NSF has introduced an Azure AI–based, agentic tool to rethink how it audits new medications for safety, efficacy, and regulatory compliance.

Traditionally, each audit required **tens of thousands of documents** to be collected, organized, checked for completeness, version-controlled, and summarized. This work was largely manual and could take **four to six weeks** per audit.

With the new solution, NSF uses:

• Azure Document Intelligence to scan and verify that all required documents are present.
• Azure OpenAI models and Model Context Protocol (MCP) Servers to classify documents into the correct, internationally regulated folder structures.
• Azure Blob Storage, Azure Cosmos DB, and the Azure Python SDK to automate version tracking and manage structured data.
• Azure OpenAI models to generate first-draft summaries that NSF experts then review and refine.

This combination has **cut audit turnaround time by about half or more**. Audits that used to take four to six weeks can now be completed in about **two weeks** while maintaining rigorous compliance standards.

For patients and healthcare providers, this means **potentially life-saving or life-improving medications can reach the market faster**, giving hospitals and clinicians more treatment options in a shorter timeframe.

NSF reports that its Azure AI–based auditing tool delivers **near-perfect accuracy** in the summaries and document handling it supports. According to NSF’s innovation leadership, the tool consistently achieves **“100% truth value”** in its outputs, with staff typically needing to make only cosmetic or style-related edits.

This level of accuracy matters because each audit involves **thousands to tens of thousands of documents** tied to strict, country-specific and international regulations. By automating tasks such as:

• Document completeness checks
• Regulatory classification and foldering
• Version control
• Drafting cross-document summaries

the solution helps **minimize the risk of human error** that can occur when teams manually process large volumes of complex information.

NSF’s experts remain in the loop. Scientists and regulatory specialists review AI-generated summaries and conclusions, focusing their time on higher-value work such as regulatory strategy and complex judgment calls, rather than on repetitive document handling. This approach both **reduces operational risk** and supports consistent, compliant audit outcomes.

NSF operates in a highly regulated environment with sensitive medical and intellectual property data, so security and compliance are central to its Azure AI approach.

Key elements of NSF’s security model include:

• Private data tenancy: Data is maintained in a private tenant within SharePoint and then moved into Azure Blob Storage for processing.
• Role-based access control: NSF uses Microsoft Entra ID and Azure role-based access control (RBAC) to ensure that only authorized users can access specific data sets and tools.
• Closed, cloud-based workflow: The entire auditing workflow runs inside the Azure Cloud. All application endpoints use private connections, reducing exposure to external networks.
• Controlled AI tool access: Azure Model Context Protocol (MCP) Servers manage how language models interact with external tools and data, allowing NSF to decide how networked or isolated each AI solution should be.

Because NSF is a “Microsoft shop,” it also benefits from the integrated security posture of the Microsoft ecosystem, including the closed environment of Microsoft 365 Copilot for broader organizational use.

This setup allows NSF to **scale AI use across more audit types**—such as medical devices, dietary supplements, and water safety—while maintaining strong controls over data privacy, access, and regulatory compliance.