This article extends our AI patent analysis series. For detailed technical analysis of AIG’s three patents, see the patent deep dives. For competitive implications, see What AIG’s Patents Mean for Carriers.

Executive Summary

The insurance industry’s AI patent landscape is splitting along a fault line that mirrors one of the oldest strategic debates in enterprise technology: build or buy.

On one side, AIG holds three granted U.S. patents covering the core architecture of its AI-assisted underwriting document processing system. These patents protect a proprietary system built by AIG’s own two-person team at the Atlanta Innovation Hub, designed for AIG’s own underwriting pipeline, and deployed against AIG’s own 370,000+ annual E&S submissions. The system names Claude (Anthropic’s LLM) and populates a Palantir Foundry ontology. It is not for sale. AIG’s patents protect a competitive advantage that its rivals cannot license.

On the other side, Quantiphi, an AI-first digital engineering company headquartered in Marlborough, Massachusetts, holds a growing patent portfolio around its Dociphi platform, a generative AI-powered intelligent document management system explicitly designed for sale to insurance carriers, brokers, and third-party administrators. Quantiphi’s patents protect a product, not a proprietary internal capability. Named to the 2025 InsurTech100 list and built on Amazon Web Services infrastructure, Dociphi is positioned as the platform that carriers who do not want to build their own AIG-scale system can license instead.

Both companies are solving the same fundamental problem: how to extract structured, reliable data from the unstructured documents that flow through commercial insurance underwriting. Both are patenting their approaches. But the strategic implications of each patent portfolio are fundamentally different.

AIG’s patents say: we have built something our competitors cannot copy. Quantiphi’s patents say: we have built something your competitors can buy from us. For insurance executives, actuaries, and technology leaders evaluating AI underwriting strategy, understanding both portfolios is essential to making informed build-or-buy decisions.

Quantiphi’s Patent Portfolio: What Dociphi Protects

Quantiphi’s intellectual property strategy centers on Dociphi, its template-free, AI-powered document processing platform. Unlike AIG’s patents, which emerged from a single focused R&D effort, Quantiphi’s portfolio is building across multiple product capabilities. Three patents are central to understanding Dociphi’s architecture.

Patent 1: Loss Run Extraction Without Manual Labeling

Granted in August 2025, Quantiphi’s patent titled "System and Method for Data Extraction and Standardization Using AI-Based Workflow Automation" addresses one of the most persistent pain points in P&C underwriting: the extreme variability of loss run reports.

Loss runs arrive in dozens of formats. Every carrier, every TPA, and every broker system generates them differently. Column headers vary. Date formats change. Some reports are machine-generated PDFs with clean tabular data. Others are scanned images of printouts from legacy systems. Traditional automation approaches required building custom extraction templates for each format, a process that scaled poorly and required extensive manual data labeling to train new models.

Quantiphi’s patent describes a method that eliminates the need for manual data labeling when processing new loss run formats. The system can extract critical data from highly variable, unstructured loss run documents even in cases of training data scarcity. According to Quantiphi’s disclosures, this approach reduces processing and training time by up to 40%.

The actuarial relevance is direct. Loss run data feeds pricing models, reserve estimates, and experience rating calculations. Every actuary who has received a stack of loss runs in inconsistent formats and spent hours reconciling them into a usable dataset understands the problem this patent addresses. The difference is that Quantiphi is solving it as a platform capability available to any carrier, not as a proprietary internal tool.

Patent 2: Three-Way Document Comparison with Human-in-the-Loop

Granted in approximately January 2026, this patent covers what Quantiphi calls Dociphi’s "Comparison Screen." The patent, titled "Validation system and method for concurrent visual validation of two or more electronic documents," describes a system that compares extracted entities across multiple versions of a document, displays all input documents alongside an AI-generated comparison view, allows users to locate exactly where each extracted value appears in its original source document, and resolves discrepancies through a secure human-in-the-loop (HITL) workflow.

