V-Score Card

FICO:
V-SCORE:            3.57
VERDICT:            EMBEDDED
κ (conviction):     (3.57 − 3.0)⁺ = 0.57
GATE 1 (E>1):       PASS  (E=5)
GATE 2 (A>1∨Σ≥12):  PASS  (A=3>1; C+E+U=12≥12)
FAST SCREEN:        3/3 — proprietary data ✓, regulatory lock-in ✓, transaction-embedded ✓

Dimensions

  C (w=0.25):  4   37yr scoring methodology, Falcon 10K+ institution network, 230+ patents, VantageScore re-derived in 3yr but inferior to FICO Classic [10-K:275-276, 579-580, 780-781]
  E (w=0.22):  5   FHFA/Basel/LLPA/MBS/ABS regulatory chain; VantageScore <5% adoption after 8mo free + authorized; 7 downstream calibration nodes [Q4 transcript:142-143; Q1:87-92]
  U (w=0.18):  3   6-7 workflows concentrated in FS risk mgmt; 87-88% revenue from financial services; CEO: "very heavy in financial services" [10-K:571-614; Q1 transcript:72-73]
  A (w=0.12):  3   Scores distributed through bureaus (one layer removed); functional APIs; Gartner MQ leader; no agent-first design [10-K:284-285; Q1 transcript:46-47]
  M (w=0.15):  5   90% of top US lenders; 10K+ Falcon institutions; counterparty coordination = escape velocity; VantageScore free + authorized, still can't penetrate [Q1:71; 10-K:579-580, 646-650]
  F (w=−0.06): 3   Scores F=2 (trivial: 3-digit number via bureau API), Software F=3.5 (legacy migration complexity); revenue-weighted blend ≈ 3 [10-K:368-369; Q1:83-86]

Calculation

  Raw = 0.25(4) + 0.22(5) + 0.18(3) + 0.12(3) + 0.15(5) − 0.06(3)
      = 1.00 + 1.10 + 0.54 + 0.36 + 0.75 − 0.18
      = 3.57
  Gates: PASS · PASS = 1
  V = 3.57

Dimension Analysis

C — Compound Cognition: 4

What's crystallized: 37 years of scoring methodology (founded 1956, FICO Score 1989). Proprietary analytic algorithms spanning linear optimization, neural systems, ML, and AI. The Falcon Intelligence Network — a consortium of 10,000+ institutions contributing transaction data — creates a compounding data flywheel that cannot be replicated without rebuilding the consortium. Patent portfolio of 230+ issued patents (204 US + 26 foreign), many AI-specific. FICO 10T identifies 18% more defaulters in the critical score decile vs alternatives and is the only score with validated performance through a complete economic cycle including the Great Recession.

What's re-derivable: VantageScore re-derived a competitive score in ≈3 years, proving c_derive is finite. But independent studies show VantageScore "can only keep pace, in some cases, can't even do that" with FICO Classic — a 20-year-old model — while FICO 10T is meaningfully superior.

Stress test (B₀ challenge): Is the crystallized cognition load-bearing, or does E do all the work? Thought experiment: if FICO's methodology were C=2 but E=5, the company still survives on infrastructure lock-in alone. But the cognition IS load-bearing — it's the reason the ecosystem doesn't WANT to switch, complementing E's reason they CAN'T. The quality gap (18% more defaulters) is the argument against switching. Without it, pressure to adopt free VantageScore intensifies. Continuous estimate: C ≈ 3.7, rounds to 4. Softer than Synopsys (EDA design rules) or SAP (50yr process logic), but deeper than Salesforce (C=3). The Falcon network data is irreplaceable (c_derive = ∞ for that component).

Risk to C: frontier AI could leapfrog bureau-based scoring with alternative data (bank transactions, employment records). Requires Section 1033 (Open Banking), currently gutted under Trump 2.0. Monitor regulatory posture changes.

