Marketplace Clearing Simulator — Upstart Model

Interactive model of how borrowers, bank partners, capital partners, and Upstart's AI clearing engine interact

📊 See Data & Methods for sources, data generation logic, and case studies.

Borrower Population
Volume500 loans
Avg Credit QualityFICO 660
APR SensitivityMedium
Underwriting Model
Model 18: APR fed back into risk model as a feature — finds near-prime borrowers underpriced by FICO alone. Upstart reports 101% more approvals, 38% lower avg APR vs. traditional models.
Capital Partners
Balance sheet is always last resort. Click to toggle a partner.
Dataset
📊 Unified Borrower Generation: Both this simulator and the Data & Methods page use identical borrower generation logic (Box-Muller FICO + LogNormal loan amounts + hidden-prime detection). Export from either page and compare side-by-side.
Clearing Rate
Avg Cleared APR
Balance Sheet Exp.
Volume Cleared
Waterfall Distribution
Where funded loans come from
Capital Partner Utilization
% of committed capacity deployed
Loan Distribution: FICO vs APR
Each dot = one borrower. Color = where it cleared or why it didn't.
Forward-flow Bank partner Spot market Balance sheet APR rejected No eligible partner
📚 Key Sources

Upstart: S-1 filing (Dec 2020), Q1–Q2 2022 earnings calls

Industry: Federal Reserve consumer credit, TransUnion research

Academic: XGBoost (Chen & Guestrin 2016), Fairness in ML (Barocas et al. 2021)

Full details: See Data & Methods page →

Borrower Journey

From first Google search to final payment — 9 stages, each with a different product, data requirement, and failure mode. Most borrowers don't experience all 9 in sequence; the drop-offs at each stage define the funnel that the Marketplace Optimization team owns.

STAGE 01
Acquisition
🔍
Discovery
Borrower searches for "personal loan" or "debt consolidation." Finds Upstart via SEO, NerdWallet/Credit Karma affiliate, or direct referral.
Details
  • ~45% of traffic from organic search
  • ~30% from affiliate/referral partners
  • ~25% direct or paid
  • No data collected yet
⚠ Trust gap — borrowers may not know Upstart vs. big banks
STAGE 02
Acquisition
📋
Pre-Qualification
Borrower enters: loan amount, purpose, name, address, income, employment. Upstart runs a soft credit pull — no credit impact.
Details
  • Soft pull = no FICO impact (key trust signal)
  • Data: SSN partial, employment type, self-reported income
  • 1800+ signals begin loading from bureau file
  • Takes ~2 minutes to complete
⚠ Form drop-off; confusion about "soft vs hard pull"
~65% reach this stage from landing
STAGE 03
Underwriting
🤖
AI Risk Assessment
Model 18 runs. APR is fed back as a feature — finds the lowest APR where P(default|borrower, APR) satisfies capital partner return floors.
Details
  • 1M+ predictions per borrower per Upstart
  • Model tests multiple APR scenarios
  • Hidden-prime borrowers identified here
  • ~91% of loans fully automated (no human review)
⚠ ~35% of borrowers don't qualify or are priced out
STAGE 04
Underwriting
💬
Rate Offer Display
Borrower sees their personalized APR, term options (36 or 60 months), and monthly payment. Multiple options shown simultaneously.
Details
  • If APR > borrower's tolerance: they leave (APR rejection)
  • Borrower may comparison-shop at this point
  • Payment calculator shown for each term
  • No credit impact yet — still reversible
⚠ Rate-sensitive borrowers shop LendingTree/competitors
~70% who receive an offer view it seriously
STAGE 05
Underwriting
Acceptance (Hard Pull)
Borrower accepts terms. Hard credit inquiry filed — now on their credit report. This is the commitment point. ~60% of offer recipients accept.
Details
  • Hard pull: FICO impact of ~5 points, fades in 12 months
  • Borrower shown explicit consent language
  • APR locked at this point
  • Marketplace now begins clearing process
~60% of offer recipients accept
STAGE 06
Verification
🔗
Income & Identity Verification
Borrower connects bank account via Plaid (preferred) or uploads pay stubs/tax docs. Identity verified via ID scan.
Details
  • Plaid: instant, real-time income + cash flow signals
  • Pay stub upload: adds 1-3 day processing lag
  • ~12-15% of applications fail or delay here
  • Fraud scoring runs in parallel
⚠ Highest-friction step; Plaid failures cause abandonment
STAGE 07
Clearing
Marketplace Clearing
Loan routed through eligibility matrix → pricing engine → waterfall router. Matched to a capital partner in seconds. Bank originates.
Details
  • Eligibility matrix: which partners can fund this loan?
  • Waterfall: forward-flow first, then spot, then balance sheet
  • Bank partner is lender of record for origination
  • Bank immediately sells to capital partner
~91% auto-cleared; ~9% to manual review
STAGE 08
Clearing
💸
Disbursement
Loan funds deposited to borrower's bank account via ACH (1–3 days) or RTP (same-day, if available). Welcome email + payment setup prompt sent.
Details
  • ACH: standard, 1-3 business days
  • RTP (Real-Time Payments): instant, growing coverage
  • Auto-pay setup: reduces delinquency risk
  • ~80% of funded borrowers enroll in auto-pay
~95% who clear verification get funded
STAGE 09
Lifecycle
📅
Repayment (36–60 months)
Monthly payments via ACH. Serviced by Upstart. Capital partner earns interest. Early Payment Default (EPD) in first 90 days is the fastest model quality signal.
Details
  • EPD = default within 90 days → signals model error
  • Upstart services loan throughout (collects, reports)
  • Hardship deferral options reduce default rate
  • Post-payoff: re-engagement flow for repeat loan
⚠ EPD spike = fastest signal that something broke in underwriting

