🔒

Access Required

Enter the email address associated with your access grant.

No access? Request here →

Architectural Directive · Insurance AI Platform

Building AI-Ready Insurance Intelligence

A canonical foundation that makes carrier data trusted, machine-readable, governed, and discoverable so that every AI workflow built on top of it works safely and can be explained. This is the primer for what we are building, why each piece belongs where it does, and how clients overlay their own reality without rebuilding from scratch.

Confidential · ElevateNow Architecture Directive
Status: Active Build
Version: 1.0 · May 2026

The Strategic Thesis

The single argument this entire architecture rests on. Every design decision below traces back to this.

Carriers do not have an AI problem. They have a data readiness problem. AI models exist and work. What fails is the layer beneath them: siloed data with no shared vocabulary, unstructured documents AI cannot parse, decisions with no traceable lineage, and knowledge locked in PDFs and tribal memory. The platform we are building closes that gap before the AI runs, not after it fails.
THE GAP TODAY

Data is siloed and semantically inconsistent

Policy admin says "Named Insured." Submission intake says "Insured Name." Claims says "Claimant." Same concept, three systems, zero linkage. AI cannot reason across boundaries it cannot see.

THE CONSEQUENCE

Every AI project rebuilds the same foundation

Each use case — eligibility, FNOL triage, BA check — hand-codes its own entity lookups, confidence logic, and lineage tracking. There is no reuse. There is no governance. There is no carrier portability.

THE RESPONSE

A governed platform that every use case inherits

One insurance ontology. One document intelligence pipeline. One knowledge governance layer. Use cases call the platform. They do not rebuild it. New use cases take weeks, not months.


The Hard Boundary: Platform vs. Use Case

This is the most critical architectural principle. Conflating the two is the single most common failure mode in enterprise AI platform builds.

Non-negotiable rule: Binding Authority is a use case. Underwriting eligibility is a use case. FNOL triage is a use case. The Semantic Layer, Document Intelligence, and Knowledge Studio are the platform. Platform components are domain-agnostic. Use cases call platform services. They do not embed platform logic.
USE CASES
Swappable · Additive · Domain-Specific · Built on top of platform
Binding Authority
BA Steward + Broker Portal
UW Eligibility
Three-gate pipeline
FNOL Triage
Claims Hub + authority
Risk & Compliance
Portfolio + Analytics
Future Use Cases
...
Use cases call platform services. Platform has no knowledge of use cases.
PLATFORM
Domain-Agnostic · Carrier-Portable · Three Pillars
Semantic Layer
Entity graph · Resolution · Standards export
Document Intelligence
OCR · Extraction · Institutional + Transactional
Knowledge Studio
Governance · Curation lifecycle · CI/CD
Platform anchors to starter kit. Starter kit is read-only reference. Never modified.
STARTER KIT
Insurance Ontology · Canonical · Carrier-Neutral · Industry Standard Format
Commercial Lines Submission
291 entities · 275 relationships · 9 source mappings
Claims Ontology
Industry upper ontology · carrier-extension ready
Code Lists + Source Maps
ACORD-aligned enumerations · field-level lineage
Existing systems are preserved as systems of record. Overlay, not replacement.
SOURCES
Carrier Source Systems · All Preserved
Policy Admin
(DuckCreek, Majesco, Guidewire)
Submission Intake
(Bold Penguin, AMS360, Applied)
Claims Systems
(ClaimCenter, Snapsheet, Reserv)
Analytics
(Snowflake, Databricks, GRisk)
Enrichment
(HazardHub, COPE, Verisk)

The Four Dimensions of AI-Ready Data

AI-ready data is trusted, machine-readable data with the context and governance required for AI to work safely and effectively. These four dimensions define the gap between data that exists and data AI can use. Every platform capability maps to one of these dimensions.

🔗

Unified Ontology

The Gap

Data is siloed. Attribute definitions are inconsistent across product lines. No enterprise MDM sync.