This capability is designed for insurance workflows where accuracy and version control directly affect operational outcomes: quote comparison, policy review, claims discrepancy detection, bordereaux processing, and regulatory reporting. The system does not simply extract data. It provides the infrastructure for humans to verify, correct, and approve the extraction before it enters downstream systems.

The HITL emphasis is a deliberate architectural choice that differentiates Dociphi from fully autonomous extraction systems. Where AIG’s patent architecture describes a critic agent that structurally verifies LLM outputs, Quantiphi’s patent places the human reviewer at the center of the validation workflow. Both approaches address the hallucination and accuracy problem, but from philosophically different positions on the automation spectrum.

Patent 3: Enhanced Search Across Large Document Volumes

Filed in June 2023 (Application No. 18/343,923) and receiving a notice of allowance in May 2025, this patent covers a capability within QDox (Quantiphi’s broader document processing platform, from which Dociphi’s insurance-specific capabilities derive). The patent, titled "System and Method for Processing One or More Electronic Documents for Enhanced Search," describes methods for effective data analysis and extraction across large volumes of electronic documents without data corruption or loss.

According to the Justia patent records, the invention involves defining bounding boxes around key-value pairs in schema files, tagging coordinates for keys and values, applying normalization operations, and creating searchable indexes with extractable key-value pairs. This is the foundational document processing layer that enables Dociphi’s extraction capabilities across multiple document types and industries.

The Platform Ecosystem: AWS, Google Cloud, and Unqork

Quantiphi’s strategic positioning is inseparable from its platform partnerships. Dociphi is built on Amazon Web Services, leveraging Amazon Textract and Anthropic Claude (via AWS Bedrock) for its document processing capabilities. QDox is available on the AWS Marketplace. Separately, Quantiphi launched its AI-led Intelligent Underwriting Platform (AiUP) in partnership with Unqork and Google Cloud, combining Dociphi’s document processing with Unqork’s codeless application development platform and Google Cloud’s infrastructure.

This platform strategy means Dociphi is not a standalone product but an embeddable capability. Carriers can integrate it into existing underwriting workbenches, claims systems, or bordereaux processing workflows via API. The platform processes multiple ACORD form types (125, 126, 127, 130, 131, 140), workers’ compensation forms, childcare forms, loss runs, policy documents, and supplemental questionnaires.

Quantiphi’s disclosed case studies describe re-engineering the submission intake process for "one of the largest US commercial insurers" from ingestion, classification, and extraction through to data enrichment with third-party APIs. They separately describe digitizing the loss run underwriting process for "a large commercial P&C insurer" to extract data and convert it into standardized formats. The unnamed client descriptions suggest Dociphi is already deployed in production environments at scale.

The Competitive Positioning: Template-Free and HITL-Centered

Two design principles distinguish Dociphi from both AIG’s proprietary system and many competing vendor platforms. First, Dociphi is explicitly "template-free," meaning it does not require pre-built extraction templates for each document format. The patented loss run extraction method works with variable, unstandardized formats out of the box. This is a direct response to the scalability limitation that plagued earlier generations of document automation: every new document format required custom template development.

Second, Dociphi places human-in-the-loop review at the center of its architecture rather than treating it as an exception path. The patented comparison screen is designed to make human review faster and more accurate, not to eliminate it. Combined with active learning loops that improve extraction models based on human corrections, this approach positions Dociphi as a tool that enhances human reviewers rather than replacing them. For carriers where regulatory or internal governance requirements mandate human oversight of AI-assisted decisions, this is a meaningful architectural distinction.

AIG vs. Quantiphi: A Technical Comparison

The two patent portfolios address overlapping problems through different architectural approaches. Understanding where they converge and diverge reveals the technical trade-offs inherent in the build-vs-buy decision.

