Open Architecture Is The $1.1T Key to Insurers’ AI Growth

AI promises to transform how insurers underwrite risk, process claims and serve customers, but the technology can only deliver value when built on open architecture that enables true connectivity across systems. Without this foundation, AI models become expensive tools operating in isolation, unable to access the diverse, real-time data they need to generate accurate insights or automate complex processes.

The difference between AI success and failure in insurance rarely comes down to the sophistication of the algorithms. Instead, it hinges on whether the underlying systems can share data freely between underwriting, claims, finance and distribution functions. McKinsey estimates that generative AI could generate between $50 billion and $70 billion in additional industry revenue, with operational costs potentially falling by up to 40% by 2030¹. Yet these gains remain theoretical for most firms. According to the FCA and Bank of England, 75% of financial services firms already use AI in some form², but actual performance improvements lag far behind the headlines. The bottleneck is architectural, not algorithmic.

Legacy systems and proprietary platforms create data silos that starve AI of context. An underwriting AI trained only on premium and loss data cannot assess risk accurately without access to claims patterns, customer behaviour, external market conditions and broker feedback. A claims automation tool that cannot verify policy terms or cross-reference fraud databases will produce unreliable outputs regardless of model sophistication. When systems cannot communicate, AI operates blind.

Open architecture solves this problem by treating data accessibility as a core design principle rather than an afterthought. API-first platforms, modular system components and standardised data formats allow AI to function as an intelligent layer across the entire business rather than a disconnected application serving a single department. This is not a technical preference. It is the fundamental requirement for AI to work at all.

Open Architecture & AI in Insurance

What Open Architecture Actually Means

Open architecture in insurance describes systems built on API-first design, modular components and standardised data formats that allow different applications to exchange information without manual intervention. Rather than locking data inside proprietary databases, open platforms expose information through documented interfaces that other systems can query in real time.

The Current State of Integration

Despite AI adoption reaching 75% of financial services firms², integration remains the critical weakness. Only 23% of insurers have mature API programmes according to recent industry analysis, whilst legacy core systems continue to operate as isolated silos. Lloyd’s Blueprint Two programme, which was designed to drive market-wide digitalisation, was effectively shelved in early 2026 after years of repeated delays³. The market engagement team was disbanded at the end of 2025, and new leadership at Lloyd’s is now exploring alternative strategies focused on incremental, peer-to-peer technology adoption rather than a single centralised re-platforming effort. The Lloyd’s Market Association reports that 41% of underwriting business leaders do not consider alignment with Blueprint Two important⁴, a sentiment that now looks prescient given the programme’s collapse.

Why Connectivity Determines AI Performance

AI models require structured, contextual data from multiple sources to function effectively. An underwriting algorithm cannot assess risk accurately using only premium data when claims histories, broker submissions, external risk databases and customer interaction records remain trapped in separate systems. Open architecture removes these barriers by allowing AI to access the complete dataset it needs across the entire insurance value chain.

Where Connectivity Unlocks AI Value Across the Insurance Lifecycle

Underwriting: Real-Time Risk Assessment

Open architecture allows AI to pull data from multiple sources simultaneously, transforming underwriting from a sequential process into an integrated analysis. When systems can communicate, underwriting algorithms access broker submissions, claims histories, external risk databases and IoT sensor feeds in real time rather than relying on static snapshots. McKinsey reports that quoting times in selected commercial lines cases have fallen from multiple days to just hours⁵, but only where platforms enable this level of integration.

The impact extends beyond speed. AI trained on comprehensive datasets identifies risk patterns that remain invisible to human underwriters working with fragmented information. However, this capability depends entirely on whether the underlying architecture permits data flow between policy administration, claims management and third-party data providers. Closed systems force underwriters to make decisions based on incomplete pictures, regardless of AI sophistication.

Claims Processing: Automated Validation and Fraud Detection

Claims automation represents one of the clearest use cases for AI in insurance, yet effectiveness varies dramatically based on system connectivity. AI can cross-reference submitted claims against policy terms, validate repair costs against market rates and flag suspicious patterns, but only if it can access the relevant databases without manual data entry or batch processing delays.

Aviva deployed over 80 AI models across its claims operations, cutting liability assessment time for complex cases by 23 days⁶. This improvement required open architecture that allowed algorithms to query policy systems, historical claims data, third-party valuation services and fraud databases simultaneously. The same AI models running on isolated systems would deliver far weaker results because they could not access the context needed for accurate decisions.

The ABI reports that fraud detection has become increasingly dependent on data sharing, with the Joint Fraud Taskforce launched in October 2024 bringing together government, law enforcement and industry bodies specifically to improve information exchange⁷. BIBA uses AI to process large volumes of data at speed for fraud analysis⁸, but this capability requires connectivity between insurers, brokers and external databases that closed platforms cannot support.