Approval & Acceptance Funnel

100% — visit rate check page
↓ 65% — complete pre-qualification
↓ 45% — receive a rate offer (qualify)
↓ 27% — accept offer (hard pull)
↓ 24% — complete verification
↓ 23% — funded ← this is what you optimize

Where Does the Simulator Fit?

The simulator models Stages 03–07: AI assessment → offer → acceptance → clearing. It does not model acquisition (traffic), verification friction, or the repayment lifecycle. Those stages are equally important — just owned by different teams. The Marketplace Optimization PM owns the clearing mechanism at the center of this funnel.

Capital Partner Journey

Capital partners are the supply side of the marketplace. They provide the funding that makes every loan possible. There are four structural types — each with different incentives, risk tolerances, and relationship dynamics. The Marketplace Optimization team must simultaneously satisfy all of them.

Forward-Flow

Eltura / Aperture

Pre-committed volume. Get first allocation in waterfall. Upstart's most stable capital source.

Target return: 7–9% net
Commitment: $500M–$600M
Contract: 12–24 months
Priority: 1st in waterfall
Spot Market

Hedge Funds / Asset Mgrs

No commitment — buy individual loans on demand. Require higher APR floor to compensate for no volume guarantee.

Target return: 9–12% net
Commitment: None (bid-by-bid)
FICO floor: 680+
Priority: 3rd in waterfall
Bank Partner

Regional Banks / CUs

Lower cost of capital (deposit funded). Higher FICO requirements. CRA credit benefit for serving underbanked.

Target return: 6–8% net
Commitment: $100M–$200M
FICO floor: 660+
Priority: 2nd in waterfall
ABS / Securitization

Structured Credit Investors

Buy rated bonds backed by loan pools. Require pool size, rating, and credit enhancement. Longer setup time.

Target return: 5–8% (rated)
Pool size: $100M+ per deal
Setup: 2–4 months
Priority: 4th (pool-based)
PHASE 01
Setup
🔬
Due Diligence
Partner reviews Upstart's model back-tests, vintage performance (actual vs. predicted defaults), and regulatory structure.
Details
  • Back-test: how did model predict vs. actual for 2019–2024 cohorts?
  • Stress test: how did loans perform in 2022 rate shock?
  • Model explainability: can risk team understand what drives the score?
  • Timeline: 1–3 months
⚠ Black-box models scare risk committees — explainability matters
PHASE 02
Setup
📝
Agreement Structure
Negotiate structure: forward-flow commitment vs. spot purchase. Define return floors, EPD provisions, repurchase obligations.
Details
  • Forward-flow: commit $X at return floor Y% for Z months
  • EPD clause: Upstart repurchases loans defaulting <90 days
  • Performance trigger: if actual default > predicted by X%, partner can exit
  • 40–100 page purchase agreement
PHASE 03
Setup
⚙️
Eligibility Config
Partner specifies their criteria: minimum FICO, APR floor, loan size range, geography, loan purpose. Live volume preview shown.
Details
  • Narrow criteria = fewer loans, more targeted exposure
  • Wide criteria = more volume, more variance
  • Preview: "X loans/month at these criteria historically"
  • Today often manual (opportunity: self-serve configurator)
⚠ Config errors cause adverse selection — test before going live
PHASE 04
Active
💰
Active Deployment
Loans allocated per waterfall priority. Partner receives matching loans in real-time. Dashboards show deployed capital and loan detail.
Details
  • Forward-flow: first look at matching loans
  • Daily reconciliation of funded loans
  • Capacity tracking: pacing toward commitment
  • Whole loan detail available for due diligence
Healthy: 80–95% commitment utilization pace
PHASE 05
Review
📈
Performance Monitoring
Monthly: actual default rate vs. model-predicted. EPD tracked at 30/60/90 days. Cohort-by-cohort attribution. Partner gets full reporting access.
Details
  • EPD = strongest early signal of model quality issue
  • If actual > predicted by >15%: escalation required
  • Partner can request segment rebalancing
  • Benchmark: compare vs. comparable credit products
⚠ Opacity = trust erosion. Partners who can't see performance leave.
PHASE 06
Review
🔄
Renewal / Expansion
6 months before commitment expiry, renewal negotiation begins. Outperformers expand. Underperformers renegotiate structure or exit.
Details
  • Upstart presents: actual return vs. target, cohort by cohort
  • If outperforming: partner typically expands commitment 20–50%
  • If underperforming: negotiate remedies, partial repurchase, model fix timeline
  • Net revenue retention of capital partners is a key business health metric
Healthy renewal rate: >80% of forward-flow partners renew

