From paper files to automated underwriting on a single lending platform

The client is a microfinance lender whose loan officers work in the field, sitting across from shopkeepers and small business owners at their places of business. Its lending model depends on that closeness. Its systems did not support it. Applications travelled as paper forms and photocopied documents. Approvals rested on the judgment of individual officers. Repayments were reconciled in spreadsheets long after the money had moved.

The approach had worked at a smaller scale. As the loan book grew, it stopped working. Each branch ran the lifecycle its own way, staff re-entered the same data at every stage, and the risk team had no way to compare decisions across the book. The lender was scaling manual work, not lending.

Kentro was engaged to build a digital lending platform end to end: a field app for officers, a back-office console for credit and collections teams, and a decision layer covering scoring, underwriting, disbursement and repayment. The brief carried a hard requirement. The risk team had to own the decision logic without asking developers to change code.

Every stage of the loan lifecycle depended on manual work, and each branch ran it differently. The problems compounded from origination through collections.

• Paper-based origination: Applications moved as paper forms and photocopied documents. Every loan depended on branch staff re-entering the same data, and every re-entry was a chance to introduce an error or a delay.
• Judgment-call approvals: Credit decisions rested on individual officer judgment with no shared scoring. Risk could not be compared across branches or officers, and lending policy existed mostly as habit.
• Disconnected money movement: Disbursements were keyed into banking portals by hand. Repayments were reconciled in spreadsheets long after the money moved, so the lender's view of its own book was always behind reality.

Kentro built an end-to-end lending platform that carries a single loan record from first application to final repayment. Node.js services handle origination, scoring, underwriting and loan servicing, with PostgreSQL as the system of record and Apache Kafka as the event backbone, all running on Kubernetes. The novelty sits in the decision logic, not the plumbing.

• Origination in the field: A Flutter app lets loan officers capture applications, KYC documents and guarantor details at the customer's place of business. Offline-first capture syncs when connectivity returns, so weak coverage does not stop a visit.
• A scoring engine the risk team owns: A configurable scorecard service combines bureau checks, repayment history and cash-flow inputs into a single score. The risk team tunes weights and cutoffs directly, without code changes or release cycles.
• Underwriting as a rules engine: Applications that sit cleanly within policy are approved automatically. Exceptions route to credit officers, and every decision is recorded with the inputs behind it.
• Disbursement wired to the ledger: Approved loans trigger disbursement through the lender's payment providers via API. Ledger entries are written to PostgreSQL as the single source of truth, so the books update as the money moves.
• Event-driven collections with a full audit trail: Repayment schedules, reminders and delinquency queues run off Kafka events, so collections staff work from live loan states rather than exported reports. Keycloak handles staff authentication and role-based access, and every state change on a loan is logged for audit and regulatory review.

The measure of success was operational: what the lender's teams stopped doing by hand, and what the platform now does for them.

• Decisions at the point of application: Applications within policy clear scoring and underwriting automatically. Officers can give a customer an answer during the same visit instead of sending paperwork back to the branch.
• Comparable risk across the book: Every loan carries a score and a recorded decision path. The risk team compares branches and tunes policy from data instead of anecdotes.
• Manual reconciliation eliminated: Disbursements and repayments post to the ledger automatically, ending the spreadsheet reconciliation that used to follow every collection run.
• A base for new products: New loan products are configuration on the same platform rather than new paper processes, and the audit trail supports regulatory reporting as the book grows.