Distribution: Broker Portal Integration and Customer Service

Open architecture extends AI value into distribution by enabling intelligent broker portals and customer-facing applications that can access real-time policy information, pricing engines and claims status without duplicate data entry. When distribution systems connect to core platforms through APIs, AI can personalise quotes based on individual risk profiles, automate routine customer queries and provide brokers with instant access to underwriting decisions.

The Lloyd’s market illustrates how fragmented systems undermine these benefits. The collapse of Blueprint Two has left the market without its planned centralised digital infrastructure, forcing individual firms to build their own connectivity solutions³. The LMA now expects a shift towards peer-to-peer technology adoption, with smaller “quick-win” projects taking centre stage⁹. This fragmented approach makes standardised data formats even more critical. The Lloyd’s Market Association continues to prioritise ACORD-compliant data standards and adoption of the Core Data Record across claims, delegated authority and underwriting⁹, recognising that AI tools cannot function effectively without standardised data formats that enable system interoperability regardless of which platforms individual firms choose.

This focus reflects the reality that AI applications depend on consistent data structures. Without standardisation, each system integration requires bespoke development that multiplies costs and limits scalability. The demise of Blueprint Two makes this work more urgent, not less, because the market now depends on multiple vendor-built solutions that must interoperate rather than a single centralised platform.

The Technical Reality: Why Legacy Systems Block AI Effectiveness

The Integration Gap

The FCA and Bank of England report that 75% of firms have already adopted some form of AI¹, yet results remain inconsistent. The critical difference lies not in the sophistication of AI models but in the architecture of the systems they attempt to operate within. AI requires access to diverse, real-time data streams to generate accurate insights. When policy administration platforms, claims systems and broker portals cannot communicate, AI operates on incomplete information regardless of algorithmic capability.

The API Deficit

Only 23% of insurance firms have mature API programmes², creating a fundamental barrier to AI effectiveness. Without application programming interfaces that enable system-to-system communication, data must be manually extracted, reformatted and uploaded between platforms. This process introduces delays, errors and gaps that undermine AI accuracy. McKinsey estimates AI could cut operational costs by up to 40% by 2030³, but only for insurers whose technical infrastructure supports the connectivity AI models require.

The Cost of Fragmentation

Closed systems force organisations to choose between expensive custom integrations or isolated AI tools that cannot access the full data picture. The result is predictable: AI deployments that fail to deliver projected value because the underlying architecture prevents them from functioning as designed. The shelving of Blueprint Two has made this problem more acute for the Lloyd’s market specifically: without a centralised re-platforming programme, each managing agent and broker must now solve the connectivity problem independently or through vendor-led peer-to-peer solutions.

The Business Case: Why Open Architecture Determines AI Return on Investment

The Deployment Cost Differential

Open architecture fundamentally changes the economics of AI implementation. When systems communicate through standardised APIs, insurers can deploy AI models across multiple functions without rebuilding integrations for each use case. Closed platforms require bespoke development for every new AI application, multiplying implementation costs and extending deployment timelines.

McKinsey reports that 92% of firms plan to increase their AI budgets within the next three years¹⁰, but spending alone does not guarantee results. The FCA found that 84% of firms have an individual accountable for their AI approach¹¹, yet accountability matters little if technical infrastructure cannot support effective implementation. The difference between productive AI investment and wasted expenditure depends on whether existing systems can share data with new tools.

Speed to Market and Competitive Advantage

Underwriting speed improvements demonstrate the commercial impact of connectivity. McKinsey found that quoting times have been reduced from several weeks to a matter of days in selected cases, with some commercial lines moving from multiple days to only a few hours¹². These improvements require AI models that can access rating engines, risk databases and policy terms simultaneously. Closed systems cannot deliver equivalent speed because manual data transfer creates bottlenecks that negate algorithmic efficiency.

The growth of MGAs illustrates how technology-enabled business models capture market share. US premium volumes channelled through MGAs have grown at approximately 14% annually, with direct premiums nearly doubling between 2020 and 2024 from $47 billion to $97 billion¹³. MGAs built on modern, open platforms can deploy AI tools faster than traditional carriers constrained by legacy systems, creating competitive pressure that forces broader market modernisation.

The Data Problem: Why AI Can’t Work in Silos

What AI Models Actually Require

AI in insurance depends on access to multiple data sources simultaneously. Effective risk assessment requires claims histories, policy terms, customer interactions, external risk databases and market conditions. Machine learning models improve through exposure to diverse datasets that reveal patterns invisible to human analysis. When systems cannot share data, AI operates on fragments rather than complete pictures.

