Key Takeaways
Agentic AI value is still stuck in window-dressing pilots. Point solutions outside the core create integration friction and limited scalability, especially where controls and auditability are required.
The automation gap is biggest in core processes. Core processes consume ~52% of total FTEs, with claims proving the least automated domain — over 90% of claims events are primarily processed manually.
Anomalies drive cost. High-variance cases, especially in claims, resist deterministic automation — resulting in leakage and taking up more expert time.
Agentic AI can close the execution gap at scale. With 20% of annual operating costs linked to FTEs, the addressable prize pool for agentic AI is estimated at ~€4.5 billion.
Core system vendors are making Agentic AI easier to implement. Embedded capabilities with built-in governance can successfully materialize savings.
Core strategy is now an AI strategy. Deciding whether intelligence lives inside or outside the core is crucial.
Introduction
Our recent studies1, 2 indicate that most insurers face three main challenges in the lead-up to 2030 and beyond:
- Profitability and margin pressure
- Changing customer expectations and demographic shifts
- Higher risk volatility and claims inflation.
At the same time, Agentic AI3 is expected to reshape core insurance processes4, but it has become increasingly difficult to simultaneously balance the needs of insurers, customers, and regulators.
Insurers operate in an increasingly tight market where margin pressure forces them to improve efficiency. In an attempt to meet customer expectations and stay competitive, many are taking steps to implement Agentic AI and significantly reduce operating cost and cycle times.
Customers increasingly expect rapid, transparent, convenient journeys (e.g., self-service, immediate responses, real-time status updates), but will only accept AI-enabled processes if outcomes feel consistent and fair, and if they believe data is being handled securely and responsibly, as highlighted in a recent BCG article.
Regulators require robust governance, auditability, and clear controls over how information is processed - and how automated decisions are designed. This constrains and slows down the implementation of AI capabilities and their integration in core processes, as outlined in another BCG article.
Placeholder figure 1
In all our client work, we see insurers experimenting with bespoke Agentic AI tools - often as pilots that deliver local wins but are far from scaled impact. Many still run on dated, highly customized core systems and face the modernization dilemma, outlined in this BCG article.
A key strategic question emerges: Modernize the bespoke core with Agentic AI–enabled tooling - or leverage a vendor core to accelerate the agentic journey through embedded, governable capabilities?
The automation gap in core processes
Boosting operations efficiency is a core field of action for many insurers, with the improvement of operating cost-to-serve being a top priority. But our BCG Insurance Excellence Benchmark (IEB) still shows a clear productivity gap driven by manual, personnel-intense work across underwriting, claims handling, and customer service.
The question is, how does Agentic AI help insurers reduce cost-to-serve, address topline potential, and absorb demographic-driven capacity constraints simultaneously?
To interpret the automation gap, we translate it into a cost-base view (Exhibit 1) using 2024 earned gross premiums (Top 10 players across P&C/Life/Health)1:
- Operating cost ratio: ~24% (P&C) vs ~10% (Life) vs ~8% (Health)
- Scale effects: only marginal in Life; largely flat in P&C and Health (i.e., no materially higher operating costs despite lower earned gross premiums)
Assuming 20% of operating costs are allocated to FTEs, the automation-addressable pool is ~€4.5bn per annum (P&C ~€3bn; Life ~€1bn; Health ~€0.5bn).
Having established this estimate, understanding where the most labor-intensive work is taking place in core operations is the next step – pinpointing where automation will have the greatest impact.
