Digital Lending Platform Launch: Automated Credit Scoring for Cambodian MFIs

Cambodia's microfinance sector faces a structural contradiction. The market is enormous: USD 16.2 billion in outstanding loans across 2.6 million borrowers, with the NBC projecting continued growth as financial inclusion expands into rural areas. Yet the operational infrastructure supporting this market remains largely manual, creating bottlenecks that limit growth, increase costs, and contribute to credit risk. The typical loan origination process at a Cambodian MFI involves six to eight distinct steps, each requiring human intervention. A field officer visits the prospective borrower's home or business, collects copies of national ID cards and household registration documents, photographs the business premises, interviews the borrower about income and expenses, and submits the application package to a branch office. A credit officer then reviews the package, may request additional documentation, checks the Cambodia Credit Bureau for existing obligations, and presents the application to a credit committee for approval. This process takes an average of seven to fourteen business days from initial contact to loan disbursement. For the borrower, this delay means lost business opportunities, as the working capital need that prompted the loan application may have passed by the time funds arrive. For the MFI, the cost is equally significant: the Cambodia Microfinance Association estimates that origination costs consume USD 85-120 per loan, eroding margins on the small-ticket loans (average USD 3,800) that constitute the majority of the portfolio. The credit data gap compounds these challenges. Only 23% of Cambodian adults have a credit history with the Cambodia Credit Bureau. For the remaining 77%, MFIs must rely entirely on field officer assessments, a subjective and inconsistent evaluation method that contributes to non-performing loan rates averaging 2.8% across the sector.

CamFinTech's approach was built on a key insight: Cambodia's government digital infrastructure, specifically CamDigiKey and CamInvoice, contains data that can serve as inputs to credit decisioning, effectively converting compliance infrastructure into lending infrastructure. CamDigiKey, Cambodia's national digital identity system, provides verified identity data that replaces manual document collection. When a loan applicant authenticates via CamDigiKey, the MFI receives a cryptographically signed identity payload that is more reliable than manually inspected paper documents and can be processed in seconds rather than days. CamInvoice, the GDT's mandatory electronic invoicing system, generates transaction records that serve as a proxy for business revenue and health. For SME borrowers, CamInvoice data reveals monthly invoice volumes, average transaction sizes, customer concentration, seasonal patterns, and VAT compliance history. This data, when properly analyzed, provides a more objective and comprehensive view of business performance than the income declarations collected by field officers. The platform combines these government data sources with additional alternative data streams including mobile money transaction patterns from Bakong-connected wallets, utility payment history, and any available Cambodia Credit Bureau records to generate a composite credit score. The scoring model was trained on historical loan performance data from participating MFIs, with the CamInvoice features contributing the strongest predictive signal for SME loan performance.

The platform architecture follows a pipeline model with five stages: identity verification, data aggregation, credit scoring, decision engine, and disbursement orchestration. The identity verification stage integrates with the CamDigiKey API to authenticate applicants via biometric verification on their mobile device. The system receives a signed identity assertion containing verified personal data, which is automatically populated into the loan application form. This stage reduces application completion time from an average of 45 minutes (manual) to under three minutes (digital). The data aggregation stage concurrently pulls data from all available sources. With the applicant's consent, the system queries the CamInvoice API for the applicant's TIN-linked transaction history, retrieves credit bureau records, and ingests mobile money transaction data through bank API connections. All data is normalized into a standardized applicant profile within a privacy-preserving data pipeline that encrypts personally identifiable information at rest and enforces strict access controls. The credit scoring stage applies a gradient-boosted decision tree model trained on 180,000 historical loan outcomes from participating MFIs. The model generates a composite score from 300 to 850, along with factor codes explaining the primary drivers of the score. For thin-file applicants with no credit bureau history, the model automatically increases the weight of CamInvoice and mobile money features, providing a score even when traditional credit data is absent. The decision engine maps the composite score to each MFI's specific credit policies, determining approval, conditional approval, or decline, along with recommended loan amount, tenor, and interest rate. Each MFI configures its own policy rules, risk appetite thresholds, and product parameters within the platform.

