount and full-time-equivalent total are not available. |
| Local opportunity pathway | Field verification and crop-centre delivery create potential for trained local verifiers and cooperative support roles. | Describe as an employment opportunity pathway; no indirect-job estimate is made. |
| New employment created | No reconciled employee or contractor register was found. | Not verified. |
| EMPLOYMENT RESULT Verified new jobs created: not available. Documented participation measures: more than 60 technical contributors from 13 countries and a 12-role core staffing plan. These measures may overlap and should remain separate. |
|---|
13. Impacts on people, communities and the planet#
| Dimension | Contribution pathway and evidence | Evidence boundary |
|---|---|---|
| People | The project enabled 11 individual farmers to receive input credit and supported larger FFO or merchant finance linked to agricultural value chains. | No verified average change in yield, income, food security, assets, resilience or well-being. Repayment is not the same as livelihood improvement. |
| Women and excluded groups | Alternative data and community verification can help people without conventional collateral or bureau histories. | No sex-disaggregated scoring, approval, loan, repayment or outcome result. Women's land visibility is an identified design risk. |
| Communities | Crop centres, field agents, cooperatives and community validation can strengthen local information and lender relationships. | No verified community-level welfare, trust, institutional-capacity or spillover result. |
| Financial inclusion | ACESS creates a pathway from agricultural activity and transaction records to formal lender assessment. | Direct individual lending was limited to 11 records; the extended FFO portfolio cannot be treated as unique smallholder access without beneficiary linkage. |
| Planet | Weather, remote sensing, agronomic and sustainability data can improve climate-risk assessment and direct finance toward appropriate practices. | No verified practice adoption, land area, soil, water, biodiversity, carbon or emissions outcome was found. No environmental impact is claimed. |
Impact evidence ladder
| Evidence level | Status | Illustrative project evidence |
|---|---|---|
| Inputs | Documented | JICA and Zowasel expenditure; technical, field and logistics resources. |
| Activities | Documented | System development, data collection, verification, score generation, lender engagement and pilot lending. |
| Outputs | Strong management evidence | ACESS prototype, ACAR, dashboards, APIs and 496 core scorable records. |
| Finance outcome | Partly established | NGN 166.95 million principal documented, but borrower counts and repayment arithmetic require reconciliation. |
| Behaviour and livelihood outcome | Not established | No verified cohort evidence of productive investment, sales, yield, income, resilience or borrower protection. |
| Community and planet impact | Not established | No verified community-level or environmental indicator. |
EVALUATION AND LEARNING
14. Evaluation against OECD-DAC criteria#
Relevance: Strong. The project responds directly to collateral barriers, fragmented farm data, high rural underwriting costs and limited agriculture-specific lender knowledge.
Coherence: Moderately strong. ACESS fits JICA's Project NINJA partnership model and links logically with ONDI, Mitsubishi, financial institutions, Zowasel's Marketplace and Crop Centres. Roles, data sharing and attribution need clearer public definition.
Effectiveness: Moderate. A functioning prototype, scorecard workflow and real disbursement were delivered. Only 25.2% of core farmer records were scorable, direct farmer lending was small and several product components remained prototype-stage.
Efficiency: Mixed. About USD 176,567 in documented co-funded expenditure created reusable infrastructure and supported lending, but low data completeness, repeated field validation and incomplete production readiness increased the cost of conversion to finance.
Impact: Promising but not established. The project enabled finance and generated credible learning. It did not measure attributable changes in livelihoods, inclusion, community welfare or environmental outcomes.
Sustainability: Emerging. ACESS continued beyond the project and was described publicly by FarmFit in 2025. Sustainable scale still depends on model validation, lender adoption, portfolio quality, data governance, unit economics and borrower trust.