E — Irreducible Infrastructure: 5

The regulatory mandate chain: FHFA requires FICO for conforming mortgages. LLPA grids calibrated to FICO. MBS rating agencies (S&P, Moody's, Fitch) reference FICO in ratings models. Basel capital calculations use FICO. Auto/credit card ABS documentation calibrated to FICO. 40+ states regulate credit-based insurance scoring referencing FICO. The CEO enumerated 7 downstream nodes where the score propagates without separate payment: originators → GSEs → rating agencies → MBS investors → mortgage insurers → prudential regulators → capital adequacy models. Each node is independently calibrated.

Empirical proof: VantageScore was authorized by FHFA in July 2025. All 3 bureaus offered it free or discounted. After 8 months: <5% adoption. The CEO identified three structural barriers: (a) LLPA grid reconciliation, (b) securitization penalties on VantageScore-scored paper, (c) scores differ by 20+ points 30% of the time, breaking downstream calibration.

Stress test (B₀ challenge): The exclusive mandate is gone — FHFA authorized VantageScore. c_ℓ is not literally ∞. FHFA could publish VantageScore-calibrated LLPA grids and the "impossible" switching cost becomes a policy update. The coordination problem is real but solvable with top-down action. However: the adverse selection argument is devastating. When lenders can choose between two scores, they submit whichever is higher — score shopping breaks the calibration that makes scoring valuable. Bi-metric scoring is self-defeating; the system converges back to one standard, and FICO has the installed base. The <5% adoption after 8 months of free availability is not a slow start — it's an ecosystem immune response. E=5 holds, but the moat is infrastructure coordination cost, not regulatory impossibility. Downgrade trigger: FHFA mandates (not authorizes) dual-score adoption with transition timeline, or VantageScore hits 15%+ share within 24 months.

U — Ecosystem Breadth: 3

6-7 distinct workflows: credit origination, fraud detection (Falcon), customer management (Strategy Director), customer communication, analytics (Workbench), decision rules (Decision Modeler), credit scoring. All concentrated in financial services risk management — one meta-domain. Department coverage limited to risk management (primary), marketing and compliance (secondary). 87-88% of revenue from financial services. CEO: "We are very heavy in financial services, have been historically." 150+ platform customers with 50%+ multi-use-case. Cross-vertical traction (telco, retail, government) is early stage.

A — Distribution & Discoverability: 3

Scores distributed through credit bureaus — FICO is one layer removed from end consumers. An agent seeking credit data talks to a bureau, not FICO. DLP program (new) adds tri-merge resellers representing ≈90% of mortgage volume. FICO Marketplace generally available. Gartner leader in Decision Intelligence Platforms (highest for ability to execute). No evidence of agent-first API design or agent-oriented endpoints. Well-known and has APIs, but agents don't route through FICO by default.

M — Ecosystem Gravity: 5

Maximum gravity. 90% of top US lenders. 75% of top 100 global banks. 600+ insurers (8 of top 10 US P&C). 10,000+ institutions in Falcon network. All 3 US credit bureaus distribute FICO scores (TransUnion passes through $185M/yr in royalties at zero margin). FICO 10T Adopter Program: $377B annual originations, $1.6T eligible servicing.

What switching requires: recalibrate all downstream models (LLPA grids, MBS ratings, securitization docs, capital adequacy) across multiple independent institutions; re-work every bureau, lender system, and GSE pipeline; retrain every loan officer, underwriter, and risk manager; coordinate all participants simultaneously to avoid adverse selection. The counterparty coordination problem creates escape velocity requirements that no competitor has achieved.

F — Ecosystem Friction: 3

Blended score. Scores side (59% revenue): F=2. A 3-digit number (300-850), per-pull via bureau API, trivially consumable. Software side (41% revenue): F=3.5. Multi-year subscriptions, professional services for implementation ($82M declining), legacy migration complexity (dual codebase: platform vs non-platform). Revenue-weighted: 0.59×2 + 0.41×3.5 = 2.6, rounded to 3. Not consultant-dependent (professional services declining, not growing). Platform once deployed is low-friction for expansion (NRR 122%).

Sensitivity Analysis

Adversarial stress test on E and C — the two highest-weighted dimensions:

Verdict is invariant across all four scenarios. κ ranges from 0.10 to 0.57 — conviction compresses but never zeroes.

