Chaz Perera, the co-founder and Chief Executive Officer (CEO) of Roots Automation, explains how intelligent automation can address the unstructured data challenges insurance companies face. This article originally appeared in Insight Jam, an enterprise IT community that enables human conversation on AI.

An IDC survey of mid-market and enterprise businesses found that 40 percent of companies use “mostly manual” document processing methods, i.e., people reviewing and extracting information of “potential value.” However, this doesn’t scale well for any business, particularly for insurance organizations.

While some data used in insurance is structured and organized into standardized formats, e.g., ACORD forms, most data leveraged across the industry is unstructured–free form, highly variable, and trapped within a profusion of paper, email, and attachments. In fact, Accenture found that unstructured natural-language data constitutes around 80 percent of all data across insurance.

What is the impact of this unstructured data? Human understanding and power are among the most common methods for addressing it, which is why the information inside this data is often incompletely analyzed, misread, misclassified, or unused. For example, according to AM Best, insurers’ inability to effectively convert unstructured data into structured data for use in underwriting and claims decisions was one of the primary reasons over $21 billion was lost in Property and Casualty underwriting in 2023.

Additionally, Accenture says the insurance industry’s reliance on human experts to perform non-essential and administrative activities like manually entering and reworking information from insurance documents will contribute to an estimated $85-$160 billion in value lost to inefficiency in underwriting alone by 2027.

There is an opportunity to prevent these losses by maximizing efficiency and effectiveness across underwriting and claims operations. To do this, though, insurers must be able to easily understand and gain value from the unstructured data they hold.

Attempts to Solve the Unstructured Data Problem with Technology

For more than a decade, insurers have attempted to solve the unstructured data issue using legacy automation technology, like Robotic Process Automation (RPA), Intelligent Document Processing (IDP), and Intelligent Process Automation (IPA). However, most insurers have seen poor results because RPA tools are brittle, take time to maintain, and heavily rely on structured data to operate successfully.

Address the Challenges and Drive Business Impact with Intelligent Automation

Generative AI has demonstrated strong capability in performing natural language tasks around unstructured data. These models understand context, extract relevant information, and generate human-like responses to user input. They are ideal for interacting with human users to automate insurance workflows and reduce errors.

According to Accenture, Generative AI will conceivably automate up to 62 percent of insurance underwriting and claims processes—an essential step in increasing the value created by human experts by empowering them to concentrate exclusively on higher-value tasks rather than incorporating and attempting to understand unstructured data for decision-making. Additionally, Swiss Re notes that by using AI to extract insights from unstructured data sources, insurers can achieve a 12-25 percent improvement in their claim loss adjustment compared with companies that don’t leverage these technologies.

However, public horizontal Generative AI (GenAI) solutions (e.g., those offered by Google, OpenAI, Meta, and others) aren’t purpose-built for insurance. They lack the necessary training and refinement to handle insurance data and interface with insurers’ systems in a contextually accurate manner. Solutions built around public AI are not practical for complex, regulated, and crucially important decision-making environments such as insurance.

The resources needed to build an in-house GenAI solution are beyond the means of all but a few multinational insurance behemoths. An Oliver Wyman/Celent poll indicates that more than 20 percent of US insurers are trying to develop internal vertical Generative AI solutions. However, less than half of these companies have effectively advanced their projects from the proof-of-concept stage to production. The reliance on internal specialized knowledge, infrastructure, and data resources to develop and maintain models with insurance domain expertise is the likely cause of this poor performance.

GenAI Specifically Built for Insurance

GenAI solutions built explicitly for insurance contain industry-specific capabilities to address the needs of underwriting and claims management. For example, with an insurance-focused GenAI solution, underwriting teams can automate document indexing, analysis, and clearance and extract 90 percent more information from unstructured data, driving 10 percent underwriting cost reductions to boost profitability and satisfy customers with faster, more accurate quotes. With straight-through processing of up to 80 percent of forms, underwriters have additional time to serve agents and brokers and concentrate on product development and other revenue-generating opportunities.

On the claims side, GenAI-powered solutions deliver impactful results throughout the entire process, including decreasing document handling times from 3-5 days to minutes—at 95%+ accuracy—and reducing manual claims set up by 90 percent. Results like these drive substantial reductions in claims overpayment while increasing operational capacity to address spikes in demand from natural disasters.

Every piece of information hidden inside unstructured data is an opportunity to enhance insurers’ underwriting and claims operational efficiency. Identifying and leveraging that data requires intelligent automation built with the insurance industry in mind. With GenAI specifically developed for insurance, organizations can reap the rewards of more effective operations and happier customers.