Financial Ecosystem — PM Visual Guide

Seven interconnected diagrams plus a PM components playbook: how money moves, where risk concentrates, where margin leaks, and which product levers actually change unit economics.

Diagram 1

The Financial Ecosystem — Concentric Layer Map

Banks sit at the center because all money ultimately lives in bank accounts. Every other layer is built on top of that foundation. Outer rings depend on everything inside them — not the reverse.

End Users
Borrowers Merchants Shoppers
Layer 3 · Financial Products
Lending
Credit Products
  • AI Underwriting
  • Marketplace Banks
  • Lead Generation
  • P2P / Institutional
Payments
Payment Products
  • Online Processing
  • In-store / POS
  • B2B Payments
  • Consumer Wallets
Embedded Credit / BNPL
Buy Now Pay Later
  • Pay in 4 installments
  • 3–36 month financing
  • Credit at point of sale
Layer 2 · Infrastructure APIs
Bank Data APIsIncome & account verification
Card Issuing APIsVirtual & physical cards
Money Movement APIsACH · Wire · RTP orchestration
Layer 1 · Payment Rails
Large-Value WireSame-day (Fedwire)
Card Networks~1.5s (Visa · MC)
Bank Transfers1–2 days (ACH)
Real-Time PaymentsInstant (RTP)
Core Foundation

Banks & Regulated Institutions

All money ultimately lives in bank accounts · Fed master accounts
Direct access to all rails · Lender of record in loan origination

Capital Sources

Funding relationship — not a technology layer. Connects directly to banks at the center, bypassing all middleware.
Bank Partners Lender of record in AI platform model
Institutional Investors Buy loan pools for yield above Treasuries
Forward-Flow Partners Pre-committed volume buyers
Depositors Low-cost funding for marketplace banks
Connect directly to banks via Fedwire and ACH — they ARE regulated institutions, so no Layer 2 middleware is needed.

How to read this: Start at the center and move outward. Banks are the foundation — all money lives in bank accounts and they have direct access to every payment rail. Each outer layer depends on everything inside it: Infrastructure APIs abstract the rails for fintechs. Financial products are built on those APIs. End users consume the products. Capital sources sit outside entirely — they're a funding relationship that connects straight to the banking center, not a technology dependency.

Diagram 2

The Four Lending Business Models

Same end result (borrower gets a loan), but the money takes a completely different path depending on the model. The model determines margins, risk exposure, and what a PM actually owns.

Model 1 Lead Generation
e.g. LendingTree, Credible
BorrowerFills out one form — loan amount, purpose, credit estimate
↓
PlatformMatches borrower to relevant lenders. Sells the lead to each.
↓
Multiple LendersEach underwrites independently and makes an offer
↓
Borrower picksCompletes full application with chosen lender. Platform exits.
Earns: Per-lead fee — regardless of whether borrower takes a loan

Key PM challenge

Lead quality — bad leads mean lenders stop paying
No control over the downstream experience
Borrower trust with shared financial data
Model 2 P2P / Institutional Marketplace
e.g. Prosper, LendingClub (pre-bank)
BorrowerApplies directly on the platform
↓
PlatformUnderwrites loan, lists it on marketplace
↓
InvestorsFund the loan (retail originally, now institutional)
↓
Platform services loanCollects payments, forwards to investors
Earns: Origination fee (upfront) + servicing fee (ongoing)

Key PM challenge

Investor confidence — model performance must be visible
Two-sided marketplace liquidity
Adverse selection if institutional capital dries up
Model 3 AI Underwriting Platform
e.g. Upstart
BorrowerApplies via platform or bank partner's white-label
↓
AI ModelRuns 1,000+ signals — outputs risk score + APR
↓
Bank PartnerLender of record — funds and owns the loan initially
↓
Capital PartnerBuys the loan. Now holds credit risk and earns interest.
Earns: Referral + platform fees. Never holds loans on balance sheet.