15. Failures, shortfalls and adaptations#
| Failure or shortfall | What happened | Learning and adaptation |
|---|---|---|
| Low data completeness | Only 496 of 1,965 core farmer records met minimum scoring requirements. | Design the minimum viable dataset with farmers and lenders, validate at collection, record missing-data reasons and use assisted revalidation before scoring. |
| Trust and survey fatigue | Past failed projects, long questionnaires and repeated visits reduced willingness to answer or disclose prior borrowing. | Use trusted local agents, explain purpose and benefit, minimise questions, give correction rights and avoid collecting data without a clear use. |
| Identity and banking gaps | BVN and bank-account gaps slowed eligibility; alternative IDs and crop-centre support were considered. | Build identity and account-readiness into the service journey, but keep lender KYC and legal requirements explicit. |
| Guarantor difficulty | Potential guarantors and community leaders were reluctant to assume legal or repayment risk. | Separate factual community validation from financial guarantee obligations and explain each role clearly. |
| Geolocation resistance | Some farmers resisted farm coordinates or were difficult to reach over poor roads and long distances. | Use transparent consent, clear benefit explanations, local verification and lower-cost remote checks without overstating precision. |
| Gender visibility | Women's farming activity could be hidden under husbands' land or household records. | Collect individual production and control-of-resource information, set inclusion targets and test scoring and approval outcomes by sex. |
| Security and infrastructure | Insecurity and poor roads constrained visits; JICA could not visit some Doma communities during one mission. | Maintain security triggers, remote-verification fallbacks, local delivery partners and realistic field schedules. |
| Prototype and small data | The model and interface were still evolving; the final report notes small data, incomplete front-end work and no complete verification pipeline. | Treat ACESS as a controlled pilot, version every model, validate on new cohorts and limit automation until performance and explanations are reliable. |
| Climate and crop failure | At least one individual facility remained outstanding after reported crop failure. | Use crop-cycle repayment, insurance or risk-sharing where appropriate, climate indicators, restructuring policy and borrower-protection controls. |
| Repayment evidence conflict | Narrative settlement and percentage claims do not reconcile to aggregate amounts. | Use a loan-level ledger, freeze reporting dates, separate principal and interest and obtain Finance and lender sign-off before publication. |
| Cost and FX pressure | Inflation, exchange-rate movement and foreign development costs affected delivery. | Use local technical capacity where feasible, staged releases, contingency budgets and transparent exchange-rate assumptions. |
16. What the project learned#
The first credit-scoring problem is often data completeness, not algorithm choice. A sophisticated model cannot repair missing or unverified borrower information.
Field validation and community knowledge are valuable, but they must be governed so social bias, coercion or informal guarantees do not replace fair underwriting.
Agricultural credit works better when loan timing, maturity, repayment source and collection are designed around crop and market cycles.
Repayment from organised, buyer-linked FFO programmes may reflect the strength of the transaction structure as much as the score. ACESS must be evaluated within the whole credit process.
A configurable scoring system needs stricter version control, approval and audit trails - not weaker governance - because changing variables can change who is approved.
Women and farmers with weak formal documentation can be systematically under-scored unless data collection measures their own productive activity and control of resources.
A pilot must include enough diversity and adverse outcomes to test discrimination between risk groups. Very small or heavily selected loan cohorts cannot prove predictive validity.
Public reporting should use counts and reconciled monetary totals before percentages, especially when restructuring, interest, settlement and default definitions differ.
Farmer consent, data minimisation, correction rights and explainable adverse decisions are core infrastructure, not compliance added after scale.
SUSTAINABILITY AND CONTRIBUTION
17. Sustainability and scale#
ACESS continued after the JICA-supported project. A 2025 FarmFit Insights Hub case study describes the system as an evolving data and credit infrastructure that converts production, delivery and transaction histories into finance-ready intelligence, uses field triangulation and community validation, and allows lender-specific variables. Zowasel and VBank also communicated a USD 100,000 credit partnership. These developments show continuity and market interest; they are not retrospective project results.
Current strategic documents describe possible ACESS 2.0 and public-infrastructure pathways. This case study does not reproduce their forecasts or commercial targets. Future scale claims should be reported as a separate phase with new baselines, governance, portfolio evidence and outcome evaluation.
