Reducto Lands $24.5 Million After Parsing 250 Million Pages

Reducto was founded in 2023 by Adit Abraham and Raunak Chowdhuri, both former engineers at Google and Nvidia, with the goal of helping enterprises extract structured information from complex unstructured documents. The company combines traditional computer vision methods with new Vision-Language Models (VLMs) to create a parsing platform that achieves higher accuracy and reliability than existing solutions.

Since launching, Reducto has parsed over 250 million pages for thousands of companies, including customers like Scale AI and Vanta, as well as multiple Fortune 500 enterprises. The platform is used to power critical workflows across finance, healthcare, tech, and legal industries, solving bottlenecks in document processing that have historically slowed down broader AI adoption.

Today, Reducto announced a $24.5 million Series A funding round led by Benchmark, with participation from First Round Capital, BoxGroup, and Y Combinator. The round brings Reducto’s total funding to $33 million. Benchmark’s Chetan Puttagunta, known for investments in companies like Elastic and Mulesoft, has joined Reducto’s board.

Reducto’s parsing platform is designed around the principle of multi-pass parsing, where outputs are not only extracted but automatically verified and corrected by Vision-Language Models in subsequent passes. This method mimics a human-in-the-loop review process without requiring manual intervention, which significantly increases accuracy, especially on complex documents involving handwriting, checkboxes, equations, and varied layouts.

To support this, Reducto recently launched an Agentic OCR framework that uses a VLM-based agent to identify and correct mistakes in parsing outputs automatically. Additionally, the company introduced a cost-optimization layer that dynamically adjusts processing based on page complexity, halving costs for simpler documents while preserving high accuracy for complex ones.

“Our approach is built on the idea that parsing is not a one-shot process,” said Adit Abraham, CEO of Reducto. “Each document might require different strategies — sometimes reading context, sometimes validating structure. Our system adapts and improves with every pass, just like a careful human reader would.”

The company’s name, Reducto, was chosen from a list of magic-themed names. Though originally inspired by the Harry Potter spell that shatters objects into smaller pieces, Abraham noted that the name ultimately fit the company’s mission of breaking down complex documents into clean, structured data — albeit unintentionally.

Reducto’s focus on parsing unstructured data aligns with an urgent need in enterprise AI. Many large organizations still rely heavily on paper-based processes particularly in accounting, finance, and healthcare making high-accuracy document ingestion a critical enabler for broader AI deployment.

“For large Fortune 500 companies, most of their accounting and finance processes are still done with paper—paper checks, paper confirmations, invoices,” said Chetan Puttagunta via email. “The workflow around this can’t be digitized, can’t be AI-enabled until the underlying documents are processed accurately for LLMs.”

Benchmark’s investment thesis centered on Reducto’s unique multi-pass accuracy approach and its ability to meet real-world enterprise demands. Puttagunta added, “In the short term, large enterprise customers need systems that can meet them where they are paper and PDFs not hypothetical future data environments.”

Existing customers are already seeing material impacts. Vanta, a leading compliance automation platform, uses Reducto for document parsing within its AI systems. Ignacio Andreu, Senior Manager of Engineering, AI at Vanta, noted, “Based on our evaluations, alternatives like Gemini models, while potentially cheaper, don’t yet match Reducto’s accuracy.”

Similarly, Liz Wessel, partner at First Round Capital, emphasized the breadth of the opportunity. “Industries like finance, healthcare, tech, and legal consistently face challenges converting complex documents into accurate inputs for LLMs,” Wessel said. “This bottleneck limits the ability of AI teams to solve bigger problems. Reducto’s parsing engine is a key enabler.”

Beyond pure parsing, Reducto has been expanding its product suite to address additional stages of the unstructured data pipeline, including:

• Splitting: Dividing multi-document PDFs and complex bundles into distinct, processable units.
• Classification: Intelligent sorting of documents by type and complexity.
• Structured Extraction: High-accuracy extraction of fields, values, and other structured data elements.

The company’s long-term goal is to launch a unified platform that enables enterprises to build complete end-to-end workflows around unstructured data ingestion, transformation, and output—without needing separate tools or pipelines.

“We’re not far off from having intelligent systems—or agents, whatever you want to call it—reasoning on every important process, from doctor’s office intake, to your financial records, insurance claims, all of it,” said Abraham.