Key PM challenge

Three-sided market must all agree simultaneously
Model accuracy — errors affect all parties
Capital market dependency — exposure to rate cycles
Model 4 Digital Marketplace Bank
e.g. SoFi, LendingClub (post-bank)
DepositorsPut money in checking/savings. This is the low-cost funding.
↓
Borrower appliesDirectly with the bank — no third-party needed
↓
Bank underwritesOwn models. Often targets prime/super-prime.
↓
Bank funds from depositsEarns interest spread. May sell some loans to market.
Earns: Interest spread (deposit rate vs. loan APR) + fees + cross-sell

Key PM challenge

Full regulatory complexity (bank regulator + CFPB)
Cross-sell — getting members to use multiple products
Holds credit risk directly on balance sheet
Diagram 3

How a Loan Clears — All Four Models

Each lending business model has a distinct clearing sequence — different participants, different handoffs, different fee moments. Select a model to trace the flow from application to funded.

How it works: The platform is a matchmaker only. It sells the borrower's lead to multiple lenders and exits. It never underwrites, never holds credit risk, never disburses funds. Revenue is earned per lead sold — regardless of whether the borrower gets a loan.
🙋Borrower
🌐Lead Gen Platform
🏦Matched Lenders (3–5)
✅Chosen Lender
Step 1 — Single Form Submitted
Fills one form — loan amount, purpose, rough credit estimate. One form reaches many lenders.
Receives lead. Validates data quality. Sells lead to matched lenders simultaneously.
Step 2 — Platform Earns Fee Here (before any loan is made)
Earns per-lead fee — collected at this point, regardless of loan outcome. Platform's financial interest ends here.
Each lender independently underwrites the borrower's lead and prices an offer
Step 3 — Competing Offers Returned to Platform
Sees multiple offers displayed side by side — different APRs, terms, lenders. Compares and picks.
Aggregates and displays competing offers. Platform's operational role ends here.
Returns approved offer to platform for display to borrower
Step 4 — Borrower Chooses a Lender (Platform Exits)
Picks best offer. Redirected to chosen lender's site to complete full application — hard pull, e-signature.
Platform is out. No further involvement in the transaction.
Borrower completes full application directly. Lender runs own hard pull and final underwriting.
Step 5 — Loan Funded Directly by Chosen Lender
Funds arrive in bank account directly from chosen lender (1–3 days)
Funds loan from own balance sheet or its own capital sources. Earns interest spread over loan life.
Platform earns: Per-lead fee — collected when lead is sold. No dependency on loan outcome, no ongoing revenue, no balance sheet exposure.   |  Chosen lender earns: Interest spread over loan life.
Diagram 4

Lending vs. Payments vs. BNPL — The Core Distinction

These three categories all deal with money, but they solve fundamentally different problems. Understanding this distinction explains why a payments PM and a lending PM need completely different mental models.

Money through TIME The borrower keeps the money — but trades a future obligation for it today. The lender gives up liquidity now and earns it back slowly, with interest, over months or years. The core question is: will the future happen as predicted? Risk is invisible at origination and only reveals itself 12–24 months later.
Money through SPACE Money teleports from one account to another — same moment, different location. No credit extension, no time horizon, no repayment. The core question is: did this transaction actually happen, and was it authorized? Risk is immediate (fraud, failed auth) and resolves in seconds.
Both simultaneously BNPL moves money through space immediately (pays merchant now) and through time concurrently (shopper repays over weeks). It is a payment method and a credit product in a single transaction. Both risks exist at once: fraud at the point of sale + credit risk on repayment.
Lending Money through TIME
Today
Lender gives borrower a lump sum
↓
Months 1–60
Borrower pays back in installments
↓
Lender earns
Original principal + interest over time
Core risk: Credit risk — will they repay? Shows up 12–24 months later.
Key metric: Default rate, vintage performance, APR spread
Layers activated
EU Borrower L3A Lending Products L2 Bank Data API L2 Money Movement L1 ACH / Fedwire Capital Sources
Payments Money through SPACE
Shopper taps card
at merchant terminal
↓ (~1.5 seconds)
Network routes it
Card network → acquiring bank → issuing bank
↓
Merchant receives
Purchase amount minus processing fee
Core risk: Fraud + settlement risk — did this transaction actually happen?
Key metric: Authorization rate, fraud rate, transaction volume
Layers activated
EU Shopper + Merchant L3B Payments L2 Card Issuing L1 Card Networks L1 ACH (settlement) No capital sources
BNPL Both — payment method + credit product
Shopper checks out
Selects installment option at checkout
↓ (milliseconds)
BNPL underwrites in real-time
Credit decision in under 1 second
↓
Merchant paid immediately
BNPL absorbs the credit risk
↓
Shopper repays in installments
Often 4 payments, often 0% APR
Both risks: Fraud at point of sale + Credit risk on repayment
Key metric: Merchant fee rate, repayment rate, repeat usage
Layers activated
EU Shopper + Merchant L3C BNPL L2 Bank Data API L2 Money Movement L1 ACH / Card Networks Capital Sources (BNPL warehouse)
Diagram 5

