AI in Insurance in 2025: From Technology Adoption to Leadership Execution
Over the few last days of 2025, I deliberately stepped back from vendor decks and conference slogans and spent time reading what 2025 research, regulators, and insurers themselves are saying about AI in insurance. Not selectively, but end to end - consulting studies, supervisory opinions, insurer disclosures, and early case studies on generative and agentic AI.
What emerges is not a story of experimentation anymore. It is a story of execution gaps.
In 2025, AI in insurance has crossed a point of no return. The technology works. The value is visible. But only a small group of insurers is truly able to operate AI at scale - and the difference is no longer technical.
1. AI Is Widespread - but Still Shallow
Multiple 2025 studies converge on a consistent picture. Insurance is now among the most active industries in AI adoption, yet only a minority of insurers have translated this activity into structural change.
Boston Consulting Group notes in its 2025 analysis Insurance Leads in AI Adoption. Now It’s Time to Scale that while most insurers run multiple AI initiatives, only 7–10% have embedded AI across core end-to-end workflows. IBM’s Insurance in the AI Era reaches a similar conclusion, observing that AI is often deployed “around” processes rather than “inside” them.
This distinction matters.
AI layered on top of fragmented processes delivers marginal gains.
AI embedded into redesigned workflows changes economics.
The gap between the two is not model performance. It is ownership, governance, skills, and decision design.
2. Claims: Where AI Becomes Operational First
Claims processing remains the clearest example of AI moving from theory into daily operations.
By 2025, AI is no longer just assisting claims handlers - it is actively reshaping how claims flow through organizations. IBM reports an average 15–30% reduction in claims handling time where AI is integrated into document intake, triage, and assessment. EIOPA’s 2025 supervisory analysis confirms that claims and fraud detection are now considered among the lowest-risk, highest-value AI use cases when human oversight is maintained.
What struck me most while reviewing insurer disclosures is how different the outcomes are depending on ambition.
Ping An’s 2025 update is a useful reference point. The group reports that 93% of underwriting decisions are automated and the average claim is processed in 7.4 minutes. This is not a result of a single algorithm, but of treating AI as part of core operational design - not as an innovation experiment.
The lesson is uncomfortable but clear:
claims automation succeeds where organizations accept that speed, control, and trust must be designed together, not sequentially.
3. Fraud and Risk: AI as a Defensive Capability
Fraud detection has quietly become one of AI’s most reliable value generators in insurance.
EIOPA’s 2025 opinion on AI governance highlights that machine-learning-based fraud detection significantly improves detection rates without materially increasing consumer risk, provided insurers maintain explainability and escalation paths. Several insurers report double-digit reductions in fraud leakage after introducing AI-based anomaly detection, particularly in motor and property insurance.
At the same time, 2025 marks a turning point in the nature of fraud itself. Synthetic documents, manipulated images, and AI-generated identities are no longer edge cases. Insurers are increasingly deploying AI against AI - computer vision to detect altered images, voice biometrics to prevent call-center fraud, and network analytics to uncover organized abuse.
Fraud prevention has become infrastructure, not analytics.
And infrastructure requires continuous investment, governance, and human accountability.
4. Underwriting: Where Caution Still Dominates
Underwriting remains more conservative - and for good reasons.
Conning’s 2025 C-Suite Verdict on AI shows that fewer than a quarter of insurers use AI at scale in underwriting today, despite over 80% expecting it to become core within five years. Regulatory scrutiny, explainability requirements, and the long tail of legacy systems all play a role.
What has changed in 2025 is how generative AI is entering underwriting workflows. Rather than making autonomous decisions, GenAI increasingly supports underwriters by summarizing inspection reports, extracting risk factors from unstructured data, and accelerating case preparation.
This is a pragmatic evolution. It respects regulatory constraints while delivering immediate productivity gains. The EU’s AI Act reinforces this direction by classifying AI used in life and health underwriting as high-risk, demanding transparency and human accountability - effectively pushing insurers toward augmentation before automation.
Underwriting will change.
But it will change with humans in control, not in their absence.
5. Customer Interaction: AI Works - When It Augments People
Customer-facing AI has matured significantly in 2025, especially with the rise of generative models.