Document Ingestion and Extraction

AIG’s approach (Patent #12,437,155) converts documents to text via OCR, uses markdown language patterns to separate tables from narrative text, creates independent table chunks and text chunks with different retrieval strategies, and feeds the appropriate chunks to an LLM (Claude) for extraction through a retrieval-augmented generation (RAG) pipeline. The system populates an ontological data store. The key innovation is the independent processing of tabular vs. textual data through separate pipelines. Full analysis

Quantiphi’s approach (Loss Run patent + QDox patent) uses AI-based workflow automation to extract data from highly variable document formats without manual template creation or data labeling. The system uses bounding box detection, coordinate tagging, and normalization to identify and extract key-value pairs. Rather than the RAG-based chunk-and-retrieve architecture AIG employs, Dociphi’s approach appears to emphasize template-free extraction that can adapt to new document formats with minimal retraining.

The key difference: AIG’s architecture is optimized for a specific, large-scale pipeline (370,000+ E&S submissions flowing through a single system with a defined LLM and ontology). Quantiphi’s architecture is optimized for adaptability across diverse client environments, document types, and formats, because a vendor platform must handle whatever its customers throw at it.

Validation and Error Control

AIG’s approach (Patent #12,437,154) builds traceability into the extraction pipeline itself. Every chunk carries unique identifiers. The LLM is required to self-report which chunks it used for each extraction. The system stores those citations alongside extracted data, enabling structural hallucination detection and automated audit trails. Full analysis

Quantiphi’s approach (Comparison Screen patent) centers validation on the human reviewer. The three-way comparison interface shows extracted entities alongside their contextual locations in source documents, allowing users to verify accuracy visually and resolve discrepancies through a structured HITL workflow. The system learns from corrections through active learning loops.

The key difference: AIG’s validation is architecturally embedded and largely automated, designed for a system processing at massive scale where human review of every extraction is impractical. Quantiphi’s validation is human-centered and interactive, designed for workflows where the human reviewer is a feature rather than a bottleneck. Both achieve accuracy improvement, but AIG’s approach scales more aggressively while Quantiphi’s approach provides more granular human oversight.

Complex Document Processing

AIG’s approach (Patent #12,511,320) uses chain-of-thought prompting to guide the LLM through a four-step reasoning process for multi-table spreadsheets: count tables, identify tables, extract metadata, reconstruct each table independently. It also includes a hybrid text/vision architecture with multi-modal language model fallback for handwritten or poorly formatted documents. Full analysis

Quantiphi’s approach does not have a directly comparable granted patent for multi-table spreadsheet processing via chain-of-thought prompting. However, Dociphi’s platform documentation describes extraction from "key value pairs, tables, checkboxes, forms, handwritten texts" using "cutting-edge AI technologies," and Quantiphi’s case studies reference standardizing loss run data into consistent formats for analytics. The platform handles multiple ACORD form types (125, 126, 127, 130, 131, 140), workers’ compensation forms, and supplemental documents.

The key difference: AIG’s chain-of-thought prompting approach represents a more architecturally specific (and therefore more defensibly patentable) method for handling the multi-table problem. Quantiphi’s platform handles document variety through a different mechanism: template-free extraction models that adapt to new formats rather than a structured prompting methodology for table decomposition.

The Coverage Gap: What Happens Beyond Loss Runs

The most revealing difference between these two patent portfolios is not what they each cover. It is what Quantiphi’s portfolio does not yet cover that AIG’s already does.

Quantiphi’s granted Dociphi patents protect specific capabilities: loss run extraction, multi-document comparison with HITL validation, and general document search. These are important components of the submission intake workflow. But a commercial insurance broker submission is not a single document type. It is a package, and the composition of that package varies dramatically by line of business.

The Full Submission Pipeline Problem

Consider a typical E&S commercial submission, the same transaction type AIG processes 370,000+ times annually. A complete E&S submission typically arrives as a package of five to ten distinct document types, each with its own structure, format variability, and extraction requirements.

The ACORD application (125, 126, 127, 130, 131, 140, depending on the line) is the backbone of the submission. These are semi-structured forms with a mix of checkbox fields, narrative question responses, named insured information, coverage requests, and schedule attachments. While ACORD forms follow a published standard, the way brokers complete them varies enormously: some fill them electronically, some scan handwritten versions, some submit partially completed forms with supplemental attachments that provide the missing information.