The Response

A shared insurance ontology defining key entities and relationships. Knowledge graph physicalizes connections. Semantic interoperability across all domains.

⚙️

Machine-Ready Context

The Gap

Manual processes and unstructured data — emails, PDFs, endorsements — slow decision-making. AI cannot consume what it cannot parse.

The Response

Document Intelligence makes unstructured content searchable and extractable. Semantic layer provides machine-ready context for AI access and consumption.

👁

Governed Lineage

The Gap

Limited ability to trace how data was enriched or used in automated decisions. No audit trail connecting AI output to data source.

The Response

Every enrichment step, every AI recommendation carries full provenance. Lineage from source through transformation to analytical output. Audit-ready by design.

🔍

Discovery & Access

The Gap

No unified catalog. Teams do not know what data exists across the enterprise, where it lives, or how to access it for AI workloads.

The Response

Catalog of catalogs provides enterprise-wide data discovery. Federated query across all source systems. Data becomes findable and consumable for AI workflows.

Mapping to our platform: Unified Ontology = Semantic Layer + Starter Kit. Machine-Ready Context = Document Intelligence. Governed Lineage = Knowledge Studio curation lifecycle + version history + decision tracing. Discovery & Access = Semantic Resolution layer + governed query + Catalog of Catalogs.

The Insurance Ontology Starter Kit

The two reference ontologies — Commercial Lines Submission and Claims — are the canonical foundation every client starts from. They are not built from scratch. They are overlaid. This distinction is what makes onboarding weeks, not months.

Commercial Lines Submission

291 Entities · 275 Relationships · 9 Source Systems

The complete intake-to-issuance entity model. Covers the full commercial lines lifecycle — submission, quote, binder, policy — plus every LOB exposure, location, hazard, coverage, rating, and regulatory entity a carrier needs.

Core Lifecycle Party + Firmographics SIC / NAICS Classification Location + Building GL + WC + Auto + Property Hazard Assessment Journey + STP Data Provenance Rating + Premium State Regulatory
Claims Ontology

Industry Standard Format · Upper Ontology Anchored · Carrier-Extensible

Claims domain entities serialized as machine-readable ontology, anchored to an established industry upper ontology. ElevateNow's canonical base is carrier-neutral. Each carrier extends it in their own namespace without modifying the shared foundation.

  • Claim, Payment, Reserve, Subrogation (financial)
  • Fraud Case, Investigation (fraud domain)
  • Injury, Medical Treatment (injury domain)
  • File Note, Attachment (adjuster file)
  • External Party, Participant, Performer (party roles)
  • Line, Item, Vehicle, Article (claim items)
  • Repair Service (service domain)
How a carrier overlays their reality: The starter kit ships as a neutral foundation covering all commercial lines entities. A carrier adds their system-specific field names, enumerations, and carrier rules on top — never modifying the base. Their configuration stacks above the canonical model, the same way an endorsement layers above a standard policy form. When the foundation improves, every carrier that builds on it benefits automatically.
USE CASE ILLUSTRATION — COMMERCIAL LINES DOMAIN

Binding Authority is a use case within the commercial lines domain. Its rule types — geographic restrictions, hazard score gates, rating rules, coverage gates, class appetite — each correspond to entity groups that already exist in the commercial lines starter kit. The table below is not a separate layer. It is a concrete example of a use case drawing directly from what the ontology already provides.

BA Rule Type (Use Case) Draws From Starter Kit Entity Commercial Lines Entity Group How the Knowledge Graph Connects It
Geographic RestrictionState Requirement + Wildfire / Flood HazardRegulatory + HazardGoverns geographic eligibility by state and hazard exposure level
Hazard Score GateCrime Hazard / Wildfire Hazard / Flood HazardHazard AssessmentApplies score thresholds to hazard severity classifications
Rating RuleMinimum Premium / Rating FactorRating + PremiumGoverns minimum premium floors and rating factor requirements
Coverage GateBuilding Construction + Wind Hazard + Endorsement FormLocation + Hazard + CoverageControls multi-variable coverage eligibility requirements
Class AppetiteIndustry Classification + Line of BusinessSIC / NAICS + LOBClassifies by industry code and governs which lines of business are eligible

The Three Platform Pillars

Each pillar has a distinct owner, a distinct concern, and a defined interface to the others. They do not bleed into each other.

