P2P Lending Software Development: Building Peer-to-Peer Platforms

• The note sold to US P2P investors is a security, a point the SEC established against Prosper in 2008, so the investor side is built around a registration or an exemption.
• Most US platforms are not the lender: a partner bank originates the loan, and no federal true lender safe harbor has existed since the OCC rule's 2021 repeal.
• The ledger is a regulated artifact: double-entry and append-only, reconciled against the bank of record, with the FDIC proposing to make daily reconciliation mandatory.
• Budget-regulated features at one and a half to two times their unregulated equivalents.

The market is measurable from bureau data. TransUnion's Q2 2026 Credit Industry Insights Report puts unsecured personal loan balances at $281 billion, up 9.6% year over year across 33.3 million loans, with 60-day borrower delinquency at 3.81%. Fintech lenders held more than half of those balances, compared with 21% for banks, as of Q3 2025, the most recent quarter for which TransUnion reported the balance split. They also originated over 40% of new personal loans in Q2 2025. Three of the five ranking guides cite the same aggregator forecast that a competing aggregator contradicts by roughly a factor of two, so neither figure appears here.

The delinquency number matters most, because managing the loss curve is the bulk of what the software does after launch. That work happens under the eye of regulators and auditors, which changes engineering practice from the first sprint:

"If I expect a regulator or auditor to look at a system, I build for traceability from day one. That means audit logs, append-only records of sensitive actions, strict production access, separate environments, reviewed changes, and sufficient history to explain who changed what and when. As a rough budgeting rule, a regulated feature typically costs 1.5 to 2 times as much as the same feature in a standard SaaS product. The UI may be similar. The extra work is in controls, auditability, testing, failure handling, and operational tooling."

Nobody publishing on this topic puts a number on that overhead, and founders consistently budget as though it were zero. The general case for secure financial software is covered in our fintech software development guide.

In November 2008, the SEC ordered Prosper, which had been selling borrower payment-dependent notes to retail investors since 2006, to stop, holding that the notes were securities under Section 2(a)(1) of the Securities Act and had been offered unregistered in violation of Sections 5(a) and 5(c). Prosper also settled with state regulators through NASAA, paying a $1 million fine coordinated through a working group of roughly 20 jurisdictions. By spring 2011, a GAO report found that Prosper and LendingClub together had made about 63,000 loans totaling about $469 million, with registrations approved in 30 states plus DC, while lenders in 20 states could not participate.

That history is why the first question on your one-page drawing is an offering question. The broad paths are a full shelf registration of the notes, a Reg D private placement limited to accredited investors, a Reg A+ offering, or Regulation Crowdfunding. The crowdfunding route caps an issuer at $5 million in any 12 months. It requires every transaction to run through an SEC-registered broker-dealer or funding portal, with resale restricted for a year. Which path fits is a question for securities counsel. Still, each one compiles into different software: accreditation checks, per-state eligibility gates, resale restrictions in the transfer logic, and disclosure documents generated from live loan data.

Calling yourself a marketplace does not move the obligation elsewhere. In 2021, the SEC charged the funding portal TruCrowd itself for failing to meet its gatekeeper duties for offerings it hosted. If your platform sits between an issuer and an investor, some gatekeeping function lives in your code and review queues.

The borrower side poses a parallel question: either your platform is the lender of record in every state it serves, which means holding state lending licenses, or an FDIC-insured partner bank originates each loan and sells it to the platform. This is the model the GAO documented WebBank running as early as 2011.

State licensing is administered through NMLS: the Conference of State Bank Supervisors reported that more than 41,000 companies and 556,000 individuals held active licenses or registrations as of 2023. Each state adds an application and an examination requirement, plus rate caps your pricing engine must apply per borrower location.

The bank-partner model carries an open risk called true lender: a court or state regulator can look past the bank and treat the platform as the real lender, stripping the exported interest rate and potentially voiding loans retroactively. The OCC's 2020 rule would have settled the question nationally, but Congress repealed it in 2021, and the OCC's Federal Register notice removing the rule states it "has no legal force or effect." No federal safe harbor replaced it. The working template is Colorado's 2020 settlement with Avant and Marlette: a 36% rate ceiling on Colorado loans, payments of $1.05 million plus a $500,000 contribution, and a requirement that the non-bank partners hold a Colorado lending license. Your state eligibility engine and rate logic should be built expecting more agreements like it.

Whichever model you pick, the compliance perimeter runs through your codebase. The FDIC's interagency guidance on third-party relationships tells banks that using third parties "does not diminish or remove a banking organization's responsibility," so a partner bank will examine your change management, logs, and models as though they were its own. That oversight surface is a deliverable.

