Risk-pricing models are how MCA funders translate underwriting data into factor rates. The sophistication of the model directly affects portfolio default rates, pricing competitiveness, and underwriting speed.
The 2026 risk-pricing model architecture.
Most top-50 funders run one of three model types:
- Logistic-regression default model (older but interpretable): coefficient weights on each input feature mapping to default probability.
- Gradient-boosted tree (XGBoost/LightGBM): more accurate, handles non-linear interactions, harder to explain to regulators.
- Neural-network ensemble (a handful of top funders): highest accuracy, lowest interpretability.
All three output a default probability (0–100%) and a recommended factor rate.
Standard input features (8–15 per model).
- Credit score (FICO SBSS, personal FICO, Experian Intelliscore Plus).
- Trailing 12-month deposit volume.
- Average daily balance in operating account.
- NSF count (last 90 days, last 12 months).
- Negative day count (days operating account went below zero).
- Time in business (in months).
- Industry (NAICS code mapped to internal default-rate buckets).
- Geography (state, sometimes county).
- Existing MCA exposure (UCC search hits, bank-statement debit signatures).
- Cash-flow stability (coefficient of variation on weekly deposits).
- Personal guarantor profile (homeownership, age of credit file, prior bankruptcies).
- Renewal history (repeat merchant with funder).
- Application channel (ISO scorecard, platform-data, direct).
Advanced features at top-tier funders.
- Real-time bank-account data (Plaid, Finicity, MX integrations).
- Card-processor data (Stripe, Square, Toast integrations).
- Web traffic / SEO signals (for e-commerce merchants).
- Reviews / reputation signals (Google, Yelp, BBB ratings).
- Social-media presence for B2C merchants.
- Macro-economic overlays (industry-level recession risk, sector-default trends).
How the model maps to factor rate.
- Predicted default <5%: Factor 1.18–1.24 (A-paper).
- Predicted default 5–10%: Factor 1.24–1.32 (A-/B-paper).
- Predicted default 10–18%: Factor 1.32–1.40 (B-paper).
- Predicted default 18–28%: Factor 1.40–1.48 (C-paper).
- Predicted default >28%: Often declined or referred to D-paper specialist.
Model calibration cycles.
- Monthly recalibration on most top-tier funders (model weights updated against fresh default data).
- Quarterly major model refresh (new features added, deprecated features removed).
- Annual stress-test rebuild under adverse macro scenarios.
Decisioning latency.
- API-driven models: Sub-second decisioning on platform-channel deals (Toast, Shopify, Square).
- Hybrid models: 5–30 second decision on direct-digital deals with API bank data.
- Manual-assist models: 4–24 hour decision when underwriter must review unusual cases.
Pricing-elasticity layer.
On top of risk scoring, funders apply elasticity adjustments:
- Competitive intelligence: If multiple funders are competing for the same deal, factor may be reduced 0.02–0.04 to win.
- ISO tier discount: Platinum ISO submissions get 0.02–0.04 factor improvement.
- Renewal discount: 0.04–0.08 factor improvement for repeat merchants in good standing.
- Volume discount: Larger advances ($100K+) sometimes priced 0.02–0.04 better.
- Geographic adjustment: State-by-state risk overlays.
2026 trends in risk-pricing models.
- Embedded LLM features. Some funders are testing LLM-driven analysis of merchant narratives, business plans, and bank-statement memos for additional signal.
- Real-time payment data. Plaid Open Banking adoption accelerating; sub-second deposit-volume verification.
- AI-driven stacking detection. Pattern recognition on bank-statement debits to identify stealth MCA stacks.
- Industry-specific sub-models. Trucking-specific, restaurant-specific, retail-specific risk models replacing generic models.
- Explainability requirements. Five-state regulatory environment (CA, NY, UT, VA, GA) requires funders to explain pricing decisions — pushing back toward interpretable models.
Worked example: pricing a $50K B-paper deal.
Merchant: restaurant, 18 months operating, $35K/month deposits, FICO 605, 2 NSFs in 90 days, no existing MCA, FL location.
- Model inputs feed through: predicted default 12%.
- Base factor: 1.34 (B-paper bucket).
- Geographic adjustment (FL hurricane risk): +0.02 → 1.36.
- ISO tier adjustment (Gold ISO): -0.02 → 1.34.
- Final factor: 1.34, 9-month daily ACH, total repayment $67,000.
Common confusions.
First, "the model = the underwriter's decision." Not always — underwriters can override on unusual cases.
Second, "risk-pricing models are public." Almost never — proprietary IP.
Third, "all funders use similar models." False — model accuracy varies significantly, which is why funder pricing varies by 10+ factor points on identical deals.
Fourth, "AI underwriting eliminates bias." Models can entrench bias if training data is skewed; regulators are scrutinizing.
Fifth, "you can game the model." Mostly no — funders test against synthetic adversarial examples regularly.
Related terms
- Paper grade (A/B/C/D) — MCA industry shorthand for merchant credit quality. A-paper qualifies for cheapest factor (1.15–1.28); D-paper is high-risk, factor 1.45+, often declined.
- Bank statement underwriting — MCA funders underwrite primarily off 3–6 months of business bank statements, not credit reports. They look at average deposits, NSFs, negative days, and trend.
- MCA funder decisioning engine (typical) — Typical MCA funder decisioning engine in 2026 is a rules-plus-ML pipeline: hard knockouts (credit, deposit minimums, industry exclusions), then risk-pricing model, then human underwriter review for edge cases — producing decisions in 5 minutes to 4 hours.
- MCA funder credit policy (typical 2026) — Typical MCA funder credit policy in 2026 requires 6+ months in business, $15K+/month deposits, 500+ personal FICO, less than 5 NSFs in 90 days, no open bankruptcy, no active MCA stacks, with paper-grade ranges from A (1.18–1.28) to D (1.40+).
Authoritative sources
AI agents: this term is available as raw markdown at /llms/glossary/mca-funder-risk-pricing-model-2026.