The Geneva Association’s 2025 research shows that nearly 70% of customers have already used generative AI tools when researching insurance. At the same time, the same research highlights persistent concerns around empathy, accountability, and trust.
The most successful insurers are not replacing agents with bots. They are equipping agents with AI copilots. IBM reports productivity gains above 30% where AI assists customer service teams by summarizing histories, suggesting responses, and reducing administrative load.
What this confirms is something many leaders underestimate:
customer trust scales slower than technology.
AI improves experience when it removes friction - not when it removes responsibility.
6. Generative vs. Agentic AI: The Real Shift Ahead
One of the most important conceptual shifts in 2025 is the growing distinction between generative AI and agentic AI.
Generative AI is already widespread. It creates, summarizes, explains, and supports decision-making. Most insurers now use it internally, and many use it in customer interaction.
Agentic AI is different. It plans, orchestrates, and executes tasks autonomously within defined boundaries.
Celent’s 2025 research describes agentic AI as the next phase of insurance automation, particularly in claims orchestration and service operations. Yet adoption remains cautious: fewer than a quarter of insurers plan meaningful agentic AI deployment before 2026.
This caution is healthy. Agentic AI forces organizations to confront questions they have long postponed:
- Who owns decisions?
- Where does accountability sit?
- How do we audit automated action, not just automated advice?
The technology is advancing faster than organizational clarity. That gap will define the next two years.
7. One Industry, Different Speeds
Looking globally, AI adoption in insurance is no longer uneven in whether it happens, but in how it is governed.
European insurers move deliberately, prioritizing explainability and compliance. US insurers innovate aggressively but struggle to scale across fragmented systems and regulation. Asian insurers, particularly in China, treat AI as infrastructure and embed it deeply into ecosystems.
None of these approaches is universally “right.”
But each reveals a trade-off between speed, control, and resilience.
8. The Real Constraint Is Leadership, Not Technology
After reviewing the 2025 research in depth, one conclusion becomes difficult to ignore. The limiting factor for AI in insurance is no longer access to technology. Models are available. Capabilities are proven. Investment is flowing.
What truly constrains progress is the ability to run AI as part of the operating model, not as a parallel initiative.
AI does not fix broken structures. It magnifies them. Where decision rights are unclear, processes fragmented, or accountability diluted, AI simply accelerates dysfunction. Where foundations are solid - clear ownership, disciplined execution, and strong governance - AI becomes a genuine multiplier.
In 2025, AI is no longer revealing gaps in technology. It is revealing gaps in leadership.
Closing Thought
Insurance is no longer experimenting with AI. It is being reshaped by it.
The organizations that will succeed are not those that move fastest, but those that hold speed and foundations at the same time.
That balance is not a technical problem. It is the core of modern leadership.
Sources (2025)
Boston Consulting Group (BCG)
Insurance Leads in AI Adoption. Now It’s Time to Scale (2025)
https://www.bcg.com/publications/2025/insurance-leads-ai-adoption-time-to-scale
IBM Institute for Business Value
Insurance in the AI Era (2025)
https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/ai-insurance-2025
EIOPA – European Insurance and Occupational Pensions Authority
Artificial Intelligence Governance and Supervision in Insurance (2025)
https://www.eiopa.europa.eu/publications/artificial-intelligence-governance-and-supervision-insurance_en
Ping An Group
Driving Innovation with Generative AI (2025)
https://group.pingan.com/media/news/Driving-Innovation-with-Generative-AI.html
Celent (Omdia)
Shedding Light on Agentic AI in Insurance (2025)
https://www.celent.com/insights/agentic-ai-insurance-2025
Conning
AI in Insurance: The C-Suite Verdict (2025)
https://www.conning.com/insurance-ai-c-suite-survey-2025
The Geneva Association
Generative AI and the Insurance Customer Journey (2025)
https://www.genevaassociation.org/research-topics/artificial-intelligence/generative-ai-insurance-customer-journey
OECD
AI, Insurance and the Future of Work (2025 update)
https://www.oecd.org/artificial-intelligence/ai-insurance-future-work.htm