| Scale condition | Why it matters | Required evidence |
|---|---|---|
| Independent model validation | Lenders and borrowers need evidence that the score predicts risk and remains stable. | Out-of-sample results, calibration, discrimination, drift, override and stress testing by crop, location and borrower type. |
| Fair and explainable decisions | Alternative data can include or exclude depending on what is observed. | Sex and group fairness tests, reason codes, adverse-action notices, appeals, corrections and human review. |
| Reliable data governance | ACESS depends on sensitive identity, farm, geospatial, transaction and financial information. | Consent, purpose, minimisation, retention, access, security, audit, deletion and partner-data agreements. |
| Portfolio quality | Scoring value is demonstrated through consistent performance across seasons and products. | Unique loan ledger, arrears, portfolio at risk, recovery, restructuring and loss by score band and cohort. |
| Lender integration | A score must fit regulated underwriting, KYC, decision and collections processes. | Production APIs, service levels, security review, decision ownership and documented lender adoption. |
| Borrower value and protection | Access is beneficial only when credit is affordable, timely and productively used. | Pricing, maturity, repeat borrowing, complaints, over-indebtedness, outcomes and user comprehension. |
| Sustainable economics | Field verification and support are costly. | Cost per verified, approved and repaid borrower; revenue; support cost; partner willingness to pay and long-run margin. |
18. Contribution to the Sustainable Development Goals#
| SDG | Contribution pathway | Evidence status |
|---|---|---|
| SDG 1 - No Poverty | Responsible productive credit may strengthen farm livelihoods and reduce exclusion. | Direct lending documented; poverty and livelihood change not measured. |
| SDG 2 - Zero Hunger | Input and mechanisation finance can support production and food availability. | No verified yield, production or food-security result. |
| SDG 5 - Gender Equality | Alternative data can widen access for women excluded by conventional collateral systems. | No sex-disaggregated lending or outcome result; land and data visibility are material risks. |
| SDG 8 - Decent Work and Economic Growth | Technical, field and finance operations create work pathways and may support agricultural enterprises. | Contributors and roles documented; new or sustained employment not verified. |
| SDG 9 - Industry, Innovation and Infrastructure | ACESS is digital infrastructure connecting agricultural data and financial decisions. | Prototype delivery documented; production performance and adoption require validation. |
| SDG 10 - Reduced Inequalities | Non-collateral assessment can include borrowers outside formal credit histories. | Approval and outcome equity by group not measured. |
| SDG 13 - Climate Action | Weather, geospatial and sustainability data can improve climate-risk assessment and appropriate finance. | No verified resilience, practice, land or emissions result. |
| SDG 17 - Partnerships | JICA, ONDI, Mitsubishi, Zowasel, lenders, technical collaborators and farmer organisations contributed complementary capabilities. | Partnership is visible; public naming, branding and attribution require clearance. |
PUBLIC NARRATIVE AND EVIDENCE
19. Public visibility and media insights#
Public communications describe ACESS as a bridge between financial institutions and farmer-facing organisations. JICA procurement records establish the contracted survey, while JICA's later Project NINJA and KAIZEN materials place the Mitsubishi-Zowasel collaboration within a wider startup-ecosystem programme. BusinessDay reported the 2022 launch and the public objective of improving smallholder finance, financial knowledge and access to banking services.
Zowasel's August 2023 Medium article explains PDS, DVS, A-F grading, ACAR and the intended lender workflow. BusinessDay's February 2024 feature repeats the extended reach and approximately USD 250,000 lending narrative and includes a named farmer story. Zowasel and founder LinkedIn posts document launch, release and VBank partnership visibility. These are valuable public records, but several originate from Zowasel or partner publicity and are not independent evaluation.
FarmFit Insights Hub's December 2025 case study is the strongest later external narrative on ACESS continuity, data triangulation and governance. It also reports a 60% Zowasel / 40% JICA co-funding split that differs from the latest project expense workbook. IFPRI's 2026 discussion of AI in smallholder finance provides an important independent caution: incomplete data, gender exclusion, reliability and governance can determine whether AI tools expand or restrict access.
JICA procurement disclosure: Survey on Credit Scoring System for Smallholder Farmers in Nigeria — official contract listing
JICA: Project NINJA - Empowering and Connecting Africa's Entrepreneurs and Tech Start-ups — official programme context
JICA: KAIZEN in Startups — official material identifying Mitsubishi x Zowasel in Nigeria
BusinessDay: JICA, ONDI launch credit-scoring project for smallholder farmers — public launch coverage, 11 July 2022
Zowasel Medium: Introducing ACESS — product and workflow description, 29 August 2023
Jerry Oche LinkedIn: ACESS release — public partner acknowledgement and launch narrative
Zowasel LinkedIn: VBank USD 100,000 credit partnership — public continuation and lender-partnership evidence
BusinessDay: How Zowasel links farmers with finance through ACESS — extended-pilot and public beneficiary narrative, 20 February 2024
TechCabal partner release: Zowasel releases ACESS — public release and media visibility
FarmFit Insights Hub: How to turn agricultural data into financial access — later external case study, 3 December 2025
IFPRI: The emerging role of AI tools in smallholder finance — independent methodological and governance context, 2026
Zowasel alternative finance and ACESS — current public product page
Zowasel Medium publication — wider company communications reviewed
Zowasel LinkedIn company page — wider public communications reviewed
20. Claim reconciliation for public release#
| Claim area | Sources differ or evidence is incomplete | Case-study decision |
|---|---|---|