Regime Context

Factor regression over T = 15 weeks (Dec 9 – Mar 27, 2026):

r_FICO = α + 0.15·r_SPY + 0.64·r_IGV + ε     R² = 0.16

α̂ (ann)     = −95.5%        Extreme negative idio alpha
σ_idio (ann) = 47.8%         High single-stock vol
IR           = −2.00         α̂ / σ_idio
ρ_intra      = 0.40          LOW — discriminate, not indiscriminate
%Idio Var    = 84.0%         Signal IS company-specific

Cumulative: FICO −42.6%, IGV −29.7%, SPY −6.7%. More than half the drawdown is idiosyncratic (−22.6% after removing factor exposure).

This is not an indiscriminate selloff. ρ_intra = 0.40, not → 1. The market is specifically repricing FICO for reasons the V-Score doesn't measure: valuation compression (P/E ≈67× at peak → 37× after −45%), leverage signal ($1B notes at 6.25%, 8-K filed 3/11), and quality-growth-to-value rotation. None of these change V dimensions.

The March catalyst: $1B senior notes offering at 6.25% due 2034 (8-K 2026-03-11, closed 2026-03-20). Use of proceeds: repay revolver, redeem $400M of 5.25% 2026 notes, buybacks. This is refinancing at higher rates for capital return — not operational distress. Market read: management levering into a falling stock for buybacks. Two-day shock (3/10-3/11): −19.0%, virtually all idiosyncratic.

IR does not gate the verdict. IR = −2.0 measures the regime — the market's 15-week repricing of FICO. V = 3.57 measures structural AI survival properties. These are orthogonal. The selloff is about valuation and leverage. The V-Score is about whether intelligence can flow around the company. It can't — the credit ecosystem's calibration standard doesn't change because the stock price fell.

δ — The Structural Discount

δ = V_structural − V_market_implied. V_structural = 3.57 (confirmed by primary source audit across six dimensions). The market is pricing FICO as if something structural broke — 84% of the drawdown is idiosyncratic. V-Score analysis says nothing structural broke. δ is large and positive.

Crucially: this δ emerged from discriminate repricing (ρ = 0.40), not indiscriminate selloff (ρ → 1). The classic V-Score entry signal is "maximum δ at maximum indiscriminate selloff." Here, δ is maximum at maximum discriminate selloff. The edge is different: not "the market is wrong about software" but "the market is wrong about FICO specifically." This requires underwriting the idiosyncratic thesis, not just buying the basket.

Thermodynamic Summary

FICO is the calibration standard for a multi-trillion-dollar credit ecosystem. Intelligence cannot flow around it because every node in the chain — originators, GSEs, rating agencies, MBS investors, insurers, regulators — is independently calibrated to the FICO score. Replacing FICO requires simultaneous re-coordination across all nodes, a coordination problem so severe that even a free alternative backed by all three credit bureaus with full regulatory authorization achieved <5% adoption in 8 months. The moat is not regulatory mandate (which is no longer exclusive) — it is irreducible infrastructure embedding where the score IS the system's shared language. The compound cognition (37 years of cycle-tested calibration, 10,000-institution Falcon consortium) is load-bearing: it provides the quality argument against switching that complements the infrastructure argument for why switching is impossible.

The deep-but-narrow pattern (E=5, M=5, but U=3, A=3) places FICO in a named exception class alongside MSCI: maximum depth in one domain, limited breadth. This caps V below FORTRESS (which requires breadth) but makes the EMBEDDED verdict structurally durable — the domain where FICO operates cannot route around it.

Basket Verdict

V = 3.57   EMBEDDED   κ = 0.57   → KEEP (include, target weight)
w_i ∝ κ_i = 0.57  (normalized across basket by tenant)

IR = −2.0 is regime context, not a gate. The structural properties that determine AI survival are intact. The market is repricing valuation and leverage — orthogonal to the thermodynamic question.

Durable revenue: ≈95% — Scores segment ($1.17B, 59% of total) is fully protected by infrastructure embedding. Platform software ($822M, 41%) protected by switching costs (NRR 112-136%) and cross-module data flows.

Exposed revenue: ≈5% — Professional services ($82M, declining) and non-platform legacy software (NRR 91-97%, actively migrating to platform). Both are small and shrinking by design.

Sources