PILLAR 1

Semantic Layer

The queryable knowledge graph. Domain-agnostic. It resolves entity context on request without knowing whether it is serving a BA check or a UW submission.

  • Entity registry (81 to 291+ entities)
  • Structural relationship graph
  • Knowledge anchoring to canonical entities
  • Semantic resolution function
  • Standards-compliant export for analytics delivery
  • Semantic query layer
INPUT
Industry class, state, lines of business, carrier
OUTPUT
Assembled knowledge context + confidence trace
PILLAR 2

Document Intelligence

Two modes, one pipeline. Makes unstructured content consumable. Domain-agnostic by design.

INSTITUTIONAL MODE

Bulletins, appetite guides, authority matrices extracted into governed rules through the Knowledge Studio curation lifecycle and into active knowledge in the Semantic Layer

TRANSACTIONAL MODE

FNOL documents, loss runs, submission packages, ACORD forms extracted into structured data scoped to individual cases and kept separate from the governed rule base

The two modes share extraction infrastructure but are governed separately. Institutional content enters the knowledge lifecycle. Transactional content stays case-scoped and ephemeral.
PILLAR 3

Knowledge Studio

Governance and authoring layer. Infrastructure product. Not a business user tool. Maintains the knowledge that use cases query.

  • Entity registry management
  • Rule curation lifecycle (draft through sanctioned)
  • Institutional document pipeline execution
  • Standards-compliant export for carrier integration
  • Ontology consistency validation
  • Immutable audit trail for all knowledge changes
Principle: Every business role — underwriter, claims director, BA steward — lives in the Workbench. Knowledge Studio is permanently the knowledge engineering backend.

How Knowledge Studio Layers In

Knowledge Studio is the bridge between the static starter kit and the live, queryable knowledge graph. It has a precise role at each step of the carrier journey.

STEP 1
Starter Kit Delivered

Insurance ontology imported into Knowledge Studio. 291 entities loaded into the registry. Structural relationships derived. Source system field mappings registered.

STEP 2
Carrier Overlays Their Model

Carrier system field names, enumerations, and carrier-specific extensions added via Knowledge Studio. Carrier layer stacks above the canonical foundation without modifying it.

STEP 3
Institutional Knowledge Authored

Bulletins, appetite guides, authority matrices flow through the DocIntel pipeline. Extracted rules are anchored to canonical entities. Curation lifecycle: draft, steward review, sanction, active.

STEP 4
CI/CD Export

Knowledge Studio serializes the live entity graph and active rules as industry-standard JSON-LD. Validated and deployed to the carrier's analytics environment in approximately 5 minutes.

STEP 5
Use Cases Query Platform

BA check, UW eligibility, FNOL triage all call the same platform resolution layer. It traverses the live entity graph and active rules. No use case rebuilds entity logic independently.

WHAT KNOWLEDGE STUDIO OWNS
  • Entity registry (add, modify, version entities)
  • Rule curation lifecycle (all status transitions)
  • Institutional document pipeline execution
  • Standards-compliant CI/CD export for carrier integration
  • Ontology consistency validation
  • Immutable audit trail (all knowledge changes)
  • Ontology visualization and coverage reporting
WHAT KNOWLEDGE STUDIO DOES NOT OWN
  • BA eligibility gate logic
  • UW three-gate pipeline decisions
  • FNOL triage routing rules
  • End-user workflows (those live in the Workbench)
  • Transactional document case management
  • Premium calculation or rating engine logic
  • Workflow orchestration

The Design Principles That Make This Work

These are architectural commitments, not preferences. Holding these boundaries is what keeps the platform flexible as use cases grow and carriers change.