Investor money that is committed but not yet deployed, and borrower repayments in flight, usually sit in an FBO account: a pooled account at a bank titled "for the benefit of" your users. The bank sees a single balance, and your ledger records which person owns which dollars. That division of labor broke down during the 2024 collapse of a banking middleware provider, and the FDIC responded with a proposed recordkeeping rule that would require reconciliation of beneficial-owner ledgers "no less frequently than at the close of business daily." The FDIC prices compliance at $250 million in the first year and $120 million annually, across 600 to 1,100 institutions.

Your money-flow page should show which account holds funds at every step, in whose name, at which bank, and what happens to each pool if the platform files for bankruptcy. Until the bankruptcy column is filled in, the structure is not finished, and neither is the schema.

Every competitor guide lists a wallet or an escrow bullet in its feature table and moves on. In practice, the ledger is where these platforms fail, because it is the one component that has to agree with an external party, to the cent, every day. My approach has stayed the same across projects:

"For money movement, I prefer a double-entry ledger with an append-only journal. Posted entries are never edited. Corrections occur through reversals, and every external transaction has a stable reference point so we can reconcile our records with the bank or payment provider. Most reconciliation problems come from timing rather than arithmetic. Duplicate callbacks, late reversals, settlement batches, and different cutoff times can all create temporary mismatches. If you only keep a mutable 'current balance', finding the cause later becomes very difficult."

In 2019, Prosper's funding entity paid a $3 million penalty because its systems excluded charged-off loans from the annualized net return figures shown to more than 30,000 investors, overstating returns for almost two years. A calculation bug in a reporting pipeline ended as a securities enforcement action. Append-only journals and daily reconciliation are cheap on day one and close to impossible to retrofit on day 900, which is why they belong in the first statement of work.

Three regulatory areas directly affect the schema and model layers, and all are cheaper to accommodate before the first migration.

Identity and screening without the acronym soup

The customer identification program minimums are set in 31 CFR 1020.220: name, date of birth, address, and an identification number. Scope is the part that surprises technical teams. FinCEN's AML program rule for loan and finance companies currently applies only to residential mortgage lenders and originators, and the 2012 rulemaking states that the definition was structured so that other loan companies can be added later. In practice, the distinction buys you little. A partner bank contractually pushes its own CIP and monitoring obligations onto the platform, and OFAC screening applies to everyone. The FFIEC BSA/AML manual shows what examiners test. The build therefore includes identity verification and recurring watchlist screening, as well as a case queue for hits.

Underwriting under fair lending law

ECOA and Regulation B require specific, accurate reasons when credit is denied. The CFPB's Circular 2022-03 states that these laws "do not permit creditors to use complex algorithms when doing so means they cannot provide the specific and accurate reasons for adverse actions." Whatever model scores applications, you need a path from each decision back to the concrete factors behind it, and from those factors, generate adverse action notices. Model explainability is a legal requirement here, and it rules out some model choices outright.

Disclosures are a computation problem

TILA and Regulation Z prescribe how APR and finance charges are calculated and displayed, so disclosure is arithmetic your platform performs on deadline for every loan. The FTC's case against LendingClub over "no hidden fees" claims produced an $18 million settlement, and the agency returned more than $17.6 million to consumers, including $9.7 million to 61,990 people in a 2022 distribution. Fee display logic is compliance code and deserves the same review rigor as the payment engine.

Most of a lending platform's code runs after the loan is funded. Money moves mainly via ACH, and Nacha's WEB debit rule has required account validation before the first debit and again after any account number change since March 2021. Bank account verification is therefore a mandatory step in the funding flow. Every failed pull becomes a servicing event: a return code to interpret and a retry decision, often followed by a delinquency status change that feeds bureau furnishing and the investor's return calculation.

Behind that sits the servicing core of payment application waterfalls, amortization schedules, interest accrual, and charge-off logic. Custom builds accumulate silent errors here that surface at year-end or in an investor return figure, which is the exact path Prosper's $3 million penalty followed. Delinquency is standing workload rather than an exception path: the New York Fed's Q2 2026 household debt report shows $18.8 trillion outstanding with 4.7% of it in some stage of delinquency. Collections states, hardship plans, and agency handoffs belong in the original scope.

"Founders often underestimate the operational work behind the happy path. Failed payments, reversals, manual reviews, reconciliation issues, historical reporting, and support investigations all need software too. Those are part of the product, not cleanup work after launch."

After enough of these builds, my split has become stable:

"I build the parts that define the product, such as lending workflows, business rules, the internal ledger, servicing logic, and operational tools. I usually buy specialist infrastructure such as Persona or Socure for identity, Plaid for bank connectivity, Stripe or Dwolla for payments, DocuSign for signatures, and established bureaus for credit data. My rule is simple. Buy infrastructure that is expensive to maintain but does not differentiate the product. Build the logic and financial records you need to understand and control yourself."