The Data Flywheel — Why the Market Leader Keeps Winning

More loans → more repayment data → better model → better outcomes → more loans. This virtuous cycle compounds over years and becomes the primary competitive moat in AI-driven lending.

Scope: Models 3 & 4 only (AI Platform and Marketplace Bank). This flywheel only exists where the platform does its own underwriting and accumulates labeled outcome data. Model 1 (Lead Gen) has no flywheel — it never underwrites, so it collects no repayment signal. Model 2 (P2P/Institutional) has a weak version — the platform underwrites and learns, but typically uses coarser risk grades rather than continuous ML signals, and institutional investors increasingly do their own analysis. The flywheel as described here — where proprietary signals compound across millions of loans — is specific to platforms that own the full underwriting stack.

More loan originations

Each approved and funded loan generates a new repayment stream — a live experiment on whether the model priced risk correctly.

More repayment data collected

Who paid on time? Who defaulted? Under what economic conditions? Across which borrower segments? This labeled data is the raw material for model improvement.

Better model calibration

More outcome data → signals can be weighted more precisely → the model improves its ability to distinguish good risk from bad risk within similar credit profiles.

More accurate underwriting

Lower false positives (bad borrowers approved) and fewer false negatives (creditworthy borrowers rejected). Approval rates rise. Loss rates fall.

Better outcomes for all parties

Borrowers get lower APRs. Capital partners earn better risk-adjusted returns. More capital partners join. More committed volume. Platform fees stay stable.

More borrowers accept → back to Step 1

Better offers → higher acceptance rate → more originations → more training data. The cycle accelerates with scale.

Why this matters for PMs

🔒
The moat is the data, not the algorithm. A competitor can hire the same ML engineers and use the same architecture. They cannot replicate years of loan performance data across millions of loans in every economic condition.
⚠️
The flywheel can run backwards. Bad model → adverse selection → capital partners lose money → they pull back → fewer loans → less data → worse model. This failure mode has played out at multiple AI lenders during rate cycles.
⏱️
Feedback is delayed 12–24 months. You won't know if today's model changes were right until loans from today mature. Early Payment Default (EPD) rate is the best leading indicator — defaults within 30–90 days signal problems early.
🆕
New segments temporarily break the flywheel. Expanding to a new borrower segment (prime borrowers, auto loans) means less historical data on that segment. Expansion trades near-term model accuracy for long-term data accumulation.
📊
Capital partners are the external validator. If they keep committing capital, the flywheel is healthy. If they pull back, the model's performance isn't meeting their return expectations — often a leading indicator before internal metrics show it.
Diagram 6

How Money Actually Moves — Technology Stack vs. Capital Flow

Capital sources are not a technology layer — they're a funding relationship. They connect directly to payment rails as regulated financial institutions, bypassing the middleware that regular fintechs need.

