Loss Runs, SOVs, and ACORD Forms: How AI Extraction Actually Works

In commercial property and casualty insurance, the speed at which you can quote is directly tied to how fast you can process the submission package. Yet the documents that make up these packages are notoriously difficult to standardize.

A single submission might contain an ACORD 125, a Schedule of Values (SOV) covering 300 locations, and three years of loss runs from two prior carriers. While basic Optical Character Recognition (OCR) can read the text on these pages, it cannot understand the context. True intelligent document processing for insurance requires AI that understands what the data actually means.

The Challenge of SOV Extraction

Schedules of Value arrive in carrier-specific Excel formats, broker-generated spreadsheets, and hand-typed tables. A 300-location SOV may contain 40 columns covering construction type, occupancy, year built, and replacement cost value.

The primary hurdle in SOV extraction is that there is no universal standard. One broker’s “Total Insured Value” column is another’s “RCV.” Traditional rules-based extraction tools break when they encounter a column header they haven’t been programmed to recognize.

AI-driven schedule of values extraction solves this through semantic understanding. The AI recognizes the context of the data within the column, allowing it to accurately map the information to your internal schema regardless of what the broker named the header.

Mastering Loss Run Automation

Loss runs present a different, often more complex, problem. Every prior carrier formats its loss run PDFs differently. Reserve amounts may appear at the claim level, the policy level, or not at all. Furthermore, open claims require different treatment from closed claims for actuarial purposes.

Effective loss run automation must do more than lift numbers off a page. It must identify claim status, extract reserves accurately, calculate incurred losses over a specific period, and flag catastrophe years. Loss run data extraction tools must also reconcile data across multipl documents, ensuring that the named insured on the loss run matches the entity on the ACORD form.

Moving Beyond Basic ACORD Form Extraction

While ACORD forms are technically standardized, they are frequently submitted as scanned PDFs, skewed images, or hand-filled documents. Basic ACORD form extraction tools often struggle with checkboxes, handwritten notes, or supplemental information squeezed into the margins.

Modern AI platforms handle these variations using computer vision alongside natural language processing. And when a system is designed with evidence traceability, every extracted field links directly back to its source document and page. If an underwriter questions a data point, they can click through to the source in one step — eliminating the “black box” problem of early AI tools.

Why Your Extraction Tool Must Learn

The variety of commercial insurance documents means no system will be 100% perfect on day one. The critical differentiator is what happens when the system makes an error.

Cazimir is built on the principle that extraction tools must learn from human corrections. When your team reviews an extracted SOV and corrects a mislabeled occupancy code, Cazimir records that correction as a training signal. The platform learns the specific formatting patterns of that broker and adapts to the variety of specialty submissions. By compounding this intelligence, Cazimir ensures your team never has to correct the same extraction error twice.