Insurers say hello to generative AI’s new partner, agentic AI

The future of agentic AI and the rapid pace of change

One thing insurers are learning is that the launch of GPT models came with a blistering pace of change. And here comes agentic AI, which refers to AI systems that can act autonomously and make decisions and execute tasks based on data analysis.

Agentic AI is slowly beginning to gain traction in the insurance industry and is quickly becoming the new topic of discussion for the next step in AI applications.

Agentic AI holds significant future potential for transforming insurance operations across various functions. This leap toward more independent AI agents promises to reshape how insurance companies operate, leading to even more increased efficiency, improved customer experiences and enhanced risk management.

Already, a small percentage of insurers are putting agentic AI solutions through pilots and proof of concepts, with a select few (4%) transitioning to production. As noted in Celent’s most recent gen AI survey, 22% of insurers stated they will have an agentic AI-based solution in production by year-end 2026.

The potential for agentic AI in insurance operations

Customer service

Virtual assistants, handling complex policy queries, can provide around-the-clock hyper-personalized support across multiple channels and guide customers through processes such as onboarding or claims submission. Agentic AI can also proactively engage policyholders with timely reminders and coverage suggestions, enhancing the overall customer experience.

Underwriting

Agentic AI analyzes vast amounts of structured and unstructured data in real time to create more accurate and dynamic risk profiles. This enables personalized policy pricing and recommendations, reducing mispriced risk. Like gen AI, agentic AI frees up even more time for underwriters to focus on complex cases and final approvals.

Claims processing

Agentic AI has the potential to automate the entire claims life cycle, from initial assessment and document verification to fraud detection and payout recommendations. It will significantly reduce processing times, increase straight-through processing by improving operational efficiency and minimizing manual touchpoints, and enhance customer satisfaction through faster claims resolution.

Fraud detection and prevention

By continuously analyzing data patterns and monitoring claims submissions in real time for anomalies, agentic AI strengthens fraud detection. It can coordinate with external databases to validate claim authenticity, helping reduce losses from fraudulent activities.

Climate risk and ESG modeling

Agentic AI can analyze real-time environmental data to improve risk modeling related to climate change and other ESG factors.

General operational efficiency

Agentic AI automates routine administrative functions such as policy updates, claims approvals and compliance reporting. It can also optimize employee resources, reduce errors and decrease the cost of rework, leading to operational expenditure savings and improved profitability.

Stronger regulatory compliance

Agentic AI streamlines reporting and supports adherence to rules.

The following chart is the response when asking insurers for their thoughts on the future impact of agentic AI across the insurance value chain:

With new AI tech comes new challenges

Despite the immense potential, several challenges need to be addressed for successful implementation of gen AI and agentic AI in insurance.

Regulatory compliance and explainability

Insurers must ensure transparency in AI decision-making to meet regulatory standards and prevent discriminatory practices. The “black box” nature of many LLM’s makes it difficult to explain their reasoning.

Addressing the challenge:

• Prioritize transparency and explainability in AI models, opting for interpretable AI techniques where possible.
• Implement strong AI governance frameworks that include regular audits for bias and fairness.
• Maintain detailed records of data used, model development and decision-making processes for regulatory review, even if not currently required.
• Engage with regulatory groups to understand evolving requirements for AI in insurance.

Bias and ethical considerations

Any LLM can perpetuate biases present in the data they are trained on, leading to unfair or discriminatory outcomes. Insurers are tasked with ensuring fairness and taking into account ethical considerations.

Addressing the challenge:

• Establish clear ethical guidelines and principles for AI. These principles should reflect the organization’s values and legal requirements.
• Provide extensive training to ensure that gen AI can reason effectively and provide reliable outputs.
• Curate and audit training data for potential biases.
• Employ bias detection and mitigation techniques during model development.
• Continuously monitor AI systems for discriminatory patterns and recalibrate models as needed.

Legacy system integration

Integrating new AI systems with existing legacy infrastructure can be complex and costly.

Addressing the challenge:

• Adopt a phased approach to integration, prioritizing key areas for AI implementation.
• Utilize APIs and middleware to connect legacy systems with new AI platforms.
• Consider cloud-based solutions for greater flexibility and scalability.

Job displacement concerns

Process automation driven by any form of AI may lead to concerns about job displacement for employees.

Addressing the challenge:

• Create a communication plan for all things AI.
• Invest in training and upskilling existing employees.
• Create an AI university and/or AI certification program.
• Partner with external AI vendors and consultants to assist in developing in-house skills or bridging the skills gap.

High initial implementation costs

Investing in the necessary technology, talent and infrastructure for gen AI and agentic AI can involve significant upfront costs.

Addressing the challenge:

• Leverage cloud platforms to access scalable and cost-effective computing resources.
• Optimize AI models for efficient performance.
• Invest in the necessary IT infrastructure and expertise to support AI deployments.

Potential for unintended consequences

As AI agents operate with increasing autonomy, there is a risk of unintended actions or outcomes that could negatively impact customers or the business.

Addressing the challenge:

• AI agents should undergo extensive testing in simulated environments that mimic real-world scenarios, including edge cases and potential failure modes.
• Employ formal methods to verify the correctness and safety properties of AI agent components.
• Design agents with modular components and prioritize interpretability (understanding why an agent made a particular decision) to make it easier to debug, monitor and understand potential  failure points.
• Design agents with only the necessary permissions and access to data and systems required for their specific tasks.