(Placeholder figure 2)
Our IEB results, covering major players in Germany and retail insurance, illustrate today’s FTE allocation (see Exhibit 2):
- On average, ~52% of total FTEs are allocated to core insurance processes (44%) and line-specific execution activities (8%), indicating that productivity gains in core operations have a direct, material impact on the operating costs (and profitability as a result)
- In terms of core insurance processes, applications (14%), contract and policy management (32%), and operative claims handling (41%) make up the largest FTE buckets
(Placeholder figure 3)
The IEB results also include degrees of automation for the above FTE allocation. Based on this, we have focused on select P&C lines (personal and motor) and Life insurance to see where automation already works - and where manual effort still dominates (see Exhibit 3):
- Across benchmark participants, straight-through processing remains limited: for Underwriting/Applications and Policy Administration, the average is just above 50% in personal lines, and just over one third in motor lines
- Claims handling is the least automated domain: >90% of claim events are still primarily being processed manually across insurance types, validating the high staffing intensity mentioned earlier (and highlighting substantial efficiency potential)
(Placeholder figure 4)
Not only can Agentic IT reduce FTE-driven operating costs by automating exceptions, it can also improve core insurance processes and customer outcomes. For example, in claims, Agentic AI can dynamically analyze documentation and detect potential fraud, and in operations and servicing, it can resolve ~80% of interactions (providing proactive nudges and next-best action recommendations), as highlighted in a recent BCG article.
Agentic AI uses probabilistic reasoning to handle factors like context and ambiguity, dynamically adapting business rules as policies evolve - unlike deterministic, rule-based automation. Using core-system vendor building blocks, insurers can deploy it safely at scale as a “digital regulated employee.”
Deep dive
BCG Insurance Excellence Benchmark
The BCG Insurance Excellence Benchmark is an annual study in which BCG collects and analyzes operational and performance metrics from major insurance players. The benchmark has been conducted for multiple years, providing consistent year-over-year comparisons and a fact base on how the industry evolves across efficiency, digitalization, and customer service outcomes.
Data collection focuses on core insurance and operational areas, including FTE deployment and productivity indicators. It also captures process performance metrics like cycle times, and digitalization measures such as degree of automation and straight-through processing across key processes.
To ensure comparability, participating insurers provide a standardized set of business and volume indicators, including premiums, insured persons, and insured risks. This ensures that operational metrics can be interpreted in context. The benchmark is complemented by thematic deep dives on current topics so as to validate findings, explain drivers of performance differences, and identify practical improvement levers.
The data is aggregated and analyzed to derive industry averages and reference ranges that represent the current market baseline. Participating insurers receive a tailored view of their own results and relative positioning versus peer groups, as well as the overall benchmark perspectives and observed trends. This enables insurers to quantify their performance in operational efficiency, digital readiness, and customer-centricity, while also identifying structural gaps.
Ultimately, it equips them to prioritize targeted initiatives based on a comparable, market-tested fact base.
If you are interested and want to know more, please reach out to us here:
<contact email>
Architectural building blocks: Closing the automation gap with Agentic AI
While it is deceptively easy to launch Agentic AI pilots, it is extremely difficult to deliver value in production.
True automation gains require insurers to move beyond standalone pilots, embedding Agentic AI strategically where it can reliably change the way work gets done. The main goal is to use the technology to reduce manual effort and shorten cycle times; please see this BCG article for reference.
Based on our project experience, three architectural layers are involved in achieving value in production:
- The Agentic AI Layer - agents add probabilistic reasoning to interpret context, handle ambiguity, and orchestrate tools and workflows beyond deterministic rules (especially in high-variance, exception-heavy work)
- The Data Layer - constraints typically do not affect LLM-models, but they do impact the data foundation: fragmented repositories, incomplete document capture (e.g., OCR), missing metadata, and inconsistent master data can limit reliable automation
- The Infrastructure & Cloud Layer - legacy integration patterns, limited API exposure, brittle interfaces, and long release cycles constrain secure, scalable, and operable connections between agents and core workflows
How vendors tackle these challenges in their commercial core systems with Agentic AI
Vendor core systems increasingly offer standardized, embedded AI out of the box - helping insurers scale adoption with governance and compliance by design.
To provide current insights into the architectural building blocks, we invited selected vendors to share perspectives. The sections that follow reflect the vendors’ own inputs and provide a neutral market view - not a comparison, evaluation, or recommendation provided by BCG.