How Fintechs Reach the Rails
A company without a bank charter uses Layer 2 APIs to access Layer 1 rails.
Fintech Product (Layer 3)
Lending / Payments / BNPL App
Needs to disburse a loan or settle a payment — but has no Fed master account or direct rail access.
↓
API call
Infrastructure API (Layer 2)
Money Movement / Card Issuing / Data
Translates the API call into rail-specific commands. Handles compliance, routing, error handling, reconciliation.
↓
ACH file / card authorization / wire
Payment Rails (Layer 1)
ACH / Fedwire / RTP / Card Networks
The actual infrastructure that moves money between bank accounts. Regulated by the Federal Reserve, Nacha, and card networks.
↓
1–3 days (ACH) · Instant (RTP) · 1.5s (Card)
End User
Borrower / Merchant / Shopper
Money arrives in their bank account. They never see the infrastructure layers that made it happen.
How Capital Moves (Institutional Flow)
Capital sources are regulated institutions — they already have direct rail access. No Layer 2 needed.
Capital Source
Institutional Investor / Hedge Fund
Holds a custodian bank account with direct Fedwire access. Wires purchase price for whole loan acquisitions.
↓
Fedwire — same-day, large-value
Bank Partner (lender of record)
Licensed Bank or Credit Union
Receives wire. Transfers loan ownership. Uses deposit pool to fund new loans. Earns origination fee before selling.
↓
ACH (1–3 days) or RTP (instant)
Borrower
Loan Recipient
Loan proceeds arrive in bank account. Monthly repayments leave via ACH pull, collected by servicer, forwarded to loan owner.
↑
ACH pull — monthly over 3–7 year loan life
Capital Partner (loan owner)
Receives Principal + Interest
Servicer collects and forwards repayments. Capital partner earns yield above Treasuries over the life of the loan.
↕
DTCC — if loans are securitized into ABS
Optional — Securitization
Asset-Backed Securities Market
Loan pools can be bundled into ABS and sold to bond investors. Settlement via DTCC — entirely different infrastructure from payment rails.
💡
Why capital sources bypass Layer 2: Infrastructure APIs (Layer 2) exist because most fintechs are NOT banks — they don't have a Fed master account or direct rail connections. Capital sources (hedge funds, insurance companies, bank partners) ARE licensed financial institutions with direct Fedwire and ACH access. They're peers of the rails, not customers of them. The technology stack describes how fintechs access the financial system. Capital flow is a separate axis — it describes who funds the products those fintechs offer.
Diagram 7

User Journeys — Five Scenarios

What actually happens at each step, from the user's perspective and behind the scenes. Green steps = money moves. Purple steps = a model or decision runs.