Loss runs are the claims history component, arriving in formats that vary by every prior carrier and TPA that generated them. Column headers differ. Date formats change. Some are machine-generated PDFs with clean tabular data. Others are scanned images from legacy systems. This is the one document type Quantiphi’s patent explicitly covers.

Financial statements arrive as multi-page PDFs or spreadsheet exports containing income statements, balance sheets, and ratio analyses. The tabular data is interleaved with footnotes, auditor’s notes, and narrative management discussion. A single financial statement may report figures in dollars, thousands, or millions, with the multiplier specified in a header or footnote that is easily missed by automated extraction.

Supplemental questionnaires are line-specific: a D&O questionnaire asks about corporate governance, litigation history, and executive compensation; an environmental questionnaire asks about site contamination history, remediation status, and regulatory compliance. These documents combine narrative responses with embedded tables and are among the most variable documents in the package because each carrier designs its own.

Broker cover letters and correspondence provide context that does not appear in any structured form: the broker’s market commentary, pricing expectations, incumbent carrier information, account history, and relationship notes. This narrative text often contains critical underwriting intelligence that informs risk selection decisions.

Property schedules and statements of values arrive as spreadsheets listing every insured location with building values, contents values, business income estimates, construction types, occupancy codes, and protection class data. These are complex multi-column spreadsheets where the layout is determined by the broker or insured, not by any standard format.

Where AIG’s Patent Coverage Extends and Quantiphi’s Does Not

AIG’s Patent #12,437,155 explicitly names "applications, broker correspondence, financials, summary of claims, historical claims filed under business insurance policies (‘Loss Run’) and historical claim losses" as document types the system processes. The patent’s claims are not limited to loss runs. They cover the full document pipeline: any document that contains tables and text can be processed through the markdown-based separation, independent chunking, and RAG-based extraction architecture.

AIG’s Patent #12,511,320, the chain-of-thought spreadsheet patent, is directly relevant to property schedules, statements of values, and financial statement processing. These are the exact kind of complex, multi-table spreadsheets that the four-step prompting methodology (count tables, identify tables, extract metadata, reconstruct each independently) was designed to handle. The unit standardization component addresses the financial data variability problem: when one financial statement reports in thousands and a statement of values reports in actual dollars, the system normalizes both to a consistent unit before extraction.

AIG’s Patent #12,437,154, the traceability patent, applies across all document types equally. Every extraction from every document in the submission package gets chunk-level traceability, source attribution, and version tracking. When an underwriter questions a total insured value pulled from a property schedule or a loss ratio derived from a financial statement, the system can trace the extraction back to the specific page and paragraph of the source document.

Quantiphi’s granted patents, by contrast, protect a specific extraction method for loss runs (one document type in the package) and a validation interface for comparing documents. If Dociphi processes ACORD applications, financial statements, property schedules, supplemental questionnaires, and broker correspondence in production, those processing methods are not yet protected by granted patents.

The IP Exposure Question

This coverage gap creates an interesting strategic dynamic. If Quantiphi expands Dociphi’s capabilities to handle the full E&S submission pipeline, including complex spreadsheet processing, multi-format financial data extraction, and end-to-end document orchestration, the methods it develops to do so would need to navigate around AIG’s existing patent claims.

Specifically, if Dociphi were to process property schedules or financial statements by separating tabular content from text using markdown patterns, chunking tables independently, and retrieving them through a vector embedding index, that approach could fall within AIG’s Patent #12,437,155. If Dociphi were to process multi-table spreadsheets like statements of values by using an LLM-driven chain-of-thought methodology to count, identify, and reconstruct individual tables, that approach could fall within AIG’s Patent #12,511,320. If Dociphi were to build automated traceability by requiring the LLM to self-report which source chunks it used for each extraction, that approach could fall within AIG’s Patent #12,437,154.