Deep dive
Vendor assessment methodology
Methodology
We invited selected insurance core system vendors to share their perspectives via a structured question template and interview format. Inputs combined written documentation and live sessions to understand how vendors embed Agentic AI into commercial core systems — addressing key constraints like governance, compliance, and legacy integration.
The question set covered vendor and platform profiles, current Agentic AI usage in core processes, differentiators and production-ready use cases, required architectural building blocks (e.g., smart business layer, Agentic AI layer, data layer, core transaction layer, cloud), and strategic outlooks.
Participants
MSG
ADESSO
FAKTOR ZEHN
PEAK3
Role of BCG
Participating vendors completed the template and contributed interview inputs; BCG performed a consistency check to support comparability across responses and alignment with BCG editorial standards. BCG did not review or evaluate provided vendor statements. Vendor-specific content can be accessed via the vendor's website.
Disclaimer
This list of participants is not comprehensive and does not represent a ranking or endorsement. The objective is to provide a neutral, broad market perspective on approaches and offerings; it is not intended to recommend or promote any product.
(vendor input)
Select a vendor to explore their responses. All inputs reflect the vendor's own perspective — not a BCG evaluation.
msg
adesso
Faktor Zehn
peak3
Q1 — Vendor and platform profile
📎 Slides to be embedded — placeholder for msg platform profile slides
msg's core insurance platforms are the msg.Insurance Suite and the msg.P&C Factory. AI capabilities are deeply embedded in these core systems and support key value-creation processes across the insurance value chain. Focus areas include sales, Pricing & Underwriting, claims management, and policy administration.
The Suite already provides functions such as policy information via chatbot or email bot, automated processing of change requests, end-to-end handling of broker submissions (commercial lines), extraction and classification of documents, recognition of damage patterns using ML, and "Next Best Actions" for claims handlers. The msg.AI Platform / msg.Insurance Data Platform powered by Databricks serves as the centralized data backbone for scalable, governed AI use.
Q2 — Agentic AI in insurance
msg provides Agentic AI and AI-supported capabilities across the insurance core, especially in sales, Pricing & Underwriting, claims management, and policy administration.
- Sales & Underwriting — AI completes application data, plausibilizes inputs, and derives tariff attributes from unstructured data
- Claims — documents are extracted and classified automatically, ML models detect damage patterns, cumulative losses are identified early, and "Next Best Actions" support handlers. AI-supported workflows are a core part of the msg.P&C Factory today (AI in a Process), while AI agents can also be used as individual sub-processes (AI as a Process)
- Policy administration — change requests are processed automatically and broker submissions can be handled end-to-end
- Customer service — chatbots and email bots provide policy information and answer standard inquiries
Productive AI agents in customer implementations include claim:it, process:it, and ask:it.
Q3 — Differentiating Agentic AI characteristics
Production-ready use cases in active client deployments include: policy information via chatbot or email bot; AI-supported assistance for policy inquiries; automated processing of change requests; end-to-end handling of broker submissions (commercial lines); extraction and classification of documents; recognition of damage patterns using ML; early detection of cumulative losses; "Next Best Actions" for claims handlers; identification of recovery potential and consequential damages; automatic reserve setting; completion of application data; automatic plausibility checks; derivation of tariff attributes from unstructured data; and straight-through processing via ML rulesets.
Productive AI agents already delivered in projects: claim:it, process:it, and ask:it.
Key differentiators: deeply embedded AI in core systems; ready-to-use productive agents; the msg.AI Platform as a centralized data lake and foundation for scalable, governed AI use; and a platform approach with shared governance, reusable components, and a future architecture with communicating agents.
Q4 — Architectural building blocks
Smart Business Layer — msg offers AI-based customer chat through its Process:it product, allowing customers to ask about contracts, make changes, and receive optimization suggestions. The same concept is planned for internal staff. In claims, the Claim:it agent checks information against product rules and LLM-based reasoning and can process a claim almost fully automatically (with human review before payout). Image recognition includes forensics to flag potentially AI-generated fraud.
