At the core of transformation is a practical objective: removing friction in how agents and clients interact with flood products. Marissa Skinner (pictured), managing director of Poulton Associates, a division of Wright National Flood Insurance Services, has been advancing a focused transformation agenda across underwriting, data, and client delivery. Poulton Associates, headquartered in Salt Lake City, Utah, runs the Natural Catastrophe Insurance Program (NCIP), as well as other natural catastrophe insurance offerings. It owns and operates the CATcoverage.com web platform and has been a leader in provisioning risk services since 1989.
Skinner’s work sits between operational efficiency and risk precision, where flood insurance remains one of the most complex and evolving lines in the industry. The effort reflects a wider push to integrate artificial intelligence while maintaining underwriting discipline. It also underscores the tension between innovation speed and data reliability in catastrophe-exposed portfolios.
Skinner points to a persistent challenge in the MGA space, where many agents write only a handful of flood policies annually. “Learning to navigate a website, understanding a workflow, and retaining that knowledge can be difficult,” she said. “Agents are calling in, needing handholding, regardless of how intuitive the user interface is.”
To address this, Poulton and Wright Flood have been experimenting with natural language tools that can streamline transactions. The goal is to enable agents to process policy changes or obtain quotes through conversational interfaces rather than traditional portals. This approach reflects a broader industry trend toward embedding AI in front-end interactions, particularly in low-frequency, high-friction product lines. At the same time, the company is exploring AI-driven claims processing that preserves human oversight.
Skinner emphasized that these initiatives are designed to balance efficiency with service quality. “The goal is to ensure clients are serviced the way they need to be - with a person available when necessary - while also processing claims with high efficiency,” she said. This dual focus highlights a recurring theme in insurance transformation: automation must enhance, not replace, the human element.
Skinner described a leadership-led approach, with a dedicated focus on responsible AI adoption embedded at the executive level. “It is absolutely a top-down initiative,” she said, noting that leadership alignment has been critical in setting priorities and governance.
However, the technical challenges remain significant, particularly around data quality and model reliability. Skinner identified data drift and model accuracy as central risks in deploying AI at scale. “We’re all aware of the hallucinogenic effects within AI and making sure that the data we’re acting on is valid and accurate is critical,” she said.
This concern has led to a strong emphasis on continuous model tuning and validation. The principle of “good data in, good data out” has become a guiding standard, particularly in underwriting where small errors can materially affect risk selection. For flood insurance, where exposure is highly sensitive to localized variables, maintaining data integrity is not simply a technical issue but a core underwriting requirement.
The company’s approach to external partnerships also reflects this cautious posture. Historically, Skinner noted, the organization built much of its technology in-house and relied heavily on internal data capabilities. Today, while the market offers a wide array of third-party solutions, Wright Flood remains highly selective in choosing partners. “We don’t want to over-invest in a solution that ultimately delivers nothing,” she said. “This emphasis on rapid testing and iteration suggests a disciplined innovation model, where experimentation is encouraged but tightly controlled.”
Measuring the effectiveness of these innovations requires close alignment between analytics and underwriting expertise. Skinner described a collaborative approach involving analysts, data scientists, and actuarial leadership, with continuous monitoring of model outputs. The objective is not only to assess performance but also to ensure that results align with underwriting intuition.
“The challenge is ensuring the science of underwriting, and the art of underwriting are in alignment,” she said. This balance is particularly important in flood insurance, where models continue to evolve and no single approach has emerged as definitive.
Skinner pointed out that transformation in flood and catastrophe risk is inherently more complex than in standard lines. Over the past decade, increased interest in the sector has led to the development of multiple modeling approaches. Yet, she noted, “I don’t think we’ve arrived at a definitive best solution just yet.”
As a result, the company has adopted a “best-of-breed” philosophy, combining multiple data sources and model outputs to form a more comprehensive view of risk. This approach reflects the fragmented nature of flood modeling, where different tools may perform better in different geographies or scenarios. It also reinforces the need for underwriting judgment in interpreting model results.
Geographic variability remains a key challenge. Skinner highlighted how risk scores can behave differently across regions, with east coast exposures showing higher frequency and west coast risks skewing toward severity. These distinctions complicate the application of standardized models and require a more nuanced interpretation of data.
Despite advances in modeling, Skinner also pointed to the continued relevance of traditional data points. Variables such as first-floor elevation, which may be underweighted in some models, still provide valuable insights into underwriting decisions.
This highlights a broader theme: innovation does not eliminate the need for foundational risk indicators.
