Fiji's FDB pilots AI credit assessment to speed lending and widen access
Fiji has moved on a practical use of AI in banking. The Fiji Development Bank (FDB) has launched an AI credit assessment pilot that Finance Minister Esrom Immanuel says will accelerate lending decisions, widen financial inclusion, and modernise core processes.
Speaking in parliament, Immanuel confirmed that existing FDB customers can now apply fully online through the bank's mobile app for loans up to FJ$100,000, with unsecured options up to FJ$10,000. The system generates instant recommendations for FDB staff, who keep full authority over final decisions.
How the pilot works
- Online applications for existing customers up to FJ$100,000; unsecured up to FJ$10,000.
- Digital identity verification and document upload. All data is encrypted.
- Hosting on Amazon Web Services in Sydney, with data residency set to Australia and governed under Fijian law. AWS Australia (Sydney) Region
- AI produces an immediate risk view for credit officers; staff remain accountable for approvals. Over 100 customers have registered interest so far.
Data protection and vendor controls
Assistant Minister Sakiusa Tubuna pressed for details on data security and vendor use of sensitive information. Immanuel outlined legally binding safeguards with the provider, DROC:
- Mandatory data residency in Australia on secure AWS infrastructure.
- Security standards equivalent to ISO 27001 and SOC 2, plus the right for FDB to run annual security audits.
- 24-hour breach notification and a full incident report within 48 hours.
- Strict prohibition on using FDB data for anything beyond the contracted service.
- Data must be deleted or returned when services end.
Regulatory track
Opposition MP Faiyaz Koya questioned the lack of updated laws for data privacy and AI-led financial services. Immanuel said new legal frameworks are in the pipeline under Fiji's National Digital Strategy to provide oversight of emerging technologies.
Why this matters for finance leaders
- Cycle time: Faster "time-to-yes" without cutting corners if the human-in-the-loop remains strong.
- Unit economics: Digital intake and instant triage can lower acquisition costs, especially for sub-FJ$10k tickets.
- Portfolio mix: A digital channel can widen reach to thin-file and MSME segments while keeping controls intact.
- Claims vs. controls: Data residency, encryption, and audit rights are promising-verify them in practice with ongoing testing.
Model risk and control expectations
- Governance: A documented AI credit policy with clear thresholds for auto-approve, auto-decline, and manual review.
- Validation: Independent testing, challenger models, back-testing, and stability monitoring across segments.
- Explainability: Credit officers need clear reasons for recommendations and a simple way to document overrides.
- Drift and quality: Alerts for data drift, missing fields, and out-of-distribution applications; defined rollback steps.
- Fair access: Track approval, pricing, and delinquency by segment to catch proxy bias early.
- Compliance: KYC, AML/CTF, consent management, and complete audit trails across all decision points.
- Vendor risk: Enforce audit rights, penetration tests, incident drills, and a clean exit plan with data return/deletion.
Metrics to track from day one
- Application-to-decision time (median and 90th percentile).
- Auto-decision rate, override rate, and reasons for overrides.
- Approval, PD, LGD, and NPL by product, ticket size, and segment.
- Model drift indicators and data quality failure rates.
- Acquisition cost and break-even for loans under FJ$10k.
Practical next steps
- Stand up a cross-functional committee (risk, compliance, IT, credit) for weekly pilot reviews.
- Run a limited production pilot with shadow approvals before scaling.
- Set clear exception handling rules and keep a manual fallback if the model or vendor is unavailable.
- Train credit teams on reading model outputs and documenting decisions. For structured upskilling, see curated AI tools for finance.
This move shows momentum in applying AI to core credit workflows while keeping people accountable. If controls hold and regulation lands on time, lenders can capture speed and reach without giving up risk discipline.
Your membership also unlocks:
