Artificial intelligence (AI) is evolving from chatbots into fully autonomous systems capable of reasoning, using tools, and executing multistep workflows. These new AI agents don't just summarize policies and other rote tasks (as powerful as these abilities can be); they can also perform such complex tasks as initiating claims adjustments, negotiating subrogation settlements, or updating risk profiles in real time.
Insurers are already bringing in AI agents to help with
claims. In commercial lines, AI agents are being used to ingest complex
broker PDFs, cross-reference them with appetite guides, and trigger "quote-to-bind"
sequences without human intervention. Hiscox transformed the underwriting of
sabotage and terrorism risks by slashing quote turnaround times from days to
minutes. The promise is clear: a radical reduction in cycle times and a massive leap
forward in operational efficiency.
When AI Agents Hit the Data Wall
Despite the enthusiasm, many insurers are finding that moving agentic AI from a pilot phase to enterprise-wide production is fraught with challenges, usually surrounding the issue of data.
The hallucination trap. AI agents operating on stale or disconnected data "hallucinate" coverage details, leading to wrongful claim denials or incorrect premium quotes.
The silo stumble. An AI agent tasked with assessing a commercial risk may have access to the policy admin system but not the real-time telematics or weather feeds, resulting in incomplete, risky underwriting decisions.
The governance gap. Without a clear lineage, insurers may struggle to explain why an AI agent took a specific action, creating significant "black box" liability and regulatory risk.
The Special Requirements of Agentic AI
The reason for these struggles is that agentic AI has
fundamentally different data requirements from those of traditional business
intelligence or even standard generative AI (GenAI). To function safely and
effectively, an AI agent has the following three basic needs.
Live data. AI agents are action-oriented,
and they live in an operational, real-time world. An AI agent coordinating a
catastrophe response cannot rely on a data warehouse that updates every 24
hours. It needs a live view of the current landscape—streaming weather data,
active policyholder locations, and real-time adjuster availability.
Semantic context. For AI agents to reason effectively, they need to understand the relationship among the different data points. It isn't enough to see a "loss ratio" number; the AI agent needs the semantic context of how that ratio was calculated, which specific perils it includes, and the historical metadata surrounding it.
Built-in governance. Because AI agents act autonomously, security and compliance must be embedded at the data layer. The AI agent must be prevented from seeing data that it is not allowed to see (e.g., personally identifiable information versus anonymized data) and follow guardrails that prevent it from violating state regulations, moving outside internal risk appetites, or otherwise "going rogue."
Bridging the Gap with Logical Data Management
Traditional data management relies on moving data into a central
repository, where it can then be accessed by a person or application. But this is
just not feasible for agentic AI, which needs live data. Logical data management, in
contrast, enables real-time access to data across disparate data sources without
requiring data to first be copied into a central repository. Logical data management
can satisfy all three of agentic AI's core needs, and it can be flexibly deployed
alongside any existing data infrastructure without costly, complex, rip-and-replace
maneuvers.
Here's how logical data management delivers on the three requirements of agentic AI.
Real-time connectivity. Logical data management provides AI agents with access to live data—whether it sits in a 30-year-old COBOL mainframe, a modern cloud lakehouse, or an external Internet of Things feed. With access to live data, AI agents always act on the most current information available.
A universal semantic layer. Logical data management provides a unified business glossary above an insurance company's disparate data sources. When an AI agent asks for "incurred losses," the logical layer translates that request across the various systems and provides the AI agent with a consistent, contextually accurate data product that summarizes the losses, rather than a raw, confusing string of numbers from different data sources.
Centralized policy enforcement. With logical data management, governance isn't an afterthought; it's built into the infrastructure. Insurers can define global security policies at the logical level, and they will be deployed across the entire data estate. If an AI agent attempts to access restricted health data for a life insurance claim, the logical data management layer can automatically mask or block that data based on the AI agent's credentials and the current regulatory environment.
Turning Data into Agentic Action
As the insurance industry moves toward a future defined by autonomous agents, the winners will not be the ones with the largest data lakes, but those with the most agile data delivery systems. By adopting a logical data management approach, insurers can provide their AI agents with the live, contextual, and governed data that they need to perform effectively and safely. In a world of increasing volatility, the ability to turn data into trusted, autonomous action isn't just an IT upgrade; it's a necessity.
Opinions expressed in Expert Commentary articles are those of the author and are not necessarily held by the author's employer or IRMI. Expert Commentary articles and other IRMI Online content do not purport to provide legal, accounting, or other professional advice or opinion. If such advice is needed, consult with your attorney, accountant, or other qualified adviser.
Artificial intelligence (AI) is evolving from chatbots into fully autonomous systems capable of reasoning, using tools, and executing multistep workflows. These new AI agents don't just summarize policies and other rote tasks (as powerful as these abilities can be); they can also perform such complex tasks as initiating claims adjustments, negotiating subrogation settlements, or updating risk profiles in real time.
Insurers are already bringing in AI agents to help with claims. In commercial lines, AI agents are being used to ingest complex broker PDFs, cross-reference them with appetite guides, and trigger "quote-to-bind" sequences without human intervention. Hiscox transformed the underwriting of sabotage and terrorism risks by slashing quote turnaround times from days to minutes. The promise is clear: a radical reduction in cycle times and a massive leap forward in operational efficiency.
When AI Agents Hit the Data Wall
Despite the enthusiasm, many insurers are finding that moving agentic AI from a pilot phase to enterprise-wide production is fraught with challenges, usually surrounding the issue of data.
The Special Requirements of Agentic AI
The reason for these struggles is that agentic AI has fundamentally different data requirements from those of traditional business intelligence or even standard generative AI (GenAI). To function safely and effectively, an AI agent has the following three basic needs.
Bridging the Gap with Logical Data Management
Traditional data management relies on moving data into a central repository, where it can then be accessed by a person or application. But this is just not feasible for agentic AI, which needs live data. Logical data management, in contrast, enables real-time access to data across disparate data sources without requiring data to first be copied into a central repository. Logical data management can satisfy all three of agentic AI's core needs, and it can be flexibly deployed alongside any existing data infrastructure without costly, complex, rip-and-replace maneuvers.
Here's how logical data management delivers on the three requirements of agentic AI.
Turning Data into Agentic Action
As the insurance industry moves toward a future defined by autonomous agents, the winners will not be the ones with the largest data lakes, but those with the most agile data delivery systems. By adopting a logical data management approach, insurers can provide their AI agents with the live, contextual, and governed data that they need to perform effectively and safely. In a world of increasing volatility, the ability to turn data into trusted, autonomous action isn't just an IT upgrade; it's a necessity.
Opinions expressed in Expert Commentary articles are those of the author and are not necessarily held by the author's employer or IRMI. Expert Commentary articles and other IRMI Online content do not purport to provide legal, accounting, or other professional advice or opinion. If such advice is needed, consult with your attorney, accountant, or other qualified adviser.