AI in the insurance industry: From pilot projects to enterprise-wide performance

Artificial intelligence is no longer merely a topic of innovation in the insurance industry. It is increasingly becoming the sector’s new operating system.

Across Europe, insurers are experimenting with copilots, chatbots, generative AI models and automation solutions. Nevertheless, many organisations fail to translate isolated innovation initiatives into structural performance improvements. The result is activity without company-wide impact.

The real challenge no longer lies in experimentation. It lies in scaling AI across all business areas in such a way that it is economically effective, operationally embedded and regulatory-compliant.

The central industry problem: AI as a collection of isolated tools

Most insurers take a tactical approach to the use of AI:

• A chatbot in customer service
• OCR in claims processing
• A fraud model in underwriting
• A productivity assistant in IT

Each use case optimises locally. Few organisations are holistically redesigning entire business domains.

Without domain-driven integration, AI remains an efficiency lever rather than a driver of transformation. At the same time, customer expectations continue to rise. Policyholders no longer compare insurers with competitors, but with digital-native platforms that offer real-time services and AI-powered personalisation.

The structural consequences of fragmented AI

A fragmented rollout of AI not only increases architectural complexity – it also limits structural value creation. If underwriting intelligence is not systematically linked to claims experience, learning effects remain partial. If conversational AI operates without integration into central policy and portfolio systems, customer journeys remain inconsistent. If productivity tools are implemented without redesigning the underlying workflows, operating models remain fundamentally unchanged.

The result is local optimisation rather than company-wide leverage. AI generates activity, but no operational scaling effects.

From tools to AI-augmented domains

An AI-native insurer does not merely implement models into existing structures; it redesigns specialist domains around augmented workflows.

Underwriting is supported by embedded risk intelligence that synthesises structured and unstructured data in real time. Claims functions integrate triage engines that support prioritisation and documentation processes. Sales teams work with contextual co-pilots that provide recommendations during customer interactions. Governance functions continuously monitor model behaviour and defined escalation thresholds.

AI impact at the domain level across insurance lines

A sustainable competitive advantage does not arise from a single AI technology. It results from the combination of predictive models, decision intelligence, cognitive automation and generative systems with the respective insurance lines.

Life insurance

Life insurance combines structured actuarial data with extensive unstructured documentation.

The impact of AI extends across several levels:

• Predictive risk models for mortality and lapse behaviour
• Machine learning-based segmentation in underwriting
• NLP-supported synthesis of medical documentation
• Automated document verification and fraud detection
• Generative systems for creating personalised insurance quotes
• Advisor co-pilots to support complex customer discussions

Here, predictive precision and generative augmentation work together. The aim is not merely automation, but improved underwriting quality and greater effectiveness in advisory services.

Health insurance

Health insurance requires real-time decision support and strict regulatory compliance.

AI capabilities include:

• Predictive models for identifying high-risk populations
• Anomaly detection in claims patterns
• Computer vision for validating medical documentation
• NLP engines to support pre-authorisation processes
• Conversational AI for interacting with service providers and policyholders

In this context, explainability and auditability are just as crucial as performance. AI must be designed with transparency in mind.

Commercial SHUK insurance

Risk assessment in commercial SHUK insurance requires the interpretation of technical reports, site inspections, satellite imagery and complex contractual structures.

AI enhances these capabilities through:

• Scenario-based catastrophe modelling
• Portfolio optimisation algorithms
• Geodata analytics and interpretation of satellite imagery
• Automated synthesis of technical reports
• Fraud detection and modelling of reinsurance recourse

Predictive intelligence strengthens the underwriting discipline, whilst generative systems accelerate documentation and reporting.

Private SHUK insurance

Private SHUK insurance is characterised by high volumes and standardisable workflows.

AI applications include:

• Dynamic pricing models
• Real-time fraud scoring
• Behaviour-based risk segmentation
• Automated FNOL (First Notice of Loss) classification
• Claims complexity forecasting
• Conversational and generative AI for customer interaction

Here, scalability and automation efficiency are the key success factors.

Scaling AI securely: Engineering before algorithms

European insurers operate within the regulatory framework of GDPR, DORA, Solvency II, IFRS 9/17 and, increasingly, the EU AI Act and regional data sovereignty requirements.

Scaling AI requires three specifically designed foundational building blocks:

AI-enabled data architecture

Fragmented legacy systems and inconsistent data hinder any AI scaling. Insurers require unified, domain-specific data models, structured and unstructured ingestion pipelines, semantic layers, and clearly defined access patterns.

The performance of AI depends less on model complexity than on the integrity of the underlying architecture.

A stable foundation for successful AI enablement is created primarily through a robust data base and clearly defined standards. The transformation process towards an AI-driven enterprise presents a key opportunity to consolidate data sources, enhance data quality and clarify responsibilities, as well as to establish uniform interfaces and governance structures. ‘Smoothing out’ the data foundation not only improves efficiency across business units but also enhances the integration capabilities and reliability of AI solutions – from development through to operation and scaling within the organisation.

Integrated Governance & Responsible AI

Insurance decisions influence financial security and individual life circumstances. Enterprise-ready AI must therefore encompass:

• Traceability of model outputs
• Tools for explainability
• Bias monitoring within MLOps pipelines
• Human-in-the-loop escalation mechanisms

Regulatory compliance thus becomes a competitive trust factor.

Business-driven transformation KPIs

AI initiatives must be directly linked to measurable economic impact:

• Improvement in the combined ratio
• Reduction in the cost-to-serve
• Increase in the sales conversion rate
• Reduction in claims processing times

Model accuracy without an impact on the P&L is not transformation.

Regulation as an architectural guardrail

European regulation – from GDPR to DORA and the EU AI Act – defines clear requirements for transparency, traceability and model validation.

Insurers that integrate these requirements directly into their architecture scale more sustainably than those that add compliance as an afterthought.

Regulation thus becomes a design parameter – not a corrective tool.

Build, buy or partner?

Insurers should build differentiating AI capabilities themselves – such as proprietary risk assessment logic or specific risk appetite models. Standardised enablers such as OCR or basic analytics can be efficiently procured. For complex modernisation projects and scalable AI engineering, a partnership is advisable.

The decision should be guided by three questions:

• Differentiation potential – Does the capability create strategic differentiation?
• Complexity and depth of integration – How deep is the required integration into core systems?
• Speed requirements – How great is the time pressure?

Regardless of the model, architectural sovereignty and model governance must remain with the insurer. Long-term autonomy is non-negotiable.

What it means to become AI-native

AI-native insurers consistently design risk assessment, claims management, sales and portfolio management around AI-augmented workflows. Employees work with digital agents – not alongside them.

Those who act now set new performance standards. Those who hesitate will face competition in the future from structurally faster and more intelligent operating models.

AI is not an upgrade, but a structural transformation.

Our perspective

At adesso, we combine in-depth insurance expertise with AI engineering expertise, regulatory sovereignty and scalable delivery models across Europe. We support insurers in shaping the transition:

• From pilot projects to productive scaling
• From experiments to measurable economic impact
• From isolated tools to AI-native domains

Sustainable AI transformation requires strategic clarity, solid technical foundation architectures and disciplined implementation.