Docusign Swapped Large Models for Specialized AI

The company reduced contract processing costs by 90 percent by moving from general-purpose to optimized models.

Updated on Sept. 21, 2026 in Artificial Intelligence

Isometric editorial illustration showing stylized, nested geometric volumes that represent modular digital processing components. — Docusign has transitioned its document processing engine to smaller, task-specific AI models, significantly increasing throughput while reducing computational costs by 50-fold. AI Illustration. Upload story photo >

Live Poll

Do you believe companies are successfully turning their current investments in AI into profitable results?

Docusign has transitioned its document processing engine from large general-purpose AI models to smaller, specialized alternatives. This architecture shift has enabled an eightfold increase in throughput while cutting per-document processing costs by 50-fold.

Why it matters

The move addresses a growing industry tension where rising production volumes have made massive models economically unsustainable. By optimizing for specific tasks, firms can maintain performance while managing the 320 percent increase in enterprise AI bills observed since 2022.

Docusign reduced per-document AI costs by 50-fold compared to its previous system while maintaining accuracy within 2 percentage points of larger models. The company now processes over 1 million agreements daily by extracting 50 facts per document.

The players

Docusign

A provider of cloud-based e-signature and agreement management software that is integrating agentic AI into its platform.

Microsoft

A global technology company providing cloud infrastructure, foundational model development, and the Foundry training platform.

The details

Docusign built this system by training smaller, task-specific models using the Microsoft Foundry platform. To improve efficiency, the company introduced a filtering layer that identifies and segments relevant contract passages before routing them to the model, eliminating the need for a previous cleanup step meant for model output correction.

2022: Per-token AI prices began a downward trend.
May 2026: Docusign unveiled its Iris AI assistant and agent tools.
September 15, 2026: Microsoft released the case study detailing this infrastructure shift.

The Tech Race

As enterprise AI spending surges, companies are moving away from massive, generalized models toward smaller, specialized architectures to achieve cost-efficiency. This transition seeks to solve the current paradox where 81 to 95 percent of large enterprises use AI, but few have achieved full investment payback.

Users will see faster contract processing speeds without a decrease in data extraction accuracy. This transition allows Docusign to scale its AI agent product line while keeping costs lower than would be possible with general-purpose, high-latency models.

The takeaway

The transition to specialized models demonstrates that infrastructure efficiency is now as critical as model intelligence for enterprise adoption. Watch for the next quarterly performance reports to see if these cost savings drive higher adoption rates for Docusign's AI agent services.

What happens next

Market analysts are tracking the 80 percent of large U.S. enterprises that plan to increase their AI budgets next year, a trend expected to push more companies toward these optimized, smaller-model architectures.