Docusign Iris Connects Contract AI to Company Policies and Approval Workflows

For enterprise contract agents, the hard part isn’t summarizing a document. It’s determining which terms govern a business decision when agreements have amendments, suppliers use different names across files, and internal policies dictate who can approve an exception. Docusign is building its agreement AI around that wider context.
The effort sits within Intelligent Agreement Management (IAM), Docusign’s platform for creating, signing and managing agreements. Introduced in April 2024 and made available to U.S. customers the following May, IAM extends Docusign’s work beyond electronic signatures. Its AI engine, Docusign Iris, is designed to connect contract information with the workflows and decisions surrounding it.

“General LLMs are great. It’s just that you need more around it to really unleash its full power,” Docusign CTO Sagnik Nandy tells The New Stack. His point is practical: a model may read a clause, but an enterprise system must also know how that clause relates to other documents and company rules.
“Context is one of the things that really differentiates a generic solution from a solution that speaks your language.”
Four layers behind a contract question
Nandy describes four forms of context that matter to agreement agents: information within one agreement, relationships across agreements, organizational policies, and details about the user.
Within a single contract, a system may need to connect a defined term on one page to a reference much later in the document. That task is relatively contained. The harder case arrives when a contract has been amended, repriced or otherwise changed over time.
“Imagine you have a single agreement that’s gone through two or three amendments, or a price change. To an LLM, they look like three or four independent PDF files,” Nandy says.
For the business, those files represent one continuing relationship. A useful answer to a pricing question must reflect the terms currently in effect, not return every price mentioned across the original agreement and its amendments. That means identifying which documents belong together and understanding how later changes affect earlier clauses. A search result by itself can’t establish which version controls.
Company policy adds another layer. Legal teams may maintain playbooks that set acceptable terms, while procurement and sales have their own approval responsibilities. The agent needs those rules to distinguish a routine case from one that calls for review. User context, such as language preferences, prior questions and feedback, can shape how the system responds as well.
General-purpose models can often help with information inside one document. Nandy says the other forms of context require more specialized systems and agreement-specific logic.
Iris combines search, extraction and agent workflows
Iris is built from multiple capabilities rather than depending on a language model to handle every step. Nandy describes a system that brings together information extraction, semantic search, agreement data and an agent layer. Docusign also offers AI contracting capabilities for finding and working with contract information.
That combination matters because locating a renewal date and deciding whether the renewal needs attention are different jobs. Extraction can identify dates and terms. Search can find relevant agreements. Agreement-specific logic can connect documents, amendments and counterparties. An agent can then use that assembled information to support a task.
Take a procurement manager asking which software agreements worth more than $1 million are due to renew in the next three months. Nandy says that answering such a question has often meant asking several people to gather records, merge spreadsheets and resolve supplier names that vary between contracts. With agreement information connected, Iris can help identify relevant contracts and obligations, then check them against organizational playbooks to surface potential issues before renewal.
A supplier payment question follows a similar chain. The system must identify the supplier as an organization, find its agreements, connect amendments and work out how payment terms changed. That’s why a seemingly simple question can depend on a repository, rich search and a way to combine related agreements into a coherent record.
“While it seems like a straightforward question, to make it truly happen, you need rich search functionality. You need a repository,” Nandy says. He estimates that answering a question of this kind may involve five or six tools working with Docusign’s contract-specific data.
“The beauty is to hide the complexity and make it look easy to the end user. But there’s a lot going on behind it.”
The technical implication is clear: agent quality depends on the data model and workflow around the model. If the system can’t resolve which documents form one agreement, or which clause remains active, fluent answers won’t make its conclusions dependable.
Automation needs approval boundaries
More context can help an agent make better recommendations, but it doesn’t mean every decision should be automated. Docusign pairs agent capabilities with business rules and approval flows so organizations can set boundaries for each process.
A company might allow routine renewals under a spending threshold to follow a standard path. A more expensive agreement, an unusual term or a policy exception could be sent to a person. The agent can collect relevant information and apply a playbook, while a configured workflow determines the next step and routes the case according to the company’s rules.
This division matters in contract operations. An agent that misses an earlier obligation or treats a sensitive term as routine could send a decision down the wrong path. A workflow can keep approvals explicit, with people reviewing the cases the organization has designated for human judgment.
Docusign’s ready-to-deploy agents can run through the Iris AI assistant’s chat experience or appear as steps in Workflow Builder. Teams can also build custom contract agents in Docusign Agent Studio. As described in Docusign’s developer guide to Iris agent capabilities, Workflow Builder can combine agent reasoning with rule-based routing and approvals: the agent supplies a result, and the workflow uses it to choose what happens next.
Availability depends on plan, region and language. Docusign lists the assistant and Agent Studio agents for eligible IAM plans in North America data centers, in English.
Agreement data can follow the full lifecycle
The broader product ambition is to support work before and after signing, as well as the signature itself. Nandy divides the agreement lifecycle into creating an agreement, committing to it and managing it afterward.
At creation, a system could help identify the right template and the terms a company usually accepts. During negotiation, it could compare proposed changes with prior agreements and help determine who must approve them. After signing, it could track obligations and flag approaching renewals. Each stage depends on connecting contract language to the decisions and processes around it.
Docusign points to customer results from agreement information entering business workflows, though those outcomes don’t establish that autonomous agents caused the improvements. In its 2026 Customer Awards, the company says Experian reduced contract cycle times from about 10 days to hours and made contract data easier to access and understand by connecting Docusign with its configure-price-quote (CPQ) solution.
For teams building enterprise contract agents, Iris illustrates the work behind a useful answer: extract dependable data, link amendments to the agreements they change, apply organizational rules and make approval boundaries part of the workflow. Reading contract text is a starting point. The harder engineering task is turning it into business context that people can act on.
More Docusign in The New Stack
- Building the agentic agreement enterprise: How developers are unlocking agentic experiences with Docusign’s MCP server and platform (May 21, 2026)
- Building integrations with AI assistance that go beyond vibes (July 31, 2025)
- Auto-read data into agreement workflows with Docusign extension apps (June 25, 2025)
- How to use Docusign APIs to embed custom document signing in an app (May 28, 2025)
- Build API-driven custom agreement workflows with Docusign Maestro (May 15, 2025)
- Inside Docusign’s AI-powered agreement management platform (January 6, 2025)
- Docusign expands its developer community with new platform (November 20, 2024)
The post The context problem: How Docusign built agents that understand the business behind contracts appeared first on The New Stack.