V-Score Card

VRSK
V-SCORE:            3.87
VERDICT:            EMBEDDED
κ (conviction):     (3.87 − 3.0)⁺ = 0.87
GATE 1 (E>1):       PASS (E=5)
GATE 2 (A>1∨Σ≥12):  PASS (A=4>1; C+E+U=13≥12)
FAST SCREEN:        3/3 — proprietary data ✓, regulatory mandate ✓, transaction-embedded ✓

DIMENSIONS
  C (w=0.25):  4   Deep domain encoding; litigation-tested coverage language across 50 states [10-K:321]
  E (w=0.22):  5   Statistical agent designation in all 50 states + 38.9B contributory records [10-K:348-349, 364-366]
  U (w=0.18):  4   15 distinct workflows, 7 departments; single-industry (insurance) caps at 4 [10-K:270]
  A (w=0.12):  4   140+ Xactimate integrations, MCP servers in development [Q4:112, 149]
  M (w=0.15):  5   Industry standard, counterparty network effects across carriers/adjusters/contractors [10-K:579]
  F (w=−0.06): 3   Standard enterprise learning curve; active modernization (Reimagine, SaaS) [Q4:73, 154]

CALCULATION:
  Raw = 0.25(4) + 0.22(5) + 0.18(4) + 0.12(4) + 0.15(5) − 0.06(3)
      = 1.00 + 1.10 + 0.72 + 0.48 + 0.75 − 0.18
      = 3.87
  Gates: PASS · PASS = 1
  V = 3.87

REGIME CONTEXT (T=15w):
  IR = α̂ / σ_idio = −1.29   (not significant: t=−0.69, p=0.49)
  δ  = V − V_market = 3.87 − 3.0 = 0.87
  ρ_intra = 0.58 (normal dispersion)

BASKET:             KEEP (κ > 0)
  w_VRSK = κ / Σκ = 0.87 / 9.04 = 9.6% (minimum weight in 8-name basket)

Dimension Analysis

C — Compound Cognition: 4 (corrected from initial 5)

The initial score of 5 conflated data assets with crystallized cognition. After adversarial review, the data was reassigned to E where it belongs.

What remains as pure cognition: (1) standardized coverage language tested in litigation across all 50 states — 250+ experts and lawyers review 17,600 legislative actions, 22,000 regulatory actions, and 2,000 court decisions annually [10-K:331-332]; (2) actuarial models calibrated on proprietary loss experience; (3) inter-module dependency logic linking underwriting → claims → fraud → cat modeling.

What was reassigned to E: 38.9B statistical records, 1.9B claims records, 143M property records, 280M driver records. These are infrastructure, not cognition.

The re-derivation test: actuarial methods are well-known mathematics — re-derivable in months given data access. Cat models already exist at competitors (Moody's RMS, Karen Clark). Coverage language is genuinely irreplaceable (requires decades of litigation outcomes). Quality validation rules (3,000 checks per record) are re-derivable in 6-12 months with ML.

C=5 comps are SAP (cross-industry, 130+ countries) and ServiceNow (cross-department, every enterprise function). VRSK is deep-but-narrow: cross-workflow within a single industry. Framework definition for C=5 requires "cross-domain" — VRSK is cross-workflow within one domain. Score: 4, "deep domain encoding, 1-3 years to re-derive."

E — Irreducible Infrastructure: 5

Three layers of irreducibility. The regulatory mandate is the hardest:

Layer 1 — Statutory designation. P&C insurers are legally required to report statistical data to regulators in every state they operate [10-K:342-344]. VRSK is the licensed statistical agent in all 50 states, DC, Puerto Rico, Guam, and the U.S. Virgin Islands [10-K:348-349, 315-316]. This has been in effect for 50+ years [10-K:346]. No state has ever re-designated. There is no mechanism to "switch" — the regulatory apparatus was built around ISO/Verisk. Government relations teams interact with all 50 state regulators daily [Q4:37]. This alone makes c_ℓ = ∞ for the statistical agent function.

Layer 2 — Contributory data network. 38.9B records accumulated via bilateral contribution agreements with insurers [10-K:364-366]. 3.6B new records contributed annually [10-K:362]. Contributors rarely opt out [10-K:855]. The bootstrapping problem is genuine: no carrier contributes to a system with no value; no system has value without contributions. A well-funded competitor faces a 5-7 year minimum ramp for common perils, and the long-tail data (30-year hurricane return periods) cannot be replicated at all. Network is expanding: 10 new statistical contributors and 106 digital media forensics contributors in 2025 [Q3:49-51].

Layer 3 — Transaction embedding. Xactimate IS the claims settlement workflow — carriers, adjusters, and contractors all use it as the common reference for claim resolution [Q4:185]. ClaimSearch processes 193,000 claims per day [10-K:623]. Forms, rules, and loss costs ARE the regulatory filing mechanism [10-K:310-319]. These aren't adjacent to the transaction — they are the transaction rail.

Scope nuance: the statutory designation directly covers ≈20% of revenue; the contributory model extends protection to ≈50-60%; transaction embedding covers another 20-30%. The layers overlap and reinforce. E = 5 holds, consistent with SPGI (NRSRO) and ICE (exchange infrastructure) comps.

U — Ecosystem Breadth: 4

15 distinct workflows spanning underwriting, claims, risk management, actuarial, compliance, product development, and financial reporting [10-K product sections]. Cross-module data flows are real: property data feeds both underwriting and claims; cat models serve underwriters, reinsurers, and risk managers; loss experience cycles from claims → actuarial → forms/rates → underwriting → claims [10-K:519-520].

Capped at 4 by single-industry concentration. After divesting Marketing Solutions in December 2025 [10-K:528-532], VRSK operates in one segment: Insurance [10-K:270]. U=5 requires 20+ workflows across multiple industries (SAP). VRSK has the workflow count but not the industry breadth.