Journey 1
Taking out a personal loan (AI Platform model)
Borrower applies online, AI underwrites, bank partner funds, capital partner buys
Application: 5–10 min · Funds: 1–3 business days
Borrower
Enters loan amount, purpose, income estimate. Clicks "Check your rate" — no credit impact yet.
AI Platform (soft pull)
Pulls soft credit inquiry. Runs initial signals. Returns estimated APR range in seconds. No credit impact.
Borrower
Sees APR range. Decides to continue. Connects bank account for income verification. Agrees to hard pull.
AI Model + Bank Partner (hard pull)
Full underwriting runs across hundreds of signals. Model outputs firm APR. Bank reviews compliance, becomes lender of record. Majority fully automated.
Borrower
Receives firm offer: loan amount, APR, monthly payment, term. E-signs. Total process: ~8 minutes.
Bank Partner → Borrower's bank
Bank disburses funds (minus origination fee) to borrower's bank account.
ACH — 1–3 business days
Capital Partner → Bank Partner
Bank sells loan (whole loan sale). Capital partner wires purchase price. Now holds credit risk and earns all future interest.
Fedwire — same-day
Rails: ACH (disbursement) Fedwire (loan sale) Platform earns: referral + platform fee. No balance sheet risk.
Journey 2
Buying with BNPL at checkout
Shopper selects installment option, BNPL underwrites in real-time, merchant paid same day
Checkout decision: <30 sec · Merchant paid: same day · Shopper pays over weeks
Shopper
Selects items, proceeds to checkout. Sees "Pay in 4" installment option. Selects it.
BNPL provider (real-time underwriting)
Runs soft credit check + proprietary signals in under 1 second. Approves installment plan. Merchant discount rate covers the 0% APR to shopper.
Shopper
Sees approval. Confirms first payment + schedule. Quick verification step. 20 seconds total.
BNPL provider → Merchant
BNPL pays merchant the full purchase amount immediately, minus merchant discount rate (2–8%). Merchant takes zero credit risk.
ACH — 1–2 days to merchant
Shopper → BNPL provider (x4)
Payment pulled from shopper's bank every 2 weeks. 0% APR = no interest, but late fees apply for missed payments.
ACH pull — 4 installments
Rails: ACH (merchant payout + repayments) BNPL earns: merchant fee (2–8%). Merchant earns: higher conversion. Shopper pays: $0 if on time.
Journey 3
Paying for something with a card
Shopper enters card at checkout, card network routes it, merchant settles 2 days later
Authorization: ~1.5 seconds · Settlement to merchant: 2 business days
Shopper
Enters card details at checkout. Clicks pay. Has no visibility into what happens next.
Payment processor (e.g. Stripe)
Encrypts card data. Sends authorization request to card network. Handles PCI compliance — merchant never touches raw card data.
API → Card network
Card network → Shopper's bank (issuer)
Network routes request to shopper's issuing bank. Bank checks: sufficient funds? Fraud signals? Approves or declines. Total: ~1.5 seconds.
Card Network
Shopper
Sees "Payment confirmed." Gets receipt. Charge shows as "pending" on bank statement. Transaction done from their perspective.
Card network settlement → Merchant
Settlement runs overnight in batch. Merchant receives purchase amount minus processing fee (~3%). Charge finalizes on shopper's statement.
Card settlement + ACH — 2 business days
Rails: Card network (auth + settlement) ACH (final bank transfer) Processor earns: ~3% fee split across processor, card network, issuing bank.
Journey 4
Depositor-funded lending (marketplace bank model)
One customer deposits savings, another takes a loan. Bank earns the interest rate spread between both.
Deposit: 1–2 days · Loan approval: minutes · Funds: 1–2 days · Bank earns spread over years
Depositor → Marketplace Bank
Transfers savings into bank's high-yield account. Bank now has low-cost deposit funding — the core advantage of the bank charter model.
ACH — 1–2 business days
Borrower
Applies for a loan directly with the bank. No third-party bank partner needed — the bank IS the lender of record.
Bank (underwriting)
Runs credit check, income verification using own models. Often targets prime/super-prime borrowers. Approves loan. Will hold on balance sheet.
Bank → Borrower
Funds loan from its deposit pool. Borrower receives proceeds (minus origination fee).
ACH — 1–2 business days
Borrower → Bank (monthly)
Borrower repays monthly over the loan term. Bank earns loan APR (~10%), pays depositor savings rate (~4%). The ~6% spread is the margin.
ACH pull — monthly over loan term
Rails: ACH (deposit + disbursement + repayments) Bank earns: interest spread. No capital marketplace needed — cleaner margin, but bank holds all credit risk.
Journey 5
Finding a loan via Lead Gen marketplace (Model 1)
Borrower submits one form, receives competing offers from multiple lenders, leaves the platform to complete their chosen loan
Form: 2–3 min · Offers returned: seconds · Full application with lender: 5–10 min · Funds: 1–3 days
Borrower
Visits lead gen platform. Fills one short form — loan amount, purpose, rough credit estimate, contact info. No SSN yet. No credit impact.
Lead Gen Platform
Validates lead quality. Matches borrower profile to relevant lenders. Sells the lead simultaneously to 3–5 matched lenders. Platform earns its fee here — before any loan decision.
Matched Lenders (independently, in parallel)
Each lender runs its own pre-qualification using the lead data — soft pull, internal scoring. Each returns an offer: APR range, loan amount, term. Lenders compete on rate.
Borrower
Sees a comparison table of offers — multiple lenders, different APRs and terms side by side. Picks the best one. Key difference from other models: borrower leaves the platform at this step.
Borrower → Chosen Lender's website
Redirected to the chosen lender to complete the full application — SSN, employment details, bank account. Hard pull runs here. E-sign loan agreement. Platform has no further role.
Chosen Lender → Borrower
Lender funds loan from its own capital sources. Disburses directly to borrower's bank. Platform never touched this money — it exited at Step 4.
ACH — 1–3 business days
Rails: ACH (disbursement by chosen lender only) Platform earns: per-lead fee (Step 2). No credit risk. No ongoing relationship. The borrower's experience after Step 4 belongs entirely to the lender.
Reference

