AI in General Insurance and the Rise of Agentic Systems - Actuaries Digital

In Part 1 of this multi-part series, we examine how AI fits into the general insurance value chain, recent examples of real insurance use cases from insurers, and the implications from ‘silent AI’.

How agentic AI differs from generative AI: From single responses to ongoing processes

Actuaries have long been early adopters of analytical techniques that turn complex data into better insurance decisions. From the widespread adoption of generalised linear models to increasingly sophisticated machine-learning techniques, the profession has continually adapted as new technologies have become available.

Artificial intelligence is the next step in that evolution, but the pace and breadth of change are different.

Generative AI (GenAI) has already moved well beyond experimentation into practical insurance applications, from coding and analytics to underwriting and claims. Agentic AI takes this a step further: rather than simply generating an output in response to a prompt, an agentic system can work towards a defined goal through an iterative cycle of planning, acting, observing and adjusting. This repeated cycle of planning, acting and responding to new information is often described as an “agentic loop”.

Figure 1: Illustrative agentic loop

These more complex AI processes are being utilised by more and more insurers. So where is this AI actually being used in the general insurance value chain?

Where is AI used in the general insurance value chain?

AI is appearing across general insurance, from pricing and underwriting through to distribution and claims.

Table 1 below widens the lens, giving colour to where AI could create value at each stage of the insurance value chain and the challenges that come with it.

Table 1 – Opportunities and challenges across the insurance value chain

Insurance value chain	Potential AI opportunities	Key AI risks / challenges
Product design and development	Compare Product Disclosure Statements (PDSs) and policy wordings, identify coverage gaps and support faster product development. Support more tailored propositions using richer customer and claims information.	Product decisions still require judgement around commercial intent, regulation and unintended coverage. Greater personalisation raises fairness considerations and how far insurers should differentiate at individual-customer level.
Pricing and actuarial analytics	Faster pricing cycles through automating key steps in the actuarial control cycle. GenAI can help with code review, debugging and documentation, while agentic AI may autonomously handle key parts of the model and report creation process. AI can unlock unstructured data, such as documents, claim notes and images, enriching datasets for pricing and risk segmentation.	A model or piece of code can be technically correct but still answer the wrong business question. Actuaries still need to understand the data, assumptions and methods, rather than treating AI as a black box. As technical barriers fall, judgement, business understanding and communication become increasingly important differentiators.
Underwriting	GenAI can review broker submissions, risk surveys, financial statements and contracts, extract key information, identify gaps, compare risks against appetite, and summarise coverage and exclusions, especially in document-heavy commercial underwriting. The bigger opportunity may be to redesign the workflow, not just automate individual tasks.	AI-generated summaries can shape what an underwriter sees first, influencing judgement even where a human makes the final decision. Controls are needed around referrals, authority limits, hallucinations and missing information, with meaningful human review needed rather than simple approval of a pre-populated recommendation.
Distribution and customer acquisition	AI-powered search and digital assistants could make it easier for customers to compare products, understand coverage and shop around. GenAI can support lower-cost and more personalised distribution through quoting, product explanations and tailored offers. Agentic AI may increasingly act on both sides of the transaction, guiding customers through purchase journeys or comparing and transacting on their behalf.	Easier comparison may increase switching and price sensitivity, changing the role of brands, aggregators and traditional distribution channels. Inaccurate product explanations or recommendations could create conduct and customer-outcome risks. There remains a gap between frontier AI capability and practical deployment at scale.
Claims, customer service and complaints	Automate claims intake and review through document summarisation, image analysis, triage and fraud detection. Support customer communication and complaints handling by summarising claim status, drafting responses and identifying cases needing escalation. Agentic AI may progress straightforward claims end-to-end, including requesting information, appointing suppliers and settling low-complexity claims.	Over-automation may worsen customer experience, particularly for complex or stressful claims where customers prefer human support. Incorrect triage, fraud flags or coverage interpretation can directly affect customers, making clear human checkpoints and escalation pathways essential. Errors can scale quickly across many claims, increasing the need for strong audit trails, control override and outcome monitoring.

How are insurers using AI today?
Real-world examples

But these opportunities aren’t just theoretical. There are current examples of insurers using AI to change entire processes and deliver real-world value.