A — Distribution & Discoverability: 4

CEO explicitly building MCP servers for agentic AI integration [Q4:112]. 35+ AI-powered projects in production [Q4:39]. ExactGen (Feb 2026) is agentic AI that autonomously gathers content from aerial imagery, policyholder photos, and policy documents [Q4:48]. 140+ ecosystem partners on Xactimate [Q4:149], 18 integrations on ClaimSearch [Q4:79].

Not 5 because MCP servers are still in development, not production. Domain-specific — general-purpose agents won't default to VRSK for non-insurance tasks. But within insurance, VRSK's 50-year statutory role means any insurance-focused agent naturally routes through Verisk data.

M — Ecosystem Gravity: 5

$3.07B revenue, 83% subscription ($2.55B durable base), 56.2% EBITDA margin, $1.19B FCF [10-K:2399, 2201; Q4:86, 96]. Migration cost φ decomposes into five components, each independently prohibitive: φ_data (50 years of irreplaceable history), φ_integrations (140+ partners), φ_retraining (tens of thousands of trained adjusters), φ_counterparty (requires carriers + adjusters + contractors + regulators to move simultaneously), φ_regulatory (statistical agent designation cannot be migrated). Contributory agreements remain in effect unless contributor opts out — "it is rare" [10-K:855].

F — Ecosystem Friction: 3

Active modernization: cloud migration completed 2023-2024 [10-K:893-894], Core Lines Reimagine releasing 22 modules in 2025 and 25 planned for 2026 [Q4:73], AI natural language interfaces (SavvyR for regulators, ClaimSearch query tools) [Q3:44], SaaS transition (Verisk Synergy Studio) [Q3:83].

Remaining friction: client training burden is real — "dedicating a lot of time to training" on new modules [Q4:154]. Enterprise-only for core products (no self-serve path). Complex domain means inherent learning curve. No consultant dependency (unlike SAP), clean API ecosystem, AI tools reducing friction. Net: standard enterprise, score 3.

Thermodynamic Summary

VRSK is a regulated utility dressed as a data analytics company. Intelligence cannot flow to local because three barriers prevent it simultaneously: legal mandate (statistical agent designation by law in all 50 states), network physics (contributory data model where 38.9B records cannot be bootstrapped), and transaction embedding (Xactimate IS the settlement rail, not adjacent to it). AI makes VRSK more valuable — proprietary data is the input AI needs, not the output AI replaces. CEO building MCP servers is the correct strategic posture: become the minimum-energy path for insurance intelligence routing.

The C-score correction (5→4) reflects honest accounting: VRSK's cognition is deep-but-narrow (single industry), and the original score double-counted data assets already captured in E. This drops VRSK from FORTRESS to EMBEDDED — structurally durable with mild erosion risk at the margins (personal lines auto softness, advisory substitution).

Regime Context

Factor regression (T=15w, 2025-12-09 to 2026-03-27):

Intra-sector correlation (financial data infrastructure peers):

ρ_intra = 0.58 means this is NOT an indiscriminate selloff. Peers are being discriminated — ICE and MSCI barely down while FICO cratered 43%. VRSK's −16.2% drawdown is 97.4% idiosyncratic: the market found something stock-specific (growth deceleration to 6-7%, personal lines softness, multiple re-rating from ≈35x to 21x forward P/E).

The δ is the story. V measures structural properties. The market applies a uniform "data analytics growth deceleration" discount. But VRSK's structure (E=5, M=5) is not a growth story — it's a regulated utility with 83% subscription revenue, 56% EBITDA margins, and statutory entrenchment. δ = 0.87 says the market undervalues structural durability by treating VRSK like a growth-challenged software company.

IR does NOT gate the verdict. V(s) ⊥ r_sector(t). The structural assessment is regime-invariant. When the idiosyncratic re-rating exhausts itself, κ = 0.87 determines the weight.

Conviction Weight

κ_VRSK = (3.87 − 3.0)⁺ = 0.87

Basket (FORTRESS + EMBEDDED, sorted by κ):
  SPGI  κ=1.24  13.7%     ICE   κ=1.24  13.7%
  SNPS  κ=1.24  13.7%     ADP   κ=1.18  13.1%
  CCC   κ=1.12  12.4%     SAP   κ=1.09  12.1%
  MCO   κ=1.06  11.7%     VRSK  κ=0.87   9.6%

w_VRSK = 0.87 / 9.04 = 9.6%

VRSK receives the minimum weight in the 8-name basket, consistent with EMBEDDED status. The basket tenant normalizes: w_i = W_S · κ_i / Σκ_j. No IR multiplier — IR measures the regime, not the stock. During indiscriminate selloffs (ρ_intra → 1), IR → 0 for everything; κ is orthogonal.

Revenue Durability

Durable (≈90%): Contributory data and statistical agent services (required by law), ClaimSearch anti-fraud (industry-wide database, 193K claims/day), Xactimate subscription base (industry-standard settlement rail), catastrophe and risk solutions (double-digit growth), forms/rules/loss costs (regulatory-mandated), life insurance FAST platform, specialty and international solutions, Medicare compliance (federal mandate).

Exposed (≈10%): Personal lines auto (facing "competitive pressures" and "softness" [Q3:92]), weather-dependent transactional volume (cyclical, not AI-threatened), advisory/consulting services (substitutable by AI agents), standalone analytics where alternative data sources could emerge.

Verdict

V = 3.87 | EMBEDDED | κ = 0.87 | KEEP

Sources: 10-K (2026-02-18), Q4 2025 transcript (2026-02-18), Q3 2025 transcript (2025-10-29), yfinance.