Financial Components — PM Lens

Core systems PMs actually need to reason about: where risk accumulates, where margins live, and where product decisions change unit economics.

How this connects to the diagrams: Layer Map shows where each engine lives, Loan Flow shows when it is invoked, and Money Flow shows how it impacts margin and cash timing. Use this section as the PM operating view.

Core Engine Risk Decisioning & Underwriting
What it doesEvaluates default risk, fraud risk, and expected lifetime value before approving capital.
PM metricsApproval rate, first-payment default, expected loss %, contribution margin per funded account.
Failure modeGrowth team pushes approvals up, losses spike 1-2 quarters later.
Core Engine Servicing & Collections
What it doesPayment reminders, delinquency handling, hardship workflows, and recovery orchestration.
PM metricsRoll rates (30→60→90 DPD), cure rate, recovery %, servicing cost per account.
Failure modeToo aggressive collections hurts brand/trust; too soft policy hurts loss performance.
Core Engine Ledger, Reconciliation & Controls
What it doesKeeps the ground truth of balances, transactions, fees, payouts, and adjustments.
PM metricsBreak rate, reconciliation latency, unresolved exceptions, financial close timeline.
Failure modeOperational complexity explodes; teams can’t trust numbers; audit pain arrives fast.
Core Engine Capital & Funding Management
What it doesMatches product demand with warehouse lines, bank partners, deposits, or securitization windows.
PM metricsCost of funds, capacity utilization, take rate net of funding, velocity to deploy capital.
Failure modeGreat UX but no funding headroom = hard growth ceiling.

Interview shortcut: when asked about fintech strategy, split the answer into distribution (acquisition), decisioning (risk + approvals), and balance sheet mechanics (funding + losses). Most candidates stop at UX; strong PMs speak in unit economics.

Common PM Failure Patterns

Approval-rate vanity

Chasing top-line approvals without tracking 90-day loss cohorts creates delayed blowups.

Unit economics blind spot

Teams optimize conversion while ignoring cost of funds, servicing cost, and charge-off trajectory.

Reconciliation debt

Shipping money movement features fast without ledger controls creates exception backlog and trust erosion.

Collections policy whiplash

Frequent policy swings confuse ops and customers, hurting both recovery and NPS.

Decision Matrix — What to Optimize For

SituationOptimize ForGuardrail MetricsAvoid
Early growth phaseAcquisition velocity + fast approvalsFPD, fraud rate, CAC paybackLoose underwriting with no cohort gating
Rising delinquenciesPortfolio quality + recovery execution30/60/90 DPD roll rates, cure rateRate hikes only (without servicing fixes)
Funding pressureMargin durability + capital efficiencyNet take rate, cost of funds, utilizationUnprofitable promo-led growth
Audit/compliance heatControl integrity + traceabilityBreak rate, unresolved exceptions, close timeBypassing ledger/process controls

Interview Scenarios + Strong Answer Angles

“Approval dropped 12 points after model update. What do you do?”

Angle: segment by risk band + channel, run champion/challenger rollback logic, protect loss guardrails before restoring volume.

“Delinquency is rising but growth target is fixed.”

Angle: propose dual-track plan: tighten thin-file bands + improve repayment UX/collections ops, align execs on risk-adjusted growth.

“Finance says numbers don’t match dashboards.”

Angle: establish single ledger truth, reconciliation SLA, and incident taxonomy before adding new reporting layers.

Financial Ecosystem Visual Guide v2 (Generic) · PM Prep Papers Series · fullstackpm.tech · March 2026