More than 1,500 field engineers at FM now access risk guidance in seconds, not hours. The firm, in partnership with Spyglass MTG, has deployed a Microsoft Azure-based AI engineering knowledge platform that puts governed retrieval-augmented generation directly into the hands of on-site personnel. Microsoft profiled the implementation in early June 2026, spotlighting how Azure AI Search transforms fragmented documentation into an authoritative decision engine.

Every field engineer knows the frustration. A critical safety question arises on site, and the answer sits buried in a 600-page PDF, an outdated SharePoint folder, or the memory of a senior colleague who retired last month. For FM—a company that manages complex facilities and infrastructure risk across the globe—that lag time doesn't just waste productivity; it introduces genuine hazard. The new platform curbs that risk by indexing millions of engineering documents and making them queryable through a natural language interface, with guardrails that prevent hallucination or unauthorized data exposure.

The Field Engineer's Knowledge Gap

Risk assessment in heavy industry isn't guesswork. It depends on precise interpretation of engineering standards, equipment specifications, maintenance logs, and compliance mandates. FM's workforce operates in sectors where errors can lead to catastrophic failure, environmental damage, or loss of life. Yet before the AI platform, engineers often spent up to 40% of their shift hunting down the right document. Even when they found it, they had to vet the version, cross-reference related clauses, and pray nothing had been superseded.

That workflow had three big cracks: time, accuracy, and governance. Speed pressures encouraged engineers to rely on memory or informal advice rather than authoritative sources. Accuracy suffered when answers came from outdated or incomplete records. And governance—the ability to audit who accessed what, ensure data residency, and restrict sensitive content—was nearly impossible to enforce across a sprawling document ecosystem.

Azure AI Search as the Retrieval Brain

Microsoft positions Azure AI Search as a cloud-native retrieval engine that blends traditional search with semantic ranking and vector embeddings. Unlike a standard keyword search, it understands intent. For FM, that means an engineer can type, “What is the maximum allowable pressure for a Type III vessel in a seismic zone?” and the platform retrieves not just the right section of the ASME code but also the relevant FM internal policy, the last inspection record, and a safety advisory issued six months earlier.

The retrieval pipeline relies on pre-built or custom skills that chunk documents, enrich them with metadata, and store vector representations in a search index. When a query comes in, the system performs a hybrid search—combining dense vector similarity with traditional BM25 relevance scoring—to surface the most pertinent passages. Those passages then form the grounding context for a large language model (LLM).

Governed RAG: Putting Guardrails on Generative