Building a fully automated digital lending platform processing ₹50Cr+ in monthly disbursals

SwiftCredit received their NBFC-MFI licence with a 12-month window to demonstrate operational activity before facing licence review. Building a lending technology stack from scratch in 12 months is an aggressive timeline under any circumstances. Building one that complies with RBI's Master Directions on digital lending — covering KYC via CKYC and CERSAI, credit bureau integration with CIBIL and Experian, FLDG disclosure requirements, loan agreement storage, and the Key Fact Statement format — adds a layer of regulatory complexity that has derailed far better-funded programmes.

The target customer segment — salaried professionals earning ₹25,000-₹75,000 per month and MSME proprietors with 2+ years of GST filing history — had been identified as underserved by both bank lending (too small for relationship banking attention) and informal lenders (too expensive). The commercial logic was compelling. The underwriting challenge was significant: assessing creditworthiness for this segment required integrating bank statement analysis, GST return data, CIBIL scores, and alternative data sources (mobile usage patterns, social signals) in ways that traditional credit models had not.

Collections — the operational competency that separates profitable NBFCs from loss-making ones — is often underengineered in new entrants focused on acquisition and growth. SwiftCredit's founders had seen competitors grow rapidly and then collapse when early-vintage loan books went delinquent. They insisted on building collections infrastructure with the same rigour as the origination stack, even when it slowed the initial build timeline.

We structured the 12-month build into three parallel workstreams: core lending platform (LOS/LMS), data infrastructure and credit model, and regulatory compliance and audit readiness. Each workstream had its own lead architect and timeline, with weekly integration checkpoints to ensure the three streams remained aligned. The regulatory compliance workstream ran from month one rather than being addressed at the end — every feature was reviewed against the RBI Master Directions before engineering work began.

The credit underwriting model was built as a modular scoring engine with three layers: bureau-based scoring using CIBIL and Experian data (weight: 40%), financial cash flow analysis from 6-month bank statement parsing using OCR and machine learning (weight: 35%), and alternative data scoring from GST return pattern analysis and, where available, account aggregator data via the AA framework (weight: 25%). The model was initially calibrated against a portfolio of 5,000 loans provided by a partner NBFC under a data sharing agreement, allowing pre-launch validation before SwiftCredit's own loan book was large enough to drive model refinement.

Collections infrastructure was designed around a 10-touch contact strategy triggered by specific days-past-due (DPD) milestones, combining automated WhatsApp, SMS, and IVR communications with human agent escalation at DPD 30+. All collection communications were RBI-compliant — no communications outside 8 AM to 7 PM, mandatory opt-out capability, no third-party disclosure of loan status. A legal notification workflow was built for DPD 90+ accounts, integrating with a panel of debt recovery lawyers across 12 cities.

The Loan Origination System handled the complete borrower journey: online application with Aadhaar-based eKYC, automated CKYC and CERSAI registry lookup, bank statement upload with automated parsing, credit bureau pull, underwriting model scoring, digital loan agreement with eSign via Aadhaar OTP, and disbursal instruction to the bank account via IMPS or NEFT. The end-to-end journey for a pre-qualified customer was reduced to under 18 minutes. RBI-mandated artefacts — Key Fact Statement, loan agreement copy, disbursement confirmation — were automatically generated, stored, and accessible to the borrower via a self-service portal.

The Loan Management System handled all post-disbursal operations: EMI scheduling with ECS/NACH mandate management, payment reconciliation from the bank statement feed, prepayment processing with interest recalculation, NOC generation on closure, and the full regulatory reporting suite required by RBI — including SMA classification, credit bureau reporting, and CERSAI charge creation and satisfaction. All data was stored on AWS with geo-redundant backups, audit logging on every record change, and role-based access controls separating origination, collections, and finance functions.

SwiftCredit went live in month 11, one month ahead of the deadline. By month 18, the platform was processing ₹50 crore+ in monthly disbursals across 4,200 active loans. The credit model achieved a 30-DPD delinquency rate of 2.1% on the first 1,000 loans — materially below the 3.5% projection used in the business plan. Three rounds of equity funding totalling ₹45 crore were raised, with all three investors citing the quality of the technology infrastructure as a key diligence finding.