1

The platform serves every use case without being shaped by any one of them

The intelligence platform answers the same question for any workflow that calls it: what do we know about this risk, in this geography, for these lines of business? It works identically whether the caller is a BA check, a UW submission, or a future use case we have not built yet. Domain logic belongs in the use case, not in the shared foundation.

NEVER: Platform logic tailored to a specific use case OK: Use case applies its own decision logic on top of platform context
2

Use cases draw from one governed source. They do not rebuild it.

When a BA check needs to know about hazard exposure, geographic restrictions, or class eligibility, it draws from the same governed knowledge base every other workflow uses. There is no separate copy maintained for BA, a separate copy for UW, a separate copy for Claims. One source. Every decision traceable to it.

NEVER: Duplicate knowledge into use-case-specific stores OK: All use cases query the same governed knowledge layer
3

The starter kit is a shared foundation, not a template to rebuild from scratch

Carriers extend the starter kit by adding their system field names, enumerations, and carrier-specific rules on top. They do not edit the base. This is what makes the platform portable across carriers: when a rule changes at the foundation level, every carrier configuration that builds on it benefits automatically.

NEVER: Edit base ontology entities for a carrier-specific need OK: Carrier adds their configuration layer on top; foundation stays intact and shared
4

Knowledge governance is a backstage function, not a business workflow

Underwriters, claims directors, and BA stewards do their daily work in the business workbench. The knowledge engineering layer — where rules are authored, reviewed, and sanctioned — operates behind that interface. Business users never interact with the governance infrastructure directly. This separation is what keeps the audit trail clean and reliable.

NEVER: Expose the governance layer as a business user workflow tool
5

Institutional rules and case-specific data are kept separate by design

Binding authority bulletins, appetite guides, and authority matrices define how decisions are made. They are institutional, durable, governed. Individual submissions, FNOL reports, and claim documents describe a specific case. They are transactional, ephemeral, case-scoped. Keeping them separate ensures governed rules remain clean and every decision remains auditable.

NEVER: Mix institutional rule content with case-specific document data OK: Each stream has its own pipeline, storage, and lifecycle
6

AI accelerates rule authoring. Runtime decisions stay deterministic and traceable.

AI is valuable when a new BA bulletin arrives: it accelerates extraction and proposes structured rule entries for steward review. But when a BA check runs at decision time, every gate is evaluated against rules that have been reviewed, approved, and sanctioned by a human steward. This distinction is what makes decisions explainable and defensible to regulators and auditors.

NEVER: Generate a runtime authority or eligibility verdict without a traceable, governed rule OK: AI assists authoring; humans sanction; governed lookup executes at decision time

Technology Alignment

Each component maps to a platform layer. The approach is additive and technology-agnostic at the carrier layer. We deliver standards-based artifacts that feed into whatever analytics and AI infrastructure the carrier operates.