While actuarial and analytical teams can use AI to speed up core tasks like coding, data processing and modelling, GenAI and agentic AI can reshape organisational workflows. As these tools become more accessible, judgement, business understanding and the ability to challenge AI outputs become increasingly important.

Here are just a handful of recent examples of AI being used by insurers:

Pricing and risk selection: AI driving decisions

Companies like Akur8 are now releasing end-to-end agentic workflows for ground-up rate making. Their Akur8 Agents ^[1] are designed to guide actuaries from data discovery to model deployment.

Underwriting: GenAI becomes a copilot

QBE’s Cyber Underwriting AI Assistant ^[2] reviews submissions for completeness and relevant risk controls, with QBE reporting a 65% reduction in review time. The underwriter remains responsible for the final decision, but AI increasingly shapes how information is extracted and presented.

Claims: AI moves from assisting to acting

Suncorp’s Single View of Claim^[3] uses GenAI to summarise claim histories and recommend next actions.

Allianz ^[4] goes further with an agentic AI solution for food-spoilage claims, which it says can reduce processing and settlement times by around 80%.

These examples show AI moving from supporting analysis and recommendations to now increasingly taking on parts of the insurance workflow itself. As autonomy increases, clear escalation points within human accountability frameworks become increasingly important.

AI insurance coverage and 'silent AI'

AI is changing not only how insurers work, but also the types of risks they may need to cover.

There is heightened risk with professionals increasingly relying on AI in their work. In Dayal [2024] FedCFamC2F 1166^[5], a solicitor relied on AI-generated legal research containing false authorities, leading to professional and regulatory consequences and notification to the Legal Practitioners’ Liability Committee. While the judgment does not confirm professional indemnity (PI) coverage, it illustrates a potentially insurable AI-related professional liability exposure.

The insurance market is also beginning to respond. Lloyd’s describes Armilla AI ^[6] as a Managing General Agent (MGA) and Lloyd’s coverholder that is focused on risks such as algorithmic error, model drift and generative-AI hallucinations.

For commercial insurers, the broader issue is how these emerging exposures fit within existing PI, cyber, Directors and Officers (D&O) and technology errors and omissions (Technology E&O) policies. Some losses may already fall within existing wordings, creating potential “silent” exposure, while reliance on common models and vendors could increase aggregation risk. This may lead insurers to refine underwriting questions, place greater focus on clients’ controls and human oversight, and consider more explicit AI-related cover.

The next questions...

AI is moving from a supporting role into general insurance workflows. This changes how risks are assessed, how underwriting and claims are managed, and the exposures insurers may need to cover.

As AI’s role expands, the key question is not just what AI can do, but how much authority it should be given and how that authority should be governed. As insurers move from AI that predicts to AI that generates, and ultimately AI that acts, the governance challenge also shifts – from a traditional focus on model performance towards questions of decision authority, human oversight and accountability.

The General Insurance Practice Committee (GIPC) AI Working Group is continuing to monitor this evolving area. Part 2, Governing AI in general insurance: what practitioners need to ask, will focus on a practical approach to prioritising AI risks, defining accountability and applying governance principles to real insurance use cases.

^[1] Akur8 Agents: Expert Actuarial AI, built by Actuaries for Actuaries .

^[2] QBE, “QBE scales GenAI solutions across operations”, https://www.qbe.com/newsroom/news/qbe-scales-gen-ai-solutions-across-operations , accessed 14 October 2026.

^[3] Suncorp Group, “FY24 look back: a year of major tech milestones for Suncorp”, https://www.suncorpgroup.com.au/news/news/major-tech-milestones , accessed 14 October 2026.

^[4] Allianz Australia, “Allianz wins 2026 Canstar Innovation Excellence Award”, https://www.allianz.com.au/about-us/work-with-us/partners/news/allianz-wins-2026-canstar-innovation-excellence-award.html , accessed 14 October 2026.

^[5] Federal Circuit and Family Court of Australia, Dayal [2024] FedCFamC2F 1166, https://www.lexisnexis.com.au/ContentNotifier/URJNotifier/LinkMan?bcnum=2414379.htm&jurisdiction=fmca , accessed 14 October 2026.

^[6] Lloyd’s, “Armilla AI”, Lloyd’s Lab Accelerator Alumni, https://www.lloyds.com/insights/lloyds-lab/programmes-and-initiatives/lloyds-lab-accelerator/alumni/armilla-ai , accessed 14 October 2026.