April 2026 Summaries

Reducto's Smart Schema is a tool designed to streamline the creation and improvement of extraction schemas, a critical step in document-heavy workflows. It provides two modes for building schemas: Fast Mode, which prioritizes speed and offers quick baseline results by skipping initial document analysis, and Enhanced Mode, which focuses on accuracy through an interactive feedback loop that uses contextual clues at the document and field levels. Smart Schema aims to overcome common challenges teams face in schema creation, such as defining extraction fields accurately and managing iterations, by offering a more efficient and automated process. Users can further optimize their schemas by addressing flagged ambiguities and seamlessly adjusting their pipelines within Reducto Studio. This innovation promises significant enhancements in extraction accuracy and efficiency, making it particularly beneficial for teams needing to quickly establish or refine document pipelines.

Harvey, an operating system for legal and professional services used by over 1,300 organizations worldwide, enhances practitioners' efficiency by streamlining workflows in areas such as contract analysis and litigation. The company invested in document intelligence by partnering with Reducto to address challenges in processing complex legal documents, which previously led to customer concerns due to incomplete or confusing results. After evaluating various solutions, Reducto emerged as the optimal choice due to its technical superiority, reliability, and ability to handle a wide range of document types, alongside offering verifiable, auditable sources crucial for legal teams. The partnership allowed Harvey to quickly implement a production-ready solution within six weeks, significantly improving document processing capabilities and setting the stage for future exploration of AI-driven enhancements in the legal industry.

Deep Extract is an advanced data extraction tool designed to improve accuracy in processing long and complex documents by using an agent-in-the-loop approach, which autonomously verifies and corrects its outputs. By breaking down documents into manageable parts and iterating the extraction process until it meets a defined quality threshold, Deep Extract achieves field accuracy of 99–100%, surpassing traditional single-pass models and even expert human labelers. This method is particularly effective for documents such as invoices, financial statements, and other extensive records where existing extraction pipelines often fail, due to their inability to capture every necessary detail consistently. With its implementation, users can automate the extraction of thousands of line items across hundreds of pages, ensuring that outputs are not only accurate but also traceable to their original source through granular bounding boxes. Although it takes longer than standard extraction methods, Deep Extract offers a more efficient and scalable solution compared to manual verification, making it ideal for enterprises handling high-stakes documents.