Card data is the clearest rent decision of all. PCI DSS v4.0.1 became the only active version when v4.0 retired at the end of 2024, with future-dated requirements mandatory from March 31, 2025. The sound response is to keep card data out of your systems via a tokenization provider, so your assessed scope shrinks toward zero. The team shape this implies is a small core that owns the ledger and lending logic, with integration work around it, the same principle behind our web application development services applied to a regulated domain.

The most clearly documented platform failure is British. Lendy, a property P2P lender, entered administration in May 2019 with the FCA confirming an ongoing investigation. The FCA chief executive's letter to Lord Myners sets out what was wrong, and each item reads as a software and data failure:

• the quality of information shown to investors, which is a reporting pipeline
• governance and systems and controls, which is workflow, review, and access design
• a charge held over the provision fund, which is an encumbrance no system surfaced to investors
• a back book of four loans "that appear not to be P2P agreements," which is product records diverging from legal reality

Managed funds attached to a marketplace carry the same exposure. In 2018, LendingClub Asset Management and two former executives agreed to pay more than $4.2 million in combined penalties after the SEC found that fund returns had been improperly adjusted. The lesson repeats across every case here: each number an investor sees is produced by software your team wrote, and securities law applies to its output.

So the one-page drawing I asked for at the start has three tests:

• it shows what you are selling to investors, and under which registration or exemption
• it shows who originates and holds each loan and each dollar, in whose name, at which bank
• it shows what the ledger must prove, how often, and to whom

Once you can draw that page with securities and lending counsel in the room, you are ready to scope the software, and the build will be cheaper for having started there.

Is a P2P lending note a security in the United States?

Yes. The SEC's 2008 cease-and-desist order against Prosper held that its notes were securities under Section 2(a)(1) of the Securities Act, and that the platform had to stop selling and register them. Structure the investor side around a registration or an exemption from the first design session.

Do I need a lending license to run a P2P lending platform?

You either work with a partner bank that originates the loans, or you obtain licenses in each state you lend into through NMLS. No federal true lender safe harbor has existed since the 2021 repeal of the OCC rule, so this choice needs lending counsel before the architecture is fixed.

How much does P2P lending software development cost?

No published cost table for this category is verifiable, so this article does not carry one. The honest anchor from my own projects: a regulated feature costs 1.5 to 2 times its unregulated equivalent, and the overhead lies in controls, auditability, testing, and operational tooling.

What is an FBO account and why does my platform need one?

An FBO ("for benefit of") account holds user money at a bank in your platform's name for the benefit of the actual owners. The bank sees a single pooled balance, while your ledger tracks individual ownership, which is why the FDIC has proposed requiring daily reconciliation of beneficial-owner records.