Quantiphi would need to develop architecturally distinct methods for processing these additional document types, methods that achieve similar outcomes through different technical approaches. This is entirely feasible. As we noted in our carrier implications analysis, there are many paths to AI-assisted document extraction that do not practice AIG’s specific claims. Template-free extraction using bounding box detection and coordinate normalization (Quantiphi’s existing QDox patent approach) is one such path. But the engineering constraint is real: Quantiphi cannot simply replicate AIG’s RAG-based architecture for new document types without evaluating the patent landscape.

For carriers evaluating Dociphi for E&S underwriting, the question is whether the platform’s current capabilities cover the full range of document types in their submission pipeline, or whether gaps exist for specific document types beyond loss runs. A carrier whose primary pain point is loss run processing and document comparison may find Dociphi’s current capabilities sufficient. A carrier seeking end-to-end automation of the entire submission intake workflow, from ACORD application through financial statement extraction through property schedule processing, may find that the full pipeline requires capabilities beyond what Dociphi’s current patent portfolio protects.

The Build-vs-Buy IP Divide

The most significant difference between these two patent portfolios is not technical. It is strategic.

AIG’s patents create a proprietary moat. No carrier can license AIG’s system. No vendor can sell it. The patents protect methods that AIG uses internally to process its own underwriting submissions faster, more accurately, and at greater scale than competitors. When AIG reports processing 100% of its financial lines submissions without adding headcount, or compressing review timelines by 5x, those capabilities are protected by intellectual property that competitors must engineer around. The competitive advantage accrues entirely to AIG.

Quantiphi’s patents create a vendor moat. Any carrier can license Dociphi. The patents protect the platform’s specific capabilities so that competing vendors cannot replicate them. This means the competitive advantage accrues to Quantiphi’s customers collectively, not to any single carrier. A mid-market carrier that deploys Dociphi gets access to patented document processing capabilities without building a dedicated AI team or filing its own patents.

This distinction has profound implications for the industry’s competitive dynamics.

For large carriers: AIG’s approach makes sense when the carrier has the scale to justify building a dedicated AI team, the volume to achieve returns on that investment, and the strategic desire to create proprietary competitive advantages. Carriers like Travelers (which has a partnership with Anthropic and 20,000+ employees interacting with AI), Zurich, and Chubb may follow similar build-and-patent strategies.

For mid-market and specialty carriers: Building an AIG-scale AI system is neither economically viable nor strategically necessary. Licensing a vendor platform like Dociphi, Indico Data, Hyperscience, or EXL’s AI tools provides 80% of the capability at a fraction of the cost and timeline. The trade-off is that the carrier does not own the IP and its competitors can license the same platform.

For brokers and MGAs: These organizations typically lack the engineering resources to build custom document processing systems. Vendor platforms are the primary path to AI-assisted workflow automation. Quantiphi’s case studies specifically reference re-engineering the submission intake process for brokers and automating policy review workflows.

For the industry broadly: A world where carriers are building and patenting proprietary AI underwriting systems while simultaneously licensing vendor platforms that may overlap with those patents creates a complex IP landscape. As we analyzed in What AIG’s Patents Mean for Carriers, any carrier using a vendor platform whose architecture happens to fall within AIG’s patent claims could face questions. Quantiphi’s own patents add another layer: vendors themselves are now building IP moats, which affects carrier switching costs and long-term vendor lock-in dynamics.

What This Means for Actuaries

Data quality improvements are coming from both directions. Whether a carrier builds a proprietary extraction system or licenses a vendor platform, the actuarial benefit is the same: more accurate, more consistent structured data feeding pricing models, reserve analyses, and portfolio analytics. The specific technology stack matters less than the outcome. Actuaries should be evaluating the accuracy, auditability, and consistency of extracted data regardless of whether it came from an in-house or vendor system.

Governance requirements apply equally. ASOP No. 56 and the NAIC Model Bulletin on AI do not distinguish between proprietary and vendor AI systems. An actuary relying on data extracted by Dociphi has the same professional obligation to understand the system’s limitations, validate its outputs, and document its role in the modeling process as an actuary relying on data extracted by AIG’s internal system. The vendor’s patent portfolio does not transfer governance responsibility to the vendor.

