How Docusign is Bringing Contract Table Extraction to Production with NVIDIA Nemotron Parse

Organizations lose significant value every year to the friction, delays and missed obligations that come from treating agreements as static documents rather than live sources of business data. Much of that trapped value sits in tables: the pricing schedules, SLA obligations and contractor rate cards that define obligations but are routinely the hardest part of a contract to extract accurately.

Docusign handles millions of transactions a day for more than 1.8 million customers. At that scale, table extraction is foundational to delivering the agreement intelligence our customers need. Earlier this year, we began evaluating the NVIDIA Nemotron Parse open model opens in a new tab to address this exact challenge. 

Today, Docusign is accepting beta customers for table extractions in Navigator, powered by Nemotron Parse, with general availability coming soon.

Why contract tables are harder to extract than they look

Contracts routinely contain merged cells, multi-page structures, mixed formatting and nested layouts that general-purpose Vision Language Model (VLM) tools and broad AI models weren't designed to handle. The result is inaccurate extractions that require manual correction, slowing down the very workflows they're meant to accelerate.

Docusign customers experience this friction in their daily work. When a system goes down, operations teams need to know immediately which SLA notification requirements apply and to whom. When business stakeholders ask legal what hourly rate was agreed to in the last contractor engagement, the answer is often buried in a rate card table. When procurement manages a vendor renewal, pricing structures scattered across exhibits require significant manual review to piece together.

Getting tables right is foundational to agreement intelligence.

"Agreements are among the most information-dense documents in any organization, and the data inside them has historically been the hardest to reach. Our adoption of the open NVIDIA Nemotron Parse model is a meaningful step toward changing that,  giving our customers the ability to surface what's in their contracts and act on it faster." – Sagnik Nandy, CTO, Docusign.

With NVIDIA Nemotron Parse integrated into Docusign's document understanding layer:

• SLA obligation tables that once required manual review are now automatically extracted and structured

• Rate cards buried in contractor agreements are surfaced on demand

• Pricing schedules across procurement contracts become queryable data rather than static exhibits.

The pipeline: From layout detection to structured data

Docusign's document understanding pipeline processes agreements at scale, handling layout detection and Optical Character Recognition (OCR) across millions of documents. Adding reliable table extraction required a model layer that could handle the structural complexity those earlier stages couldn't fully resolve.

At the core of the integration is NVIDIA Nemotron Parse opens in a new tab, a compact vision-language model purpose-built for document understanding. Nemotron Parse model combines layout detection, OCR and document semantics to interpret and reconstruct complex tables accurately. It handles the structural variability that trips up general-purpose approaches, including the layout complexity common across contract types and industries.

Nemotron Parse is served via vLLM and integrated into Docusign's layout and OCR pipeline. Sensitive agreement data stays within Docusign's environment — a hard requirement when documents contain confidential business terms — while giving us the flexibility to run and optimize the model for our specific use case.

PDF document processing flow for table extractions inside Navigator.

PDF document processing flow for table extractions inside Navigator.

To validate the integration, we tested against real enterprise contracts rather than synthetic benchmarks, which don't capture the formatting variations, inconsistent table structures and mixed-language content that enterprise contracts actually contain. 

This gave NVIDIA the confidence to deploy Docusign IAM to manage its own agreements. 

"Open models are helping enterprises deploy AI into production with greater flexibility, transparency and control over their data and workflows. With NVIDIA Nemotron Parse, Docusign is enabling customers, including NVIDIA, to unlock and act on critical data with greater accuracy and speed at enterprise scale." — Kari Briski, VP of Generative AI Software, NVIDIA.

What's next

We're continuing to improve accuracy on more complex and varied table structures, and exploring deeper integrations with agentic workflows through the NVIDIA Agent Toolkit opens in a new tab. A public API for direct integration with downstream systems is coming soon.

Learn more about how Docusign Navigator can centrally store, manage, and analyze even the most complex agreements with AI. Download the latest Nemotron Parse model from NVIDIA here opens in a new tab.