Agentic AI Layer — msg develops its insurance agents in-house as generic tools do not fit its software, releases, and domain knowledge closely enough. The company is LLM-agnostic, using its own prompt layer and working with whichever models the customer runs (especially AWS or Azure). Camunda currently handles orchestration; Databricks is used on the data side. The roadmap includes a multi-model platform with stronger governance and agent-to-agent communication through MCP.
Data Layer — msg streams suite data into a real-time data service and combines it with other source systems, important for customers connecting legacy and new core environments. The platform feeds downstream reporting, warehouses, and AI agents. Databricks is a strategic component, and msg's data dictionaries help map structures quickly to support valid analytics and production use.
Core Transaction Layer — msg sees a selective API layer around the suite as essential for true end-to-end use cases, exposing only the business services that agents need. AI creates value only when linked to workflows, rules, and follow-up processes — not as a stand-alone recognizer. msg is pushing tighter integration between API layer, AI layer, and process layer.
Infra and Cloud Layer — Deliberately flexible cloud strategy. msg runs its SaaS mainly on AWS but also supports Azure and other common cloud setups via containerized deployment. Stack:it Cloud is also a viable option.
The msg.Insurance Suite already provides core AI functions as part of the implementation roadmap. Build-out elements include: AI-Workbench, AI Marketplace, Knowledge Management, Context-Engineering, domain-specific / universal / task agents, and a future architecture with communicating agents. In software engineering, the roadmap includes: DoR agent (productive); AI in coding (in use); requirements assistant (under expansion); test generation (under expansion); and quality gate vision.
Strategically, msg is moving from embedded single AI capabilities and productive agents toward a holistic, governed platform approach — explicitly positioning the platform solution as the target for reusable components, centralized governance, and scalable operation. The msg AI Platform provides a unified, scalable architecture for orchestrating AI agents, enforcing governance and security standards, and enabling enterprise-wide automation.
Q1 — Vendor and platform profile
Insert adesso's answer here.
Q2 — Agentic AI in insurance
Insert adesso's answer here.
Q3 — Differentiating Agentic AI characteristics
Insert adesso's answer here.
Q4 — Architectural building blocks
Insert adesso's answer here.
Insert adesso's answer here.
Q1 — Vendor and platform profile
📎 Slides to be embedded — placeholder for Faktor Zehn platform profile slides
Q2 — Agentic AI in insurance
At Faktor Zehn, Agentic AI is integrated into software modules wherever it helps support business decisions, automate complex processes, and sustainably increase efficiency across core insurance processes. Together with partner Hayuno — a software company specializing in AI agents for the insurance industry — Faktor Zehn is actively driving the use of Agentic AI within core insurance systems.
AI agents are currently deployed in the following areas:
- Knowledge assistance — intelligent support for business decisions with transparent source referencing
- Customer correspondence — automated, context-based generation of responses to customer inquiries
- Document verification — AI-supported matching of incoming documents with reference data, including plausibility checks
- Document processing — automated analysis, structuring, and summarization of complex, unstructured content
- Fraud detection — analysis of documents and images to identify potential anomalies, including calculation of a fraud score
The approach focuses on a holistic view of processes, achieving a high degree of automation through a combination of deterministic workflows and AI agents.
Q3 — Differentiating Agentic AI characteristics
Production-ready AI agents are available for all use cases listed above. Key client deployments include:
- A domain-specific knowledge tool for service employees, significantly improving response quality
- AI-powered customer service that generates concrete response suggestions, significantly reducing response times
- Automated document verification for incoming invoices through intelligent matching with existing reference data and detection of inconsistencies
The key differentiator is the consistent combination of an open, modern core system architecture with integrated Agentic AI capabilities. By providing REST services, Faktor Zehn opens its core insurance solutions for seamless AI integration, ensuring customers remain flexible and can independently manage technology decisions. Agents can be developed and operated by Faktor Zehn, customers, or third-party providers.