CamFinTech executed the implementation in three phases: model development and backtesting, controlled pilot with live lending, and scaled rollout across the MFI network. During model development, CamFinTech worked with five participating MFIs to obtain anonymized historical loan data spanning 180,000 loans disbursed between 2020 and 2024. This data included loan performance outcomes (performing, delinquent, default, written off), borrower demographics, loan product characteristics, and field officer assessments. CamFinTech enriched this dataset by matching borrower TINs to CamInvoice records where available, creating a training dataset that combined traditional lending outcomes with digital invoice data. The model was validated using time-series cross-validation to prevent look-ahead bias. The backtesting results showed that the model's predicted default probability had a Gini coefficient of 0.62 across the full population and 0.58 for thin-file applicants, compared to 0.45 for the MFIs' existing field officer assessment methodology. The improvement was most pronounced for SME loans where CamInvoice data was available, with the Gini coefficient reaching 0.71. The controlled pilot launched with 2,500 loan applications processed through the platform over a 90-day period. During this phase, both the automated scoring and the traditional field officer assessment were conducted for each application, with the credit committee receiving both inputs. This dual-track approach allowed performance comparison while managing risk during the transition. After the pilot confirmed that automated scoring matched or exceeded field officer accuracy while dramatically reducing processing time, the participating MFIs began routing applications exclusively through the platform.

The platform delivered transformative results across loan processing efficiency, credit quality, and operational cost metrics. Loan approval time collapsed from an average of 10.5 business days under the manual process to 4.2 hours for digitally processed applications. For applicants with CamDigiKey verification and CamInvoice data, the end-to-end time from application submission to fund disbursement averaged 2.8 hours, with some applications completing in under one hour. This speed improvement was not achieved by lowering credit standards; rather, it resulted from eliminating manual bottlenecks in identity verification, document collection, and data analysis. Credit quality improved significantly. Loans originated through the automated scoring platform showed a 40% reduction in 30-day delinquency rates compared to the historical baseline of manually originated loans at the same MFIs. The improvement was driven by two factors: the scoring model's ability to identify risk patterns invisible to field officers (particularly revenue seasonality detected in CamInvoice data), and the removal of subjective bias from the assessment process. Origination costs fell by 62%, from an average of USD 97 per loan to USD 37 per loan. The reduction came from eliminating most field visits for identity verification (replaced by CamDigiKey), automating income verification (replaced by CamInvoice data analysis), and reducing credit committee review to exception handling rather than reviewing every application. Field officers were redeployed from documentation collection to relationship management and portfolio monitoring, improving their productivity and job satisfaction.

The most consequential lesson was about data coverage and the cold-start problem. While CamDigiKey and CamInvoice data provided strong predictive signals, their coverage of the MFI borrower population was initially limited. CamDigiKey had 28% adult penetration and CamInvoice adoption was concentrated among urban, VAT-registered businesses. Rural borrowers and informal-sector workers, who constitute a significant portion of MFI clients, often had neither CamDigiKey credentials nor CamInvoice history. For this population, the platform fell back to mobile money transaction data and utility payment patterns, which have broader coverage but weaker predictive power. CamFinTech addressed this by implementing a hybrid model that routes thin-data applicants through a simplified digital process supplemented by a streamlined field verification, reducing but not eliminating the manual component. As CamDigiKey and CamInvoice adoption expands, the fully automated path will cover an increasing share of applicants. A second lesson involved MFI organizational change management. The technology platform was ready within four months, but the organizational adaptation required to use it effectively took considerably longer. Field officers initially perceived the automated scoring as a threat to their roles. CamFinTech worked with MFI management to redefine field officer responsibilities around relationship management, portfolio quality monitoring, and financial literacy education, roles that add more value than document collection. Finally, the regulatory environment required careful navigation. The NBC's regulations on credit scoring, data privacy, and digital lending are evolving as Cambodia's financial sector digitizes. CamFinTech maintained ongoing dialogue with the NBC's FinTech supervision team to ensure the platform's use of alternative data sources, automated decisioning, and CamDigiKey integration complied with both existing regulations and anticipated regulatory developments.

The digital lending platform demonstrates how Cambodia's government digital infrastructure can be repurposed to solve financial inclusion challenges. As CamDigiKey penetration grows toward the government's target of 70% adult coverage by 2028, and CamInvoice adoption becomes universal among VAT-registered businesses, the data foundation for automated credit scoring will strengthen dramatically. CamFinTech is expanding the platform in three directions. First, integrating CamDX (Cambodia Data Exchange) data feeds as they become available, which will provide additional government-verified data points including business registration status, land title records, and social security contributions. Second, developing scoring models specifically optimized for agricultural lending, incorporating satellite imagery data and crop cycle patterns to assess farming household creditworthiness. Third, building a loan origination marketplace that connects pre-scored borrowers with MFI products matching their risk profile and loan requirements, reducing search costs for both borrowers and lenders. For investors and MFI operators considering Cambodia's digital lending opportunity, the convergence of mandatory digital infrastructure, growing alternative data availability, and an NBC regulatory framework that supports responsible innovation creates conditions for rapid scaling of automated lending platforms.