| Project dates | Original TOR and public launch describe June 2022-August 2023; the final report covers 4 June 2022-29 February 2024 and contract amendments exist. | Use the extended final-report period and disclose the original timetable. |
| ACESS name | One final-report acronym entry says Agricultural; public and current sources say Alternative. | Use Alternative Credit Evaluation Scoring System. |
| Core and extended reach | 1,965 collected, 496 core scorable, about 700 extended qualified and more than 100 FFOs are reported without a common unique-ID rule. | Keep every funnel stage separate; do not add populations. |
| Access to finance | Public wording can imply about 700 farmers accessed finance; loan records show 11 individual farmer facilities plus 46 FFO entries. | Use 'met scoring requirements' for reach and 'received finance' only for documented funded records. |
| Disbursement | Naira principal totals NGN 166.95 million; public materials say about USD 250,000 without an exchange-rate method. | Lead with naira principal and label USD as project-reported. |
| Individual repayment | Counts support 8 full, 2 partial/restructured and 1 outstanding, while percentage labels vary. | Use counts, not conflicting percentages. |
| FFO repayment | All 46 are described as settled, but amount repaid is below principal and expected repayment. | Do not publish 100% repayment until the loan-level ledger is reconciled. |
| Budget | Latest workbook: USD 176,566.73 and near 50/50 split; final report: USD 166,730.36; FarmFit: 60/40 split. | Use the latest workbook as management-reported and disclose the discrepancy pending Finance/JICA confirmation. |
| Model performance | A crop-yield R-squared is reported, but credit-default validation metrics are not. | Do not use the crop-yield metric as ACESS credit accuracy. |
| Employment | 60+ technical contributors, 12 planned core roles and paid work categories are visible. | Keep participation and role measures separate; new jobs created remain unverified. |
| Impact | Public storytelling describes increased yield and school-fee payment for one farmer. | Treat as a testimonial candidate pending consent and verification; do not generalise. |
| Current scale | Later FarmFit and ACESS 2.0 materials describe continued development. | Report as post-project continuation, not as a JICA project result. |
21. Priority actions before publication#
Obtain Finance and JICA confirmation of the final project dates, expenditure total, co-financing split and treatment of excluded or late expenses.
Reissue the farmer and FFO funnel using unique IDs and dated definitions for collected, verified, scorable, eligible, approved, disbursed and repaid.
Reconcile the loan-level ledger to principal, expected interest, amount repaid, restructuring and final status; document the exchange rate used for any USD equivalence.
Commission independent credit-model validation and fairness testing before using 'proven', 'accurate', 'low risk' or similar performance language.
Add sex, age, disability, location and land-control measures through scoring, approval, disbursement, repayment and outcome stages.
Verify any farmer name, photograph, testimonial and outcome against consent records and the relevant anonymised loan record before website use.
Create a unique employment register that distinguishes new hires, existing staff, paid contractors, volunteers or challenge participants, short-term roles and continuing jobs.
Confirm the formal Zowasel supervisor and partner focal points and obtain written clearance for names, logos, budget, repayment and attribution language.
Publish a short methodology note with the case study explaining source hierarchy, management-reported figures, limitations and the date through which evidence was reviewed.
Management questions to close
| Question | Why it matters |
|---|---|
| Which amended contract date and approved cost statement should be treated as final? | The project period and co-funding split differ across records and later public reporting. |
| Are the 46 FFO entries unique organisations, repeated facilities or merchant transactions? | This determines the correct beneficiary and portfolio denominator. |
| What explains the difference between FFO principal and amount repaid when all entries are described as settled? | A public repayment claim cannot be approved without this reconciliation. |
| Can the 700 farmers and 100+ FFO records be mapped to the 496 core records and the funded facilities? | This is needed for one defensible reach-to-finance funnel. |
| What credit-specific validation and fairness results exist for each model version? | A credit score requires evidence of performance, stability and equitable treatment. |
| Do project participant records include sex-disaggregated approval, loan and repayment data? | Reach intent is not enough to report women's financial inclusion. |
| Which staff and contributors were newly employed, paid, part-time, retained or volunteer collaborators? | This is needed to report employment created without double counting. |
| Is documented informed consent available for the named farmer story and all public images? | Public disclosure should protect borrowers and meet privacy and dignity standards. |
Key internal source links#
Zowasel Credit Scoring Final Report, 4 June 2022-29 February 2024 — primary close-out design, implementation, reach, scoring and lending source
JICA TOR - Credit Scoring — official scope, phases, deliverables and responsibility boundaries
JICA RFP - Credit Scoring — procurement and proposal requirements
Project inception report — workplan, staffing, assumptions, risks and methodology
ACESS project press release — project-authored public narrative and extended reach claims
Updated project expense workbook — documented cost categories and JICA/Zowasel contribution figures
Expected beneficiaries for farm input credit — planning assumptions for NGN 10 million and 160 potential beneficiaries
Credit-scoring and fund-disbursement calendar — timing and sequencing of scoring and finance activity
Evidence base and source register#
Evidence hierarchy used for this case study: signed agreements and approved budgets; dated project M&E/final reports; implementation records; then public communications. Where figures conflict, the most project-specific and most recent verified source is used and the conflict is disclosed.
