A P2P lending platform launches with good intentions. Borrower onboarding is smooth. The investor dashboard looks professional. The first \u20b92 crore in loans gets funded. The team celebrates.<\/span><\/p>\n Month 6: default rate climbs to 18%. The credit scoring module was using bureau data alone, thin-file borrowers who looked acceptable on CIBIL had no bank statement analysis, no income verification beyond self-declaration, and no alternative data layer. The scoring model was optimistic by design because higher approval rates meant more transactions and more revenue.<\/span><\/p>\n Month 9: the collections module is a WhatsApp group. There are no automated reminders, no NACH bounce handling, no escalation workflows. The team is manually chasing recoveries. Three investors have filed complaints with the RBI.<\/span><\/p>\n This is not a hypothetical. The RBI’s tightened P2P lending regulations in August 2024\u00a0 explicitly forbidding platforms from guaranteeing returns, creating assured products, or using automated algorithmic investment tools without investor consent\u00a0 came directly from seeing this pattern across multiple platforms.<\/span><\/p>\n I co-founded EngineerBabu 14 years ago. The team’s CTO spent 17 years at Wishfin, one of India’s largest credit marketplaces. The team has built lending platforms that together process thousands of crores annually\u00a0 including one that has disbursed over \u20b910,000 crore and LoanOS, the team’s own modular lending platform, currently processing \u20b91,000 crore annually through a DSA. We know what breaks. This guide is how to prevent it.<\/span><\/p>\n If you’re ready to build and want a team that’s built at this scale\u00a0 email <\/b>mayank@engineerbabu.com<\/b><\/a>.\u00a0<\/b><\/p>\n India’s P2P lending market is projected to reach $10.5 billion by 2026<\/b><\/a>, growing at a CAGR of 21.6%. The global P2P lending market was valued at $176 billion and is projected to reach $1.38 trillion by 2034 at a 25% CAGR.<\/span><\/p>\n The structural drivers: 1.59 lakh registered startups in India as of 2025, a massive gig economy with 2+ crore workers, and MSME entrepreneurs who face systemic credit rejection from traditional banks. These are creditworthy borrowers with no bureau history. P2P platforms\u00a0 when built correctly\u00a0 serve them with returns of 10\u201318% per annum for investors.<\/span><\/p>\n But here’s the context the market numbers miss.<\/span><\/p>\n The RBI cracked down on India’s P2P sector in August 2024. The revised Master Directions explicitly banned guaranteed returns, algorithmic investment without investor consent, and a number of practices that platforms had been using to simulate fixed-income products. Platforms that built their product on regulatory grey areas had to reconstruct their entire model.<\/span><\/p>\n The NPA rate in India’s P2P lending sector reportedly rose to approximately \u20b91,163 crore in FY24. The platforms that survived the regulatory tightening\u00a0 and the NPA cycle\u00a0 were the ones that had built their credit scoring, collections, and compliance architecture correctly from the start.<\/span><\/p>\n A P2P lending platform is a digital marketplace<\/b> that connects individual or institutional investors directly with individual or business borrowers, enabling loans to be originated, funded, managed, and collected entirely through the platform\u00a0 without a bank as the intermediary. The platform earns fees from the transaction, not interest from its own balance sheet. The platform bears the credit risk of the marketplace design, not the principal risk of individual loans.<\/span><\/p>\n That distinction\u00a0 marketplace mechanics, not balance sheet lending, is what defines a P2P platform architecturally and what the RBI’s framework regulates.<\/span><\/p>\n India (RBI NBFC-P2P):<\/b> All P2P <\/span>lending platforms in India<\/b><\/a> must be registered as NBFC-P2P under the RBI. Key requirements: minimum net owned fund of \u20b92 crore, maximum aggregate exposure per lender of \u20b950 lakh across all P2P platforms, maximum loan exposure per borrower of \u20b910 lakh, maximum loan tenure of 36 months, and no guarantee of returns or creation of assured products.<\/span><\/p>\n The August 2024 revised Master Directions added: T+1 settlement (funds must reach borrowers within 1 business day of loan sanction), escrow account requirements through a bank or promoted escrow agent, and mandatory disclosure of NPA rates in real time.<\/span><\/p>\n United States (SEC\/State):<\/b> US P2P lending platforms that issue notes to investors are regulated as securities offerings. Platforms must register offerings with the SEC or qualify for an exemption (Regulation D, Regulation A+, Regulation Crowdfunding). State money transmitter licenses required for fund movement. LendingClub, Prosper, and Funding Circle all operate under SEC registration as a multi-year, multi-million-dollar compliance investment.<\/span><\/p>\n UK (FCA):<\/b> FCA authorisation as a P2P lending platform under the loan-based crowdfunding regulations. Investor appropriateness assessments required. Wind-down plan required. NPA disclosure requirements similar to India’s.<\/span><\/p>\n The honest answer: for most startups in 2026, the India NBFC-P2P path is the most achievable regulatory route with the clearest framework. The US path is viable but requires a minimum 18-month regulatory runway and significant legal investment before a single loan is processed.<\/span><\/p>\n P2P lending is a marketplace. Matching borrowers and lenders isn’t just a UX feature, it’s a regulatory obligation and a risk management system.<\/span><\/p>\n Under India’s revised framework, platforms cannot automatically allocate lender funds to borrowers without explicit investor consent for each loan. Every lender must actively choose which borrower’s loan to fund, or explicitly consent to an auto-invest algorithm’s decisions.<\/span><\/p>\n This changes the architecture significantly:<\/span><\/p>\n Loan listing: every<\/b> approved loan must be listed with full disclosure: borrower risk profile, credit score, purpose, tenure, interest rate, LTV ratio, and the platform’s internal risk classification. The listing is a disclosure document, not just a product card.<\/span><\/p>\n Investor allocation flow<\/b>\u00a0 the lender selects a loan, commits a specific amount, and the commitment is escrowed. When the loan reaches its funding threshold (100% of the requested amount), disbursement is triggered. If the threshold isn’t reached within the listing period, committed funds are returned.<\/span><\/p>\n Auto-invest with consent<\/b>\u00a0 if the platform offers an auto-invest feature (the investor sets criteria, the system allocates automatically), explicit informed consent must be obtained and documented for each auto-invest rule change.<\/span><\/p>\n Partial funding mechanics<\/b>\u00a0 most <\/span>P2P platforms<\/b><\/a> allow multiple investors to fund portions of one loan. The platform needs to track each investor’s contribution, calculate their proportional share of repayments, and attribute NPAs to the correct investor accounts.<\/span><\/p>\n The matching engine is not complex algorithmically. But the regulatory mechanics around consent, disclosure, escrow sequencing\u00a0 are where most platforms fail.<\/span><\/p>\n The NPA rate on a P2P platform is almost entirely determined by the quality of the credit scoring engine.<\/span><\/p>\n The failure pattern described at the start of this guide\u00a0 bureau data alone, no alternative data\u00a0 produces a specific NPA curve: low defaults for the first 6 months (the loans that were genuinely creditworthy by any measure), then a cliff at months 7\u201312 as the edge cases begin defaulting.<\/span><\/p>\n Production credit scoring for a P2P platform:<\/span><\/p>\n Bureau data<\/b>\u00a0 CIBIL, Experian, Equifax. The starting point, not the complete picture. The CTO’s 17 years at Wishfin means the team has built bureau integrations for every major Indian credit bureau, understands the edge cases in each bureau’s data model, and knows which bureau’s data is most predictive for which borrower segment.<\/span><\/p>\n Bank statement analysis<\/b>\u00a0 income estimation, expense patterns, existing EMI obligations, average balance trends. This is the most predictive alternative data source for thin-file borrowers. APIs like Perfios or Finbox can analyse 6\u201312 months of bank statements in minutes and return structured income signals.<\/span><\/p>\n GST returns<\/b>\u00a0 for self-employed and MSME borrowers, GST return data is a strong revenue verification signal. GSTN API integration is available for consenting borrowers.<\/span><\/p>\n Behavioral data<\/b>\u00a0 time-on-platform, completeness of profile, document submission quality. These signals are weak individually but contribute to a composite risk score.<\/span><\/p>\n Separate scoring models for separate segments<\/b>\u00a0 a salaried employee in a Tier 1 city has a fundamentally different credit profile than an MSME owner in Tier 3. One model applied to both will underperform for both. The team builds separate scoring models per borrower segment, validated on historical performance data before production deployment.<\/span><\/p>\n As a Google AI Accelerator 2024 team, the approach to ML credit scoring is production-oriented: champion-challenger model framework, weekly retraining pipeline, automated NPA attribution back to the scoring model that approved each defaulted loan. The feedback loop is built from day one.<\/span><\/p>\n The escrow requirement in India’s revised P2P framework isn’t just a compliance checkbox. It’s the trust infrastructure that makes the whole marketplace function.<\/span><\/p>\n Every fund movement on a P2P platform must flow through NPCI-approved escrow accounts:<\/span><\/p>\n The platform itself never holds the principal. This is the regulatory requirement that the platform is a facilitator, not a lender.<\/span><\/p>\n Building this correctly requires:<\/span><\/p>\n Escrow account integration<\/b>\u00a0 partnership with a bank that provides NBFC-P2P-compliant escrow accounts. ICICI, HDFC, and Axis all offer escrow services for P2P platforms.<\/span><\/p>\n Funds flow accounting<\/b>\u00a0 every rupee entering and leaving each escrow account must be tracked at the transaction level. The escrow ledger is the regulatory audit trail. It must be immutable.<\/span><\/p>\n T+1 settlement<\/b>\u00a0 once a loan is sanctioned and funded, disbursement must reach the borrower within 1 business day. This is an SLA that the platform’s payment processing architecture must support consistently.<\/span><\/p>\n Investor account reconciliation: Each<\/b> investor sees a real-time breakdown: committed (in escrow against a specific loan), disbursed (loan is live), repaying (EMI received), and available (ready to reinvest). This reconciliation view must be accurate to the rupee, updated in real time.<\/span><\/p>\n Collections is where P2P platforms distinguish between sustainable businesses and short-term volume machines.<\/span><\/p>\n The automated collections stack:<\/span><\/p>\n NACH mandate<\/b>\u00a0 registered at loan disbursement, not after first missed payment. Every borrower who receives a loan signs up for NACH (National Automated Clearing House) auto-debit at disbursement. The NACH mandate covers the full loan tenure.<\/span><\/p>\n Pre-due reminders<\/b>\u00a0 SMS and WhatsApp 7 days, 3 days, and 1 day before EMI due date. Not optional. The platforms with the lowest NPAs have the most aggressive pre-due reminder infrastructure.<\/span><\/p>\n Bounce handling<\/b>\u00a0 when NACH debit bounces, the system detects it within 4 hours (NPCI provides near-real-time bounce notifications), triggers a retry logic, sends a payment link to the borrower, and escalates to the collections queue. The difference between a 1-day bounce and a 7-day bounce in terms of recovery probability is enormous.<\/span><\/p>\n Escalation workflows<\/b>\u00a0 Day 1: automated reminder + payment link. Day 7: escalation to collections team with full account context. Day 30: NPA classification + legal notice generation. Day 90: writeoff + recovery agent assignment.<\/span><\/p>\n Investor NPA attribution<\/b>\u00a0 when a loan defaults, the investor who funded it sees their exposure shift to NPA in their portfolio. The NPA calculation must match the RBI’s classification criteria exactly; the platform’s NPA numbers must be auditable by the regulator.<\/span><\/p>\n LoanOS, the team’s own lending platform, has automated collections as a core module. The module was built after seeing manual collections at multiple platforms fail catastrophically at scale. It is the most underestimated investment in a P2P platform build and the one that most directly determines whether investors return.<\/span><\/p>\n Investors in a P2P platform are the liquidity providers. Losing investors is existentially dangerous. The investor portfolio management experience determines whether they return capital for new loans.<\/span><\/p>\n Production investor portfolio management:<\/span><\/p>\n Risk-diversified auto-invest<\/b>\u00a0 investors set risk tolerance and investment parameters. The platform allocates across multiple borrowers automatically, diversifying the investor’s exposure. No single borrower can represent more than a defined percentage of any investor’s total P2P portfolio.<\/span><\/p>\n Real-time portfolio dashboard<\/b>\u00a0 net returns (XIRR), outstanding principal, loan-wise performance, NPA tracking, upcoming repayments, and cash available for reinvestment. Every number must be accurate to the rupee and updated within minutes of any transaction.<\/span><\/p>\n Tax reporting<\/b>\u00a0 P2P returns are taxed as interest income in India. The platform must generate Form 26AS-compliant interest certificates for investors. This is a non-negotiable feature for institutional investors and a strong trust signal for individual investors.<\/span><\/p>\n Withdrawal mechanics<\/b>\u00a0 investors must be able to withdraw available (uninvested) funds on demand. The T+2 withdrawal SLA is standard. Building this reliably requires the escrow architecture to support immediate fund release for uninvested balances.<\/span><\/p>\n The RBI’s oversight of NBFC-P2P platforms is data-intensive. The platform must be able to produce specific reports on demand:<\/span><\/p>\n Monthly NPA disclosure<\/b>\u00a0 gross NPA ratio, net NPA ratio, NPA by tenure bucket. Must match the borrower-level ledger exactly.<\/span><\/p>\n Lender exposure report<\/b>\u00a0 proof that no lender has aggregate P2P exposure above \u20b950 lakh.<\/span><\/p>\n Borrower exposure report<\/b>\u00a0 proof that no borrower has aggregate P2P loans above \u20b910 lakh across all platforms (requires cross-platform data sharing through the CIC\/credit bureau system).<\/span><\/p>\n Escrow reconciliation<\/b>\u00a0 weekly reconciliation of escrow account balances against platform ledger.<\/span><\/p>\n IS audit support<\/b>\u00a0 annual IS audit is mandatory. The platform must provide complete audit logs, access records, and system documentation.<\/span><\/p>\n Building regulatory reporting as an engineering system\u00a0 automated, scheduled, with defined data sources and reconciliation checks\u00a0 is the difference between a platform that passes audits and one that scrambles for data every quarter.<\/span><\/p>\n P2P lending fraud has three distinct vectors:<\/span><\/p>\n Synthetic identity borrowers<\/b>\u00a0 fraudsters create <\/span>loan applications<\/b><\/a> using a mix of real and fabricated data. They pass basic KYC because the PAN is real and the Aadhaar is real. But the address, the employment, and the bank statements are fabricated or manipulated. Detection requires device fingerprinting, velocity checks across multiple applications from the same device, and network analysis of connected borrower profiles.<\/span><\/p>\n Multiple platform borrowing<\/b>\u00a0 borrowers simultaneously apply to multiple P2P platforms to exceed the \u20b910 lakh cap. Detection requires integration with CIBIL’s P2P lending database (which tracks all P2P loan inquiries) and real-time enquiry checks at application submission.<\/span><\/p>\n Investor-side fraud<\/b>\u00a0 money laundering attempts using investor accounts to layer funds. AML monitoring on investor deposits and withdrawals, suspicious transaction reporting to FIU-IND.<\/span><\/p>\n The team builds ML fraud detection for P2P platforms using the same production AI pipeline applied to the fintech platforms built under the Google AI Accelerator 2024 program.<\/span><\/p>\n Frontend: Flutter (mobile\u00a0 borrower and investor apps) + Next.js (admin and regulatory reporting panel)<\/b><\/p>\n Backend: Python FastAPI (credit scoring + ML models) + Node.js NestJS (loan lifecycle management, escrow orchestration, collections automation)<\/b><\/p>\n Database: PostgreSQL<\/b>\u00a0 the loan ledger, investor accounts, escrow tracking. ACID compliance is non-negotiable when every transaction involves real money and regulatory audit obligations.<\/span><\/p>\n Redis<\/b>\u00a0 session management, NACH queue processing, real-time portfolio calculations.<\/span><\/p>\n Escrow integration<\/b>\u00a0 NPCI NACH for auto-debit, bank escrow API integration (ICICI\/HDFC), UPI for disbursement and voluntary repayments.<\/span><\/p>\n Credit scoring infrastructure<\/b>\u00a0 CIBIL API, Experian API, Perfios\/Finbox for bank statement analysis, GSTN for GST verification. All orchestrated through a unified credit assessment service.<\/span><\/p>\n Cloud: AWS Mumbai<\/b>\u00a0 RBI data localisation requirement. All financial data on Indian servers.<\/span><\/p>\n The lending platform that has disbursed over \u20b910,000 crore taught the team one lesson above all others: the credit scoring model that works at \u20b9100 crore AUM breaks at \u20b91,000 crore AUM, and the collections infrastructure that works at \u20b9100 crore AUM is the difference between \u20b91,000 crore and insolvency.<\/span><\/p>\n For the early salary advance product (which became one of India’s first to process thousands of daily credit decisions), the critical challenge wasn’t building the scoring model. It was building the retraining pipeline\u00a0 the system that automatically ingests new performance data, detects model drift, and retrains the scoring model on a weekly cadence.<\/span><\/p>\n When the borrower mix changes, a new employer segment enters the platform, macroeconomic shift affects repayment behavior, and the scoring model that was calibrated on the previous population becomes less predictive. Without a retraining pipeline, that degradation is invisible until NPA rates start climbing.<\/span><\/p>\n LoanOS, the team’s own modular lending platform, has the retraining pipeline as a core component. Not an optional ML feature. A production system.<\/span><\/p>\n The CTO’s 17 years at Wishfin adds specific credibility to every credit scoring and collections architecture decision. He has seen bureau API failures at scale, NACH rejection cascades, and scoring model drift in production. These patterns are designed around, not discovered.<\/span><\/p>\n The team can have a scoped P2P lending architecture in your inbox within a week. <\/b>mayank@engineerbabu.com<\/b><\/a>. \\<\/b><\/p>\n Credit scoring model designed for approval rate, not default rate. Thin-file borrowers approved on bureau data alone. NPA cliff at month 7. Investor complaints follow.<\/span><\/p>\n The fix:<\/b> Multi-layer scoring (bureau + bank statement + alternative data) validated against historical performance before production deployment. Approval rate is an outcome of good scoring, not a design parameter.<\/span><\/p>\n Platform built without proper escrow architecture. Funds flow through the platform’s bank account rather than NBFC-P2P-compliant escrow. RBI inspection finds the gap. Platform operations suspended pending remediation.<\/span><\/p>\n The fix:<\/b> Escrow architecture designed in sprint one, before any other financial infrastructure. T+1 settlement SLA baked into the payment processing design.<\/span><\/p>\n Collections handled manually\u00a0 WhatsApp messages, phone calls, spreadsheets. Works at \u20b910 crore AUM. Fails catastrophically at \u20b9100 crore AUM when the collections team is managing 5,000 active loans simultaneously.<\/span><\/p>\n The fix:<\/b> Automated collections from day one. NACH mandate at disbursement. Pre-due reminders. Bounce handling within 4 hours. Escalation workflows with defined triggers.<\/span><\/p>\n Platform launches, grows to meaningful AUM, then discovers that RBI’s revised Master Directions require changes to the auto-invest feature, the assured returns language, and the lender exposure tracking. Three months of platform restructuring at peak growth.<\/span><\/p>\n The fix:<\/b> Regulatory requirements mapped to technical controls before sprint 1. RBI Master Directions for NBFC-P2P read and implemented as design constraints, not legal documents.<\/span><\/p>\n White-label P2P software:<\/b> Available, faster to market, but designed for generic use cases. The credit scoring engine is not customizable. The collections workflows don’t account for NACH-specific failure modes. The regulatory reporting doesn’t map to India’s RBI requirements specifically.<\/span><\/p>\n Custom build:<\/b> Required for any platform that needs differentiated credit scoring (the actual competitive moat), compliance with India’s specific NBFC-P2P framework, and collections infrastructure at scale. Custom build is the only path to the level of credit quality control that separates platforms with 5% NPA from those with 18% NPA.<\/span><\/p>\n P2P lending platform development starts from $15K for a production <\/span>The Market Opportunity in 2026<\/b><\/h2>\n
![\"P2P](\"https:\/\/engineerbabu.com\/blog\/wp-content\/uploads\/2026\/06\/01_npa_default_rate_curve.png\" "\\"\\"")
<\/p>\nThe Licensing Landscape<\/b><\/h2>\n
The 7 Engineering Challenges That Decide Everything<\/b><\/h2>\n
1. The Marketplace Matching Engine\u00a0 Who Lends to Whom<\/b><\/h3>\n
2. Credit Scoring\u00a0 The Difference Between 8% and 18% NPA<\/b><\/h3>\n
3. Escrow Architecture\u00a0 The Trust Infrastructure<\/b><\/h3>\n
\n
4. Collections\u00a0 The Overlooked Revenue Protector<\/b><\/h3>\n
5. Investor Portfolio Management\u00a0 The Retention Feature<\/b><\/h3>\n
6. Regulatory Reporting\u00a0 What the RBI Actually Checks<\/b><\/h3>\n
![\"India](\"https:\/\/engineerbabu.com\/blog\/wp-content\/uploads\/2026\/06\/02_p2p_market_growth_india.png\" "\\"\\"")
<\/p>\n7. Fraud Detection\u00a0 Identity, Synthetic Borrowers, and Investor Fraud<\/b><\/h3>\n
Technology Architecture<\/b><\/h2>\n
How EngineerBabu Builds Lending Platforms\u00a0 Through Stories<\/b><\/h2>\n
![\"Multi-layer](\"https:\/\/engineerbabu.com\/blog\/wp-content\/uploads\/2026\/06\/03_credit_scoring_layers.png\" "\\"\\"")
<\/p>\nThe EngineerBabu P2P Failure Framework<\/b><\/h2>\n
Failure Mode 1: The Optimistic Scorer<\/b><\/h3>\n
Failure Mode 2: The Escrow Afterthought<\/b><\/h3>\n
Failure Mode 3: The Manual Collections Cliff<\/b><\/h3>\n
Failure Mode 4: The Regulatory Surprise<\/b><\/h3>\n
Build vs. White-Label<\/b><\/h2>\n
![\"RBI-compliant](\"https:\/\/engineerbabu.com\/blog\/wp-content\/uploads\/2026\/06\/04_escrow_fund_flow.png\" "\\"\\"")
<\/p>\nCost and Timeline<\/b><\/h2>\n