Adverse Selection

Being routed the worst loans while better loans go to spot market buyers. This happens when waterfall routing logic is misconfigured. The fix: transparent eligibility matrices + balanced allocation across tiers. If a partner suspects adverse selection, they leave — and they tell others.

Model Deterioration

When actual default rates diverge upward from predicted. Can happen from: model drift (borrower population shifted), channel mix change (riskier acquisition), or macro shock (rate spike). EPD is the fastest signal — loans defaulting within 90 days are the strongest model quality indicator.

Bank Partner Journey

The bank is not optional — it's structurally required by US banking law. Understanding why, and what the bank actually does in each transaction, is essential for understanding why Upstart's marketplace is built the way it is.

🏛 Why the Bank Is Always Required

In the United States, only a federally or state-chartered bank can legally originate a consumer loan. A hedge fund, investment firm, or fintech platform — no matter how sophisticated — cannot make a loan directly to a consumer. They can own loans. They cannot make them.

The second reason is even more important for marketplace economics: a national bank can "export" its home state's interest rate laws to any borrower in any state (the "valid when made" doctrine, from a 1978 Supreme Court case). Upstart partners with banks chartered in states with no usury cap (APR ceiling). This lets Upstart's AI model operate as one unified marketplace with consistent pricing across all 50 states. Without bank partners:

  • Upstart would face different APR caps in every state — some as low as 10%
  • Their AI model might price a loan at 28% for a near-prime borrower in a strict-cap state and be legally unable to make it
  • The marketplace would fragment into 50 separate state products — killing the data flywheel

The bank is not a rubber stamp. It must genuinely underwrite and approve each loan, maintain fair lending documentation, and is subject to regulatory examination on its third-party lending program. It earns an origination fee (~1–3% of loan amount) and has zero credit risk — it sells the loan immediately.

PHASE 01
Regulatory
📜
Third-Party Program Setup
Bank applies for regulatory approval to offer Upstart-powered loans. OCC or FDIC reviews the third-party lending program before any loans are made.
Details
  • Bank submits credit policy to regulator for approval
  • BSA/AML (fraud + money laundering) procedures documented
  • Third-party risk management framework required
  • Timeline: 3–6 months for new bank partners
⚠ Regulatory approval is the biggest barrier to adding new bank partners fast
PHASE 02
Operations
✍️
Origination
Each loan application is reviewed and approved by the bank under its approved credit policy. Bank's name appears on the promissory note — it is the legal lender.
Details
  • Bank may delegate approval to Upstart within defined policy guardrails
  • All TILA (Truth in Lending Act) disclosures issued by bank
  • Fair lending documentation maintained per ECOA requirements
  • Each loan carries bank's name and routing/account info
PHASE 03
Operations
💸
Funding
Bank disburses loan proceeds from its own accounts to the borrower. Holds loan on its balance sheet for hours to a few days — never intended to hold long-term.
Details
  • ACH disbursement: 1–3 business days
  • RTP (Real-Time Payments): instant settlement, growing availability
  • Bank's balance sheet exposure is brief — measured in days
  • Credit risk clock starts ticking at disbursement
PHASE 04
Operations
📦
Whole Loan Sale
Bank sells the loan to a capital partner (or Upstart's balance sheet). Transfer is near-immediate. Bank receives principal + origination fee and exits credit exposure.
Details
  • Whole loan sale: capital partner buys the entire loan
  • Upstart handles all servicing post-sale (collections, reporting)
  • Bank retains: representations and warranties on loan quality
  • EPD repurchase obligation: if loan defaults within 90 days, bank may need to repurchase
PHASE 05
Economics
💼
Ongoing Fee Income
Bank earns: origination fee per loan (1–3%), CRA credit for underbanked lending, growth without balance sheet risk. Regulatory obligations continue.
Details
  • CRA = Community Reinvestment Act: regulators reward banks for lending to underserved communities
  • Bank must file periodic exam reports on third-party program
  • If regulator flags issues: bank can wind down with 90-day notice
  • Exit risk: bank regulatory pressure is Upstart's #1 supply-side concentration risk
⚠ Bank partner exit due to regulatory pressure = origination capacity gap, hard to replace quickly