Internal project records reviewed
Zowasel Credit Scoring Final Report (June 4, 2022-Feb 29, 2024) — primary close-out report; architecture, field work, score distributions, lending, repayment, challenges, learning and limitations.
JICA TOR and RFP - Credit Scoring — authoritative project purpose, phases, deliverables, lending test and responsibility boundaries.
Project inception report — methodology, core locations, team plan, timeline, risk assumptions and delivery approach.
Monthly and quarterly reports — implementation progress, field activity, data collection, validation, technology development and issues.
Updated project expense workbook — documented expenditure and contribution figures; management record, not independent audit.
Expected beneficiary and disbursement-planning records — planned loan sizes, locations, beneficiary assumptions and schedule.
ACESS project press release and media materials — public project narrative, extended reach, financing and testimonial claims.
Score distribution and loan tables in final reporting — core scoring funnel and principal/repayment evidence used in this case study.
Contract and amendment folder — legal project records reviewed for existence; several scans were not machine-readable and are not used for unsupported detail.
Later ACESS strategy and 2.0 materials — continuation and future direction only; not used as JICA project results.
Public sources and media reviewed
JICA procurement disclosure: Survey on Credit Scoring System for Smallholder Farmers in Nigeria — official contract listing
JICA: Project NINJA - Empowering and Connecting Africa's Entrepreneurs and Tech Start-ups — official programme context
JICA: KAIZEN in Startups — official material identifying Mitsubishi x Zowasel in Nigeria
BusinessDay: JICA, ONDI launch credit-scoring project for smallholder farmers — public launch coverage, 11 July 2022
Zowasel Medium: Introducing ACESS — product and workflow description, 29 August 2023
Jerry Oche LinkedIn: ACESS release — public partner acknowledgement and launch narrative
Zowasel LinkedIn: VBank USD 100,000 credit partnership — public continuation and lender-partnership evidence
BusinessDay: How Zowasel links farmers with finance through ACESS — extended-pilot and public beneficiary narrative, 20 February 2024
TechCabal partner release: Zowasel releases ACESS — public release and media visibility
FarmFit Insights Hub: How to turn agricultural data into financial access — later external case study, 3 December 2025
IFPRI: The emerging role of AI tools in smallholder finance — independent methodological and governance context, 2026
Zowasel alternative finance and ACESS — current public product page
Zowasel Medium publication — wider company communications reviewed
Zowasel LinkedIn company page — wider public communications reviewed
Further project information#
| Contact | Public information | |
|---|---|---|
| Zowasel partnerships | partners@zowasel.com | https://www.zowasel.com/ |
| Zowasel public product information | https://zowasel.com/alternative.html | |
| JICA | https://www.jica.go.jp/english/ | |
| Project geography | Nigeria; core work in Nasarawa and Ondo, with extended finance records in Oyo and Kwara |
Zowasel public product information
https://zowasel.com/alternative.html
JICA
https://www.jica.go.jp/english/
Project geography
Nigeria; core work in Nasarawa and Ondo, with extended finance records in Oyo and Kwara
| PUBLICATION CONTROL Before public PDF release: (1) reconcile dates, budget and co-finance; (2) approve the reach-to-finance funnel; (3) reconcile repayment to a loan-level ledger; (4) approve a credit-model validation and fairness statement; (5) confirm partner names, logos and attribution; (6) verify consent for every named farmer, image and quote; (7) remove borrower identities, coordinates, KYC, bank, loan and account data; and (8) keep the report labelled outcome-unverified until an appropriate evaluation is complete. |
|---|
Suggested public download gate: collect name, organisation, role, work email, country, area of interest and explicit consent for follow-up. Link the form to the approved public PDF - never to internal Drive folders - and provide a privacy notice, purpose, retention period and opt-out route.