The Failure Mode of Closed Systems

Legacy platforms and proprietary software create isolated databases that prevent information flow between underwriting, claims and distribution functions. McKinsey research shows that AI and advanced analytics could generate up to $1.1 trillion annually¹⁴, but only when models can access comprehensive data. Closed systems force AI to make predictions based on incomplete inputs, producing unreliable outputs that undermine business confidence in algorithmic decision-making.

Why Data Silos Destroy AI Value

An underwriting AI trained only on submission data cannot incorporate claims experience or loss ratios. A claims processing model without access to policy terms cannot validate coverage accurately. The FCA reports that 75% of firms use AI¹⁵, yet effectiveness varies dramatically based on data accessibility. Systems that prevent cross-functional data sharing condemn AI tools to operate blindly, regardless of model sophistication. The technical architecture determines whether AI enhances decision-making or simply automates existing limitations across disconnected processes.

Where Connectivity Unlocks AI Value Across the Insurance Lifecycle

Underwriting: Real-Time Risk Assessment

Open architecture enables AI to pull data from multiple sources simultaneously, transforming underwriting from a sequential process into real-time risk evaluation. Connected systems allow algorithms to access broker submissions, external risk databases, IoT sensor data and historical claims patterns within seconds. McKinsey found that quoting times have been reduced from several weeks to a matter of days in selected cases, with some commercial lines moving from multiple days to only a few hours¹². This speed improvement is impossible without connectivity that eliminates manual data transfer between systems.

AI models identify pricing patterns and risk correlations that human underwriters miss, but only when they can analyse complete datasets across policy types and geographies. Closed platforms force underwriters to request information sequentially, creating delays that negate algorithmic speed advantages. The 14% annual growth in MGA premium volumes¹³ demonstrates how firms built on connected platforms capture market share through faster, more accurate pricing enabled by AI accessing unified data sources.

Claims Processing: Automated Validation and Fraud Detection

Connected architecture allows AI to cross-reference claims against policy terms, historical patterns and third-party databases instantly. The ABI reports that cyber claims payouts have increased 230% year-on-year to £197 million in 2024¹⁶, creating pressure to process higher volumes without proportional cost increases. AI can validate coverage, detect anomalies and route complex cases to specialists, but only if systems share data across claims intake, policy administration and fraud detection functions.

Aviva deployed more than 80 AI models to improve claims outcomes, cutting liability assessment time for complex cases by 23 days¹⁷. This improvement required connectivity between claims handlers, legal databases and settlement systems. Closed platforms cannot deliver equivalent efficiency because isolated AI tools must wait for manual data extraction, eliminating the speed advantage that justifies implementation costs. The Joint Fraud Taskforce launched in October 2024 engages participation from the NCA, City of London Police, ABI and Lloyd’s¹⁸, reflecting industry recognition that fraud detection requires data sharing across organisational boundaries that closed systems prevent.

The Technical Reality: What Open Architecture Means in Practice

API-First System Design

Open architecture in insurance means building systems around application programming interfaces that expose data and functions to external applications. APIs allow different software components to communicate without manual intervention, creating automated data flows between underwriting platforms, claims systems and distribution channels. The FCA reports that only 23% of financial services firms have mature API programmes¹⁵, explaining why many insurers struggle to implement AI effectively despite substantial investment in machine learning capabilities.

Modular Components Replace Monolithic Platforms

Modern insurance infrastructure separates functions into independent modules that communicate through standardised interfaces. Rating engines, policy administration, billing and claims processing operate as distinct services rather than components locked within proprietary systems. This modularity allows insurers to replace individual functions without rebuilding entire platforms, reducing implementation risk and enabling faster adoption of AI tools designed for specific tasks. The failure of Blueprint Two’s monolithic re-platforming approach has reinforced this point: the LMA now expects the market to adopt smaller, modular technology projects rather than waiting for a single large-scale transformation³.

Data Standards Enable Cross-System Intelligence

The Lloyd’s Market Association prioritised ACORD-compliant data standards and the Core Data Record across claims, delegated authority and underwriting in 2026⁹. Joe Brace, Operations Director for the Lloyd’s Market Association, confirmed that “the LMA will be focused on ensuring the continued rollout and adoption of the CDR across the vendor ecosystem”⁹. Standardised data structures allow AI models to interpret information consistently across different systems, eliminating the translation layers that introduce errors and latency. Without common standards, connected systems still cannot share intelligence effectively because algorithms cannot reconcile incompatible data formats between platforms. With the market now pursuing multiple peer-to-peer technology solutions rather than a single centralised platform, the CDR has become the primary mechanism for ensuring these disparate systems can exchange data in a format AI tools can consume.