Bank Partner (Lender of Record)

Role: Legally makes the loan. Issues disclosures. Holds briefly.

Earns: Origination fee (1–3% per loan). Zero credit risk.

Regulated by: OCC, FDIC, CFPB. Subject to exam.

Cares about: Regulatory compliance, not losing charter, fair lending documentation.

Capital Partner (Loan Buyer)

Role: Buys the loan from the bank. Takes on credit risk. Earns interest over 36–60 months.

Earns: Interest income minus defaults. Net 7–11% target return.

Regulated by: SEC (if fund). Not subject to banking regulation.

Cares about: Model accuracy, adverse selection, EPD rates, consistent volume.

The Clearing Engine

This is what the Marketplace Optimization PM owns. Three layers — eligibility, pricing, routing — each with distinct logic, distinct failure modes, and distinct PM levers. A loan must pass all three to clear.

Input

Loan Application + Signals

FICO + tradelines Plaid income signals Employment history Cash flow patterns Loan amount + purpose
1,800+ features passed to Layer 1
Layer 1 — Deterministic Rules

Eligibility Matrix

Which capital partners can fund this loan? Binary checks — no ML here.

FICO ≥ partner minimum? Loan size in range? Purpose allowed? Geography approved? Capacity remaining?
Output: List of eligible capital partners for this loan. If empty → loan cannot proceed regardless of APR.
✗ Fail: No eligible partner → rejected (no_partner)
Layer 2 — AI Optimization (Model 18)

Pricing Engine — APR as a Feature

Novel: APR is fed back into the risk model as an input, not just an output.

P(default | borrower, APR) for multiple APR scenarios Lower APR → lower payment → lower P(default) Find APR where risk-adj. return ≥ partner floor AND borrower accepts (APR ≤ max tolerance)
Output: Optimal clearing APR — the lowest price where both sides agree. Hidden-prime borrowers get significantly lower APRs than FICO alone would suggest.
✗ Fail: No APR clears both sides → apr_too_high
Layer 3 — Waterfall Routing

Allocation Engine

Given a clearing APR, which capital partner gets this loan? Priority-based routing.

① Forward-flow (Eltura, Aperture) — committed, first access ② Bank programs (WestBank etc.) — lower cost of capital ③ Spot market — no commitment, higher APR floor ④ Balance sheet — last resort, Upstart funds it
Output: Partner assignment + loan funded. Clearing rate = % of loans that reach this output successfully.
Post-Clearance

Funded Loan + Feedback Loop

Bank disburses via ACH/RTP Bank sells to capital partner Upstart earns platform fee EPD monitoring starts (day 1) Actual vs. predicted tracking begins
Data flywheel: repayment outcomes feed back into model calibration → better predictions → lower APRs → more loans clear → more data.
APR Too High for Borrower

Model prices loan at 28% APR. Borrower's tolerance is 24%. Loan doesn't clear — borrower rejects. Demand-side failure.

PM fix: Improve model accuracy (APR-as-feature finds lower clearing price)
No Eligible Capital Partner

Borrower's FICO (610) or loan purpose (small business) doesn't meet any partner's criteria. Supply-side failure — no price fixes this.

PM fix: Onboard partners with broader criteria or lower FICO floors
Forward-Flow Capacity Exhausted

Eltura hits their monthly commitment cap. Loan falls to spot market (higher APR required) or balance sheet. Clearing rate doesn't drop — but quality of placement does.