The HITL question is becoming a design choice, not a default. AIG’s architecture moves toward automated validation with human review as an exception for flagged items. Quantiphi’s architecture places human review at the center of the workflow. For actuaries designing or overseeing AI-assisted underwriting processes, the appropriate level of human oversight depends on the use case, the risk tolerance, and the regulatory requirements, not on which vendor’s system happens to be deployed.

Vendor lock-in has an IP dimension now. When a carrier licenses a vendor platform with patented capabilities, switching to a competitor or building an in-house replacement becomes more complex than simply migrating data. The vendor’s patents may constrain what the carrier can build internally, and the carrier’s investment in workflow integration with the vendor platform creates operational switching costs on top of the IP constraints.

The Technology Stack Dimension

Beyond the patent claims themselves, the two portfolios reveal different bets on which technology stack will dominate AI-powered insurance operations.

AIG’s stack is vertically integrated with named partners. The patents explicitly reference Claude (Anthropic) as the LLM. AIG’s public disclosures confirm Palantir Foundry as the ontology layer. The system is built around a specific combination of partners with deep enterprise commitments. AIG brought both Anthropic’s CEO Dario Amodei and Palantir’s CEO Alex Karp to its 2025 Investor Day, signaling that these are not vendor relationships but strategic partnerships. The risk is vendor concentration. The advantage is deep integration.

Quantiphi’s stack is cloud-platform-native and multi-partner. Dociphi runs on AWS with Amazon Textract and Claude via Bedrock. AiUP runs on Google Cloud with Unqork. Quantiphi appears to maintain flexibility across cloud providers and LLM backends. For carrier customers, this means less risk of being locked into a single technology stack, but potentially less depth of integration than AIG achieves with its focused partner set.

This distinction matters because the technology stack choices embedded in patents tend to outlast the current generation of tools. AIG’s patents are model-agnostic in their claims (they say "may be a publicly available LLM such as Claude" rather than requiring Claude), which gives them flexibility to swap models while maintaining the same patented architecture. Quantiphi’s product-level flexibility across cloud providers serves the same purpose differently: customers can deploy on whichever cloud infrastructure they already use.

The Emerging IP Landscape

AIG and Quantiphi are not the only players building patent portfolios in insurance AI. The landscape is expanding rapidly.

EXL was recently granted 10 AI-related patents in a single year, including patents covering an insurance-specific fine-tuned LLM and knowledge graph construction for claims processing. As a major services and analytics provider to insurers, EXL sits between the carrier and vendor categories. Its patents protect capabilities that could be deployed across its client base.

Ping An holds over 100 generative AI patents globally, making it one of the most prolific AI patent filers in any industry. However, Ping An’s portfolio is concentrated in the Chinese patent system and covers a broader range of applications beyond underwriting, including healthcare diagnostics and financial fraud detection. The geographic concentration limits its direct relevance to U.S. and European carriers.

Convr, an AI underwriting platform backed by Prudential Financial, has disclosed patent-pending technology for automated commercial underwriting data aggregation and analysis. Its platform ingests submission data and enriches it with external data sources, occupying a similar space to Dociphi’s downstream data enrichment capabilities.

Indico Data, Hyperscience, and other document intelligence vendors have not publicly disclosed insurance-specific patent portfolios, but the competitive pressure from Quantiphi’s patent activity may accelerate their own IP strategies.

The trend line is clear: AI intellectual property is becoming a competitive dimension in insurance. Within five years, the industry may see patent cross-licensing agreements between carriers and vendors, infringement disputes between vendors whose architectures overlap, patent portfolios influencing M&A valuations for insurtechs (a Dociphi acquisition, for example, would include its IP portfolio as a significant asset), and open-source AI tools creating prior art that constrains future patent filings.

For now, the AIG-Quantiphi comparison offers the clearest illustration of how the build-vs-buy divide manifests in intellectual property strategy. AIG is building a moat around its own castle. Quantiphi is building a moat around the castle it sells to everyone else. Both strategies are rational. Both create real competitive advantages for their respective stakeholders. And both will shape how the insurance industry deploys AI for years to come.