By Cédric Jadoul, Partner | Intelligent Automation & Christophe Felice, Partner | Insurance Consulting — Deloitte Luxembourg

The insurance sector is entering a new phase in its digital transformation. After a decade focused on automating individual tasks—such as data extraction, document processing, and notifications- the question is no longer "What can we automate?" but "How do we orchestrate collaboration between humans and AI to execute entire processes?" 

This shift in perspective is redefining the insurance operating model.

The situation is widely recognized across the industry: while individual tasks have become faster, end-to-end processes remain slow. The industry has automated gestures without rethinking how work is coordinated. Today, thanks to the growing maturity of AI—and particularly the rise of AI agents capable of reasoning, acting, and collaborating—it is becoming possible to transform not just isolated tasks, but whole business processes.

A sector under pressure from all sides

The signals are converging: the vast majority of industry leaders now view AI as an immediate lever for transformation rather than a distant experiment. The adoption of Agentic AI ranks among the top investment priorities for the next two years, and the expected productivity gains are significant, with prospects for operational cost reductions measured in tens of percent.

"What we observe on the ground is a growing gap between the digital ambition displayed by insurers and the reality of their operations," notes Christophe Felice. "Teams are under pressure, systems are fragmented, and coordination between departments still relies largely on manual exchanges. The current model is reaching its limits."

The insurance operating model built on manual coordination between distribution, underwriting, policy management, and claims, no longer scales.

From task automation to process transformation

Over the past decade, the industry has progressed in successive waves with workflow, robotic process automation (RPA), intelligent document processing, and more recently, AI assistants. Each wave has created value, but predominantly at the level of individual tasks.

"Most insurers we work with have already automated dozens, even hundreds of tasks," explains Cédric Jadoul. "But automating tasks and transforming a process are not the same thing. The real breakthrough happens when AI moves beyond assistance to actively contributing to the coordination of work, within a controlled framework."

This represents a fundamental paradigm shift. The goal is no longer simply to accelerate human actions. It is to give AI the ability to coordinate an entire process, mobilizing the right resources (human, software, and decision-making) at the right time, within a governed framework.

In practice, this means that specialized AI agents, each mastering a specific domain such as risk analysis, coverage verification, fraud detection, or customer interaction, work together to execute a process end to end. Human involvement does not disappear; it evolves. Responsibilities shift from execution to oversight and from processing to decision-making.

"The human remains at the center," insists Christophe Felice. "But their role evolves. A claims handler or an underwriter should no longer spend their time searching for information across five different systems. They should focus on what only they can provide: judgement, client relationships, and decisions in complex cases."

Concrete examples across the entire value chain

This model already applies to core insurance processes. For example:

• Life claims: Upon a death notification, AI verifies document completeness, retrieves policy data, analyzes coverage and exclusions, detects fraud signals, and presents the handler with a complete file and a recommendation. The handler focuses on the final judgement. Processing time is significantly reduced, at a moment when the beneficiary expects speed and empathy.

• Underwriting: When an application is received, AI extracts medical and financial data, assesses the risk profile, identifies cases requiring referral, and proposes pricing with its rationale. The underwriter no longer searches for information; they only challenge it.

• Policy servicing: For requests such as a beneficiary change, an address update, or a surrender, AI classifies the request, checks eligibility rules, executes simple modifications, and escalates complex cases to an expert. The customer receives a response in hours rather than days.

• Distribution: AI qualifies leads in real time, analyzes protection needs, and prepares a personalized proposal. The advisor enters the meeting with a pre-structured file and can focus on the client relationship and quality of advice.

"These are not theoretical concepts," emphasizes Cédric Jadoul. "We have designed and demonstrated these models on real insurance processes. The pattern is replicable: specialized agents that collaborate, humans that oversee, and a governance layer that ensures control at every step."

Regulation as an enabler, not a barrier

Many decision-makers assume that regulation will slow down the large-scale deployment of AI. In reality, field experience shows the opposite.

The AI Act requires traceability, explainability, and human oversight. The Digital Operational Resilience Act (DORA) demands monitoring and resilience. Solvency II calls for clear governance and documented controls.

"A well-designed operating model, where humans and AI collaborate within a structured framework, natively meets these requirements," explains Christophe Felice. "Every action is traced, every decision is justifiable, every exception is escalated to a human. Insurers that embed this governance from the start are not slowed down by regulation; they turn it into a competitive advantage."

The institutions that will scale AI fastest will not be those with the most advanced models. They will be those that can prove how their processes work, how decisions are made, and how risks are controlled.

What this means for decision-makers

If you are leading an insurance transformation, on either the business or technology side, here is what this evolution means in practice:

• Think processes, not use cases: The value no longer lies in an isolated chatbot or a document extractor. It comes from redesigning complete workflows, from notification to settlement, and from application to issuance.

• Prepare for the role shift: Business experts will increasingly move from executing tasks to supervising AI-driven processes. This requires new skills, new KPIs, and serious change management.

• Embed governance from the start: Traceability, auditability, escalation to humans, these are not constraints added after the fact. They are the foundation of trust and scalability.

• Start with one process, prove the value, then scale: Begin with one domain, one mixed human-AI team, and one proven control framework. Then industrialize and replicate.

"Our recommendation is always the same," concludes Cédric Jadoul. "Start with a process that hurts, one that is heavy on documents, rich in rules, full of exceptions. Show what AI can do within a controlled framework. Then industrialize and extend the model."

Final thought

The future of insurance will not be shaped by isolated AI use cases, but by a new operating model in which humans and AI work together in a controlled, transparent, and customer-centric way.

"The insurers that will win the next decade will not be those with the most advanced AI technologies," summarizes Christophe Felice. "They will be those that managed to make them work together from underwriting to claims."

Cédric Jadoul is a Partner at Deloitte Luxembourg, specialized in Intelligent Automation. Christophe Felice is a Partner at Deloitte Luxembourg, specialized in Insurance Consulting. Together, they support insurers in transforming their operations and operating model through AI.