PM fix: Pacing alerts; expand commitment or diversify partners
Bank Policy Rejection

Upstart model approves the loan. Bank partner's own credit policy rejects it. Creates "late decline" — worst UX (borrower thought they were approved).

PM fix: Align Upstart model + bank policy criteria; earlier bank-policy pre-check
Adverse Selection Cascade

Routing error sends worst-performing loans to forward-flow partners, best to spot market. Partners see EPD spike. Begin exit process. Capital supply collapses.

PM fix: Balanced allocation testing; segment-level performance monitoring per partner
Decision Log

Assumptions & Notes

What this simulator models accurately, what it simplifies, and what real Upstart does differently. Knowing the gap between a model and reality is as important as knowing what the model shows.

⚠ What This Simulator Simplifies

1
Borrower population uses normal distribution — Real data has complex bimodal clusters (prime and near-prime) with different behavioral patterns.
2
"Hidden prime" is a binary flag — Real Model 18 uses continuous risk scoring across 1,800+ signals including income volatility, employment tenure, geographic cash flow patterns.
3
APR is FICO-band lookup + offset — Real Model 18 is a gradient boosting model with non-linear interactions between hundreds of features. APR-as-feature means it runs iterative scenarios, not a lookup table.
4
Eligibility is FICO floor + APR floor only — Real partners specify loan purpose, geography, debt-to-income ratio, employment type, and dozens of other segment criteria.
5
Clearing is instantaneous — Real marketplace has timing windows: forward-flow partners get first-look periods (hours), spot market runs in batches, some loans wait overnight for bids.
6
No income verification step — Real system: ~12–15% of accepted applications fail or delay at Plaid/income verification, reducing actual funded rate from simulated rate.
7
No fraud detection layer — Real system runs parallel fraud scoring that declines applications post-offer. Affects near-prime segment disproportionately.
8
Single APR per borrower — Real system offers 36-month and 60-month options simultaneously. Borrower choice affects actual default risk (shorter term → higher payment → different risk profile).
9
Static capital partner preferences — Real partners update eligibility criteria frequently based on portfolio performance, macro conditions, and capital availability. This is a dynamic, not static, supply.
10
No macro factors — Fed rate changes directly affect capital partner return requirements. A +100bps rate increase can shift all partner APR floors upward, collapsing clearing rate for near-prime borrowers within weeks (this is exactly what happened in 2022).
11
No EPD feedback loop — Real system feeds EPD data back into model calibration within 30–90 days. The data flywheel is not represented; simulator is a static snapshot, not a dynamic system.
12
Balance sheet is always open — Real Upstart's balance sheet capacity is constrained by its own capital position. During 2022–2023, Upstart's balance sheet grew to ~$1B+ because it had to fund loans it couldn't place — creating real financial risk.

✅ What This Simulator Correctly Models

1
Three-sided marketplace tension — borrower APR tolerance, capital partner return floor, and Upstart's clearing rate must all align simultaneously.
2
APR-as-feature concept — AI model finds a lower clearing price for hidden-prime borrowers; these loans fail in classic model but succeed in Model 18.
3
Waterfall routing priority — forward-flow before bank before spot before balance sheet, with capacity constraints at each tier.
4
Supply vs. demand constraint distinction — the two failure modes (APR too high vs. no eligible partner) have different causes and different PM responses.
5
Capital crunch dynamics — removing forward-flow partners forces overflow to spot/balance sheet, raising APRs and reducing clearing rate.
6
Balance sheet as a cost signal — high balance sheet exposure is a leading indicator that the marketplace is not clearing efficiently.
7
Commitment capacity exhaustion — forward-flow partners hit their caps and overflow to lower-priority tiers, which is a real operational pattern.

📐 Key Parameters Used

FICO distribution: Normal(mean=creditQualitySlider, σ=65), clamped 520–820
Loan amount: Log-normal(μ=$12K, σ=0.45), clamped $2K–$45K
Hidden prime rate: 28% of FICO 580–720 borrowers have "hidden prime" flag
APR reduction (Model 18 for hidden prime): Normal(μ=8.5%, σ=1.5%) below classic FICO pricing
Classic FICO APR bands: 780+=10.5%, 740+=13%, 700+=16.5%, 660+=21.5%, 620+=27.5%, 580+=33%, <580=38%
Borrower max APR: Decreases with FICO; hidden prime borrowers are 5% more rate-sensitive (they know their true risk)
Capital partner capacities: Eltura $4M, Aperture $3M, WestBank $2M, SpotFund = unlimited in simulation (represents deep spot market)