The approach combines technological openness, process integration, and regulatory security — with built-in support for data protection, data sovereignty, and the EU AI Act.
Q4 — Architectural building blocks
Smart Business Layer — AI-centric applications are positioned above systems of record in a layer consisting of workflows, task management, and AI agents. Capabilities include policy and coverage inquiries with transparent reasoning, intelligent customer response generation, invoice and document validation, automated document analysis, and fraud scoring. Available via chat interface or integrated into other front ends via REST. Adoption path ranges from standalone assistants to semi- or fully autonomous agents.
Agentic AI Layer — LLM-agnostic approach; model selection depends on insurer context, existing vendor relationships, and target infrastructure. Agents interact with operational systems via tools/REST services and are not limited to Faktor Zehn-owned systems. No lock-in to a Faktor Zehn cloud; operation by Faktor Zehn or by the customer is both possible.
Data Layer — The Faktor Zehn Suite supports all common relational databases (Oracle, PostgreSQL, Microsoft SQL Server). For AI use cases such as knowledge bases or vector search, no fixed platform standard is prescribed — the data architecture is adapted to the customer's existing landscape. Key requirement is access to structured core data and, depending on the use case, additional document or knowledge sources.
Core Transaction Layer — Technology-agnostic and modular target architecture. End-to-end use cases integrate with surrounding systems (inbox/task management, document sources, operational systems) via REST. No proprietary document management or input management solution is imposed. AI outputs can prepare or trigger follow-up actions within the core process.
Infra and Cloud Layer — Flexible deployment: on premises, private cloud, or customer-specific/partner-operated cloud environments. For Faktor Zehn–managed cloud scenarios, operations are delivered in collaboration with codecentric AG, supporting multiple providers including AWS. Sovereign cloud setups (e.g., hosting within Germany) are supported, ensuring compliance with regulatory and data-residency requirements.
Faktor Zehn's roadmap for the next 12 months focuses on three areas:
- Knowledge assistance (rolling out over 12 months) — standardized integration of AI services for context-based rule searches with transparent source references, targeting improved quality and traceability of business decisions
- Claims automation — further automation of the claims notification process to significantly increase STP rates and reduce processing times
- Service operations — full integration of AI-supported assistance solutions, including employee support for domain-specific questions and automated quality-assured responses to customer inquiries
Strategically, Faktor Zehn is focused on driving holistic process optimization through AI — particularly in claims — while embedding AI into the insurance core in a controlled, transparent, and compliant manner. The goal is to combine technological innovation speed with regulatory security, creating a robust foundation for productive AI use and long-term value creation.
Q1 — Vendor and platform profile
Peak3 offers two core products: Graphene, an enterprise-grade, modular, cloud-native SaaS insurance core platform, and Fusion, an insurance sales and orchestration platform for scalable distribution models. Peak3 additionally provides a growing portfolio of agentic AI solutions that enhance and complement its main product suite.
Graphene supports all major lines of business on one platform: property & casualty (retail and commercial), life, individual and group health. Its no-code product configurator enables rapid product development and iteration. The platform covers the entire insurance value chain end-to-end: product management, distribution/sales, underwriting & rating, policy administration, claims & benefits, billing, finance, and digital engagement.
Graphene is deployed as a cloud-native, cloud-agnostic SaaS platform with flexible deployment options: regional public instance (Peak3-managed, multi-tenant public SaaS), private single/multi-tenant instance (Peak3-managed private SaaS), or private single/multi-tenant instance (client/partner-managed). Its full microservices architecture supports end-to-end deployment or selective module implementation.
Peak3 has been recognized as a Technology Standout in Celent's XCelent Awards 2025 (Life & Health, P&C), a Major Player in IDC MarketScape for worldwide cloud-based P&C platforms 2025, and a Leader/Rising Star in ISG Provider Lens 2024.