Component Layer What It Does Integration Path
Operational Knowledge Store
Knowledge graph database
Semantic Layer Houses the entity registry, relationship graph, governed rules, and reference data. The queryable foundation every use case draws from. Schema evolves incrementally as the ontology grows. Current. No change required. Adaptable to carrier's preferred database infrastructure.
OWL Ontology Library
Python-native, open source
Semantic Layer Serializes the entity registry as industry-standard JSON-LD for delivery to carrier analytics environments. Enables logical reasoning over the knowledge graph for derived decision gates without LLM dependency at runtime. Added to platform dependencies. Powers standards-compliant export and ontology validation.
Ontology Authoring Tool
Open source, desktop and web
Knowledge Studio GUI used by knowledge engineers during carrier onboarding to review and extend the entity model. Supports direct import from Excel entity registries, accelerating initial setup for carriers with existing data dictionaries. Knowledge engineering workflow only. Not a runtime dependency.
OWL Standards Layer
Industry-standard ontology interface
Knowledge Studio The JSON-LD we produce is compatible with industry-standard OWL tooling. Carriers that already have ontology infrastructure in their analytics environment can consume ElevateNow output without changes to their existing pipelines. Compatibility layer. No carrier tooling changes required.
Knowledge Studio Backend
Python · REST API
Knowledge Studio All curation lifecycle management, entity registry, document intelligence pipeline, and standards-compliant export. The production governance backend that powers the entire knowledge layer. Current. Continuous improvement as platform matures.
Carrier Analytics Platform
Snowflake, Databricks, or equivalent
Carrier System AI and analytics consumption layer receiving the JSON-LD knowledge manifests from ElevateNow. Carriers query semantic views rather than raw tables, accelerating AI and BI workload development on governed data. ElevateNow delivers standards-based JSON-LD. Carrier's own CI/CD handles deployment. No ElevateNow code lives in the carrier's analytics environment.
Workflow Orchestration
Use case workflow engine
Use Case Layer Orchestrates use case workflows — BA bulletin ingestion, submission intake, FNOL processing. Calls platform services as discrete, auditable workflow steps. Each use case defines its own workflow independently. Current. New use cases add their own workflows. Platform layer does not change.

AI Readiness Across Insurance Domains

Each domain requires specific structured and unstructured data to be AI-ready before automation is achievable. The platform — built once — serves all four domains simultaneously. A carrier that implements Binding Authority gets Underwriting, Claims, and Risk for the marginal cost of the domain extension.

Binding Authority
STRUCTURED MUST BE AI-READY
  • Appetite parameters and class restrictions
  • Geographic eligibility rules by state and hazard
  • Authority thresholds by class and amount
  • Rating rules and minimum premium requirements
UNSTRUCTURED MUST BE AI-READY
  • BA bulletins and carrier circulars
  • Underwriting guidelines and appetite guides
  • Endorsement forms and authority letters
WORKFLOW STAGES

Bulletin Ingestion, Rule Extraction, Authority Assignment, Submission Check, Audit Trail

Underwriting Value Chain
STRUCTURED MUST BE AI-READY
  • Submission, policy, quote records
  • Customer and location attributes
  • Risk scores and classifications
UNSTRUCTURED MUST BE AI-READY
  • Broker emails and cover notes
  • Loss run PDFs and SOVs
  • Prior endorsements and UW notes
WORKFLOW STAGES

Account Creation, Data Enrichment, Technical Rating, UW Review, Policy Issuance

Claims Lifecycle
STRUCTURED MUST BE AI-READY
  • Claim events and FNOL data
  • Reserve amounts and payment history
  • Litigation status and regulatory flags
UNSTRUCTURED MUST BE AI-READY
  • Adjuster file notes and correspondence
  • Demand packages and medical records
  • Police reports and legal documents
WORKFLOW STAGES

FNOL Intake, Assignment, Evaluation, Litigation, Continuous Feedback

Risk & Compliance
STRUCTURED MUST BE AI-READY
  • Enterprise exposure positions by line
  • Treaty and facultative structures
  • Appetite parameters and thresholds
UNSTRUCTURED MUST BE AI-READY
  • Treaty wordings and slip documents
  • Manuscript endorsements
  • Emerging risk bulletins
WORKFLOW STAGES

Exposure Aggregation, Clash & Accumulation, Compliance, Scenario Testing, Emerging Risk Signals

Why binding authority is more than a rule engine: A BA solution built on this platform is not a lookup against a spreadsheet of rules. It is a governed intelligence layer where every BA decision traces back to the specific bulletin that authorized it, runs against a knowledge graph connecting hazard, geography, class, and rating entities, and produces an audit trail a carrier can defend to regulators. That foundation — once built for BA — extends to every other AI workflow at near-zero marginal cost.

