Data Architecture Is Key for 'Agentic AI'
Upcoming TDWI webinars are highlighting the need for modern data architectures, like the lakehouse, to power the next wave of "agentic AI" — AI capable of autonomous action. The discussion emphasizes that robust, unsiloed data foundations are a prerequisite for enterprises, including insurers, looking to deploy AI for operational decision-making.
The leap from current AI to "agentic AI" is defined by autonomy; instead of just analyzing or suggesting, these systems are designed to take ownership of entire workflows like claims management or policy issuance, operating within predefined regulatory guardrails. This shift moves insurers from a reactive "detect and repair" model to a proactive "predict and prevent" stance on risk. For underwriting departments, agentic AI acts as an intelligent coordinator, extracting data from broker submissions and ACORD forms, verifying it against third-party datasets, and triaging applications in real-time. This allows straightforward submissions to be auto-quoted while escalating more complex cases to human underwriters, transforming their role from manual data processing to focusing on specialized risk assessment. In claims processing, agentic AI can automate the First Notice of Loss (FNOL) across chat, web, and voice channels, immediately initiating the claims process. By using computer vision to assess damage from photos and flagging suspicious patterns for fraud detection, these systems can reduce claims resolution times from days to hours, leading to faster payouts. The adoption of agentic AI in the insurance sector is projected to grow from 14% to 70% by 2028. Companies successfully scaling these systems are seeing tangible results, including loss ratio improvements of 3-5% and a 60-99% reduction in quote-to-bind times for commercial property and casualty insurers. This advanced automation is heavily dependent on modern data architectures like the lakehouse, which unifies structured and unstructured data from various sources. Legacy systems, where data is often fragmented in silos, create a significant barrier to training reliable AI models and require substantial pre-cleaning efforts. However, implementation faces challenges beyond technology. Organizational readiness is a primary success factor, requiring an estimated 70-80% of digital talent to be in-house to manage and adapt these systems. Change management, including reskilling employees and fostering a culture that embraces AI-assisted workflows, is considered a major hurdle. Regulatory compliance is another critical aspect, with frameworks like the NAIC Model Bulletin requiring insurers to establish board oversight, fairness testing, and transparent accountability for their AI systems. The "black box" nature of some AI, where the decision-making process is not easily understood, is a key concern for both insurers and regulators.