Q2 — Agentic AI in insurance
Peak3 builds agentic AI solutions deployable across the value chain, following a twofold approach: pre-built AI agents that are production-ready and can be deployed quickly, and an AI orchestration platform that enables insurers to co-build and manage custom agents across the lifecycle.
AI agent capabilities are organized into three categories:
- Assessor and Triage Agents — autonomously assess transactions, develop risk reports and scores with transparent evidence chains, and assist experts with final decisions
- Intelligent Document Processing Agents — vision-powered AI agents that intelligently recognize, extract data from, and automate the validation of documents
- Voice & Text Customer Agents — manage and automate customer interactions 24/7 through conversational multi-modal AI agents
Current main focus has been on building these capabilities fully for claims, with expansion into sales and underwriting underway. Separately, agentic AI is being incorporated into Graphene to automate backend configuration (currently in POC stage).
Q3 — Differentiating Agentic AI characteristics
The following agentic AI use cases are production-ready and in implementation with clients today:
- Voice & Text Customer AI Agents — conversational FNOL agents (voice and text) that guide customers through claim submissions, collect structured data, validate documents, and integrate directly into claims workflows
- Intelligent Document Processing (IDP) Agents — combining OCR with multimodal LLMs to automatically extract, classify, validate, and understand information from complex claims documents
- AI Assessor & Triage Agents — agentic AI for claims investigation that deconstructs claim review tasks and executes risk checks using a deep research framework
Key differentiators include: AI-native core integration with full microservices and open architecture; standalone or integrated deployment of AI agents decoupled from the Graphene core; LLM and cloud agnosticism with long-term optionality; and compliance-first architecture with explainability, observability, and human-in-the-loop governance.
Q4 — Architectural building blocks
Smart Business Layer — AI solutions are designed to be multi-modal and support multiple languages. Clients can configure underlying capabilities to fit their needs on top of pre-defined capabilities (e.g., FWA rules in claims assessment).
Agentic AI Layer — LLM-agnostic multi-model framework with a Model Gateway/Proxy for routing and load balancing. Protocols supported include MCP, A2A, Open APIs, webhooks, tool calling, and SSE APIs. Key components include Agent Management, Agentic Workflow Engine, RAG Pipeline, LLM Operations, and a CI/CD + LLMOps factory.
Data Layer — Graphene provides built-in big data platform capabilities including a Data Warehouse and Customer Data Platform (CDP). Real-time data synchronization, event streaming, hot case caching, and third-party data API integration are supported. Additional data such as SOPs, rule libraries, policy terms, and compliance guides are managed via RAG pipelines with vector databases.
Core Transaction Layer — Graphene is designed as an end-to-end core insurance platform with most required systems built in or pre-integrated. External systems (scanning solutions, telephony, payment gateways) are integrated as needed.
Infra and Cloud Layer — Cloud-native, cloud-agnostic, and flexible deployment. Built entirely on microservices with containerized deployments, primarily on public hyperscalers with private/sovereign cloud options available.
Peak3's agentic AI roadmap for the next 12 months:
- Available now / Q2 (continuous enhancements) — Agentic Claims: end-to-end AI-powered claims processing including FNOL chatbots, IDP, AI assessor/FWA, and claims assistant with HITL. Target value: STP uplift, cycle time reduction, improved fraud detection, reduced claims leakage
- Q2/Q3 — Expanded AI Orchestration Platform: enhanced agent lifecycle management, broader MCP/A2A integration library. Target value: faster time-to-market for custom AI agents
- Q3 — Agentic Underwriting: AI-assisted risk assessment, automated data gathering, underwriting decision support. Target value: faster quote-to-bind cycle times, improved risk selection accuracy
- Q3/Q4 — AI telemarketing and customer-facing agentic applications. Target value: increased conversion rates, reduced cost per acquisition
Strategically, Peak3 is focused on deepening and scaling pre-built AI agents — expanding from claims into underwriting and sales — while co-building with customers and maintaining long-term optionality across deployment options and model choices, including Sovereign AI.