Build Roadmap

Three milestones. Each delivers tangible business capability. Organized by the four dimensions of AI-ready data.

MILESTONE 1: FOUNDATION
Platform Established

Unified Ontology: Full starter kit entity model loaded and validated. Structural relationship graph complete. Carrier source system mappings registered.

Machine-Ready Context: Submission documents and BA bulletins extractable via Document Intelligence. ACORD form extraction feeding intake workflows.

Governed Lineage: Every extraction step traceable to source document and field. BA rules linked to the specific bulletins that authorized them.

Discovery: Semantic resolution layer live with full BA domain awareness. Platform and use case logic cleanly separated.

MILESTONE 2: OPERATIONAL
Use Cases Running on Platform

Unified Ontology: Claims domain entities added. Cross-domain entity resolution live — submission to claim to loss history.

Machine-Ready Context: Loss runs, endorsements, BA bulletins all searchable. Document Intelligence feeding UW, Claims, and BA workflows.

Governed Lineage: End-to-end provenance captured. Audit trail connects automated decisions to the source data and the specific rules applied.

Discovery: Standards-compliant CI/CD export operational. Carrier's analytics environment updated on each Knowledge Studio publish.

MILESTONE 3: INTELLIGENCE
Platform-Driven Differentiation

Unified Ontology: Enterprise ontology covering all priority lines. Derived decision gates replace hand-coded logic. New lines of business onboard in days, not months.

Machine-Ready Context: Full Document Intelligence across UW and Claims. Unstructured content is a queryable asset, not a manual bottleneck.

Governed Lineage: Any regulatory question about how a decision was made is answerable directly from the system. Rule changes propagate automatically to all affected workflows.

Discovery: Impact preview: changing a rule shows which submissions are affected before the change goes live. Full semantic query across the enterprise knowledge graph.


What We Build First

Four deliverables that establish the governed foundation and activate the binding authority use case. Every item is additive. Nothing in your current environment is disrupted.

FIRST DELIVERABLE

A complete, governed rule base for binding authority

We ingest your binding authority bulletins, appetite guides, and authority matrices through the document intelligence pipeline. Every rule is extracted, linked to its source bulletin, reviewed by a human steward, and entered into the governed knowledge base. From day one, every BA decision is traceable to the specific document that authorized it.

✓ Every BA decision is explainable, auditable, and defensible

SECOND DELIVERABLE

Your carrier rules layered above a neutral foundation

We deliver the full commercial lines ontology and overlay your carrier's specific rules, geographic parameters, class eligibility logic, and rating requirements on top. Your rules extend the foundation — they do not overwrite it. When the foundation improves, your configuration benefits automatically. No proprietary lock-in.

✓ Your BA rules are portable assets, not embedded code

THIRD DELIVERABLE

One source of truth that every workflow draws from

When a BA bulletin changes a geographic restriction or a hazard threshold, the update enters the governed rule base once. The BA check, the UW eligibility pipeline, and every future workflow calling the platform inherit the update automatically. No manual synchronization across systems. No drift between teams.

✓ Rule changes propagate automatically — one update, consistent everywhere

FOURTH DELIVERABLE

A knowledge foundation you own, in a format you can use

The governed knowledge foundation we build with you is serialized in industry-standard formats and delivered to feed your existing analytics infrastructure — whether Snowflake, Databricks, or your carrier's current AI and BI platform. You retain the asset. No proprietary format. No ongoing dependency on ElevateNow for data access.

✓ Standards-based output consumable by any downstream AI or analytics stack

The question insurance data leaders are asking is not "Can we use AI for binding authority?" It is: "How do we ensure that AI-assisted BA decisions are explainable, consistent with our rules, and defensible to regulators?" The answer is not a better AI model. It is a governed knowledge foundation that every decision traces back to. That is what we build first — and it becomes the foundation for every AI workflow that follows.