Docusign processes over 1 million contracts daily and has optimized its infrastructure by replacing large general-purpose AI models with smaller, task-specific ones. This transition, detailed in a Microsoft case study, reduced AI processing costs by 90% while increasing throughput by up to eight times.
The company now uses smaller models for routine data extraction, such as identifying contract values and renewal dates. Frontier models are reserved only for complex tasks requiring reasoning. Additionally, Docusign filters documents to send only relevant passages to the models, rather than full contracts, which further improves efficiency.
The new approach has led to a 50-fold reduction in per-document costs compared to the previous system. These savings support the company's Iris assistant and AI agents, which automate workflows in legal, HR, and procurement departments. The extraction models pull over 50 specific facts from each agreement.
Large enterprises often manage backlogs of tens of thousands of legacy contracts across multiple systems. Firms like SEI use the Docusign system to integrate legal obligations directly into platforms like Workday and HubSpot. This ensures that new work orders are automatically populated with relevant client terms.