• U.S. Securities and Exchange Commission, "Order Instituting Cease-and-Desist Proceedings: Prosper Marketplace, Inc." (Securities Act Release No. 8984, 2008). https://www.sec.gov/files/litigation/admin/2008/33-8984.pdf. Cited for the holding that Prosper's notes are securities and the Sections 5(a) and 5(c) violations.
• U.S. Government Accountability Office, "Person-to-Person Lending" (GAO-11-613, 2011). https://www.gao.gov/assets/a320698.html. Cited for the WebBank origination model, the March 2011 loan volumes, and the state registration counts.
• North American Securities Administrators Association, "Prosper Marketplace, Inc. Enters Settlement with State Securities Regulators" (2008). https://www.nasaa.org/5622/prosper-marketplace-inc-enters-settlement-with-state-securities-regulators-over-sales-of-unregistered-securities/. Cited for the $1 million multi-state settlement.
• U.S. Securities and Exchange Commission, "Regulation Crowdfunding" (current). https://www.sec.gov/resources-small-businesses/exempt-offerings/regulation-crowdfunding. Cited for the $5 million cap, intermediary requirement, and one-year resale restriction.
• U.S. Securities and Exchange Commission, press release 2021-182, "SEC Charges Crowdfunding Portal" (2021). https://www.sec.gov/newsroom/press-releases/2021-182. Cited for the TruCrowd gatekeeper enforcement.
• U.S. Securities and Exchange Commission, press release 2019-58, "Prosper Funding to Pay $3 Million Penalty" (2019). https://www.sec.gov/newsroom/press-releases/2019-58. Cited for the overstated investor returns penalty.
• Office of the Comptroller of the Currency via Federal Register, "National Banks and Federal Savings Associations as Lenders" (2021). https://www.federalregister.gov/documents/2021/08/05/2021-16619/national-banks-and-federal-savings-associations-as-lenders. Cited for the repeal of the true lender rule and the "no legal force or effect" language.
• Colorado Attorney General, "Attorney General Phil Weiser announces settlement with Avant and Marlette Funding" (2020). https://coag.gov/press-releases/8-18-20/. Cited for the 36% ceiling, payment amounts, and licensing requirement.
• Conference of State Bank Supervisors, "Nonbank Licensing and Examination" (2023 figures). https://www.csbs.org/nonbank-licensing-and-examination. Cited for NMLS licensing counts.
• Federal Deposit Insurance Corporation, FIL-29-2023, "Interagency Guidance on Third-Party Relationships: Risk Management" (2023). https://www.fdic.gov/news/financial-institution-letters/2023/fil23029.html. Cited for the responsibility language banks apply to platform partners.
• Federal Deposit Insurance Corporation via Federal Register, "Recordkeeping for Custodial Accounts" (proposed rule, 2024). https://www.federalregister.gov/documents/2024/10/02/2024-22565/recordkeeping-for-custodial-accounts. Cited for the daily beneficial-owner reconciliation proposal and the FDIC's cost estimates.
• U.S. Treasury eCFR, 31 CFR 1020.220 (current). https://www.ecfr.gov/current/title-31/subtitle-B/chapter-X/part-1020/subpart-B/section-1020.220. Cited for the four customer identification program minimums.
• Financial Crimes Enforcement Network via Federal Register, "Anti-Money Laundering Program and Suspicious Activity Report Filing Requirements for Residential Mortgage Lenders and Originators" (2012). https://www.federalregister.gov/documents/2012/02/14/2012-3074/anti-money-laundering-program-and-suspicious-activity-report-filing-requirements-for-residential. Cited for the current scope of the AML program rule for loan and finance companies.
• Federal Financial Institutions Examination Council, "BSA/AML Examination Manual: Customer Identification Program" (current). https://bsaaml.ffiec.gov/manual/AssessingComplianceWithBSARegulatoryRequirements/01. Cited for what examiners test.
• Consumer Financial Protection Bureau, "Circular 2022-03: Adverse action notification requirements in connection with credit decisions based on complex algorithms" (2022). https://www.consumerfinance.gov/compliance/circulars/circular-2022-03-adverse-action-notification-requirements-in-connection-with-credit-decisions-based-on-complex-algorithms/. Cited verbatim for the explainability requirement.
• Federal Trade Commission, "Federal Trade Commission Returns More Than $9.7 Million to Consumers Harmed by LendingClub's Deceptive Hidden Fees" (2022). https://www.ftc.gov/news-events/news/press-releases/2022/08/federal-trade-commission-returns-more-97-million-consumers-harmed-lendingclubs-deceptive-hidden-fees. Cited for the settlement and refund figures.
• NACHA, "Supplementing Fraud Detection Standards for WEB Debits" (effective 2021). https://www.nacha.org/rules/supplementing-fraud-detection-standards-web-debits. Cited for the account validation requirement.
• PCI Security Standards Council, "Just Published: PCI DSS v4.0.1" (2024). https://blog.pcisecuritystandards.org/just-published-pci-dss-v4-0-1. Cited for version status and the March 2025 mandatory date.
• TransUnion, "Q2 2026 Credit Industry Insights Report" (2026). https://newsroom.transunion.com/Q2-2026-CIIR/. Cited for unsecured personal loan balances, growth, loan counts, and delinquency.
• TransUnion, "Q3 2025 Credit Industry Insights Report" (2025). https://newsroom.transunion.com/q3-2025-ciir/. Cited for fintech and bank shares of balances and originations.
• Federal Reserve Bank of New York, "Quarterly Report on Household Debt and Credit, Q2 2026" (2026). https://www.newyorkfed.org/newsevents/news/research/2026/20260811. Cited for total household debt and the aggregate delinquency rate.
• Financial Conduct Authority, "Lendy Ltd enters administration" (2019). https://www.fca.org.uk/news/statements/lendy-ltd-enters-administration. Cited for the administration and investigation.
• Financial Conduct Authority, Andrew Bailey letter to Lord Myners regarding Lendy (2019). https://www.fca.org.uk/publication/correspondence/lord-myners-lendy-ltd.pdf. Cited for the enumerated failures.
• U.S. Securities and Exchange Commission, press release 2018-223, "LendingClub Asset Management and Former Executives Settle Fraud Charges" (2018). https://www.sec.gov/newsroom/press-releases/2018-223. Cited for the $4.2 million combined payment and the adjusted fund returns finding.

This article is a technical guide for founders and CTOs commissioning software, written from an engineering perspective. It is not legal, financial, or investment advice. How securities, lending, and money transmission law treats a specific platform depends on its structure and jurisdictions, so engage securities and lending counsel before the architecture is fixed. Descriptions here are current as of August 2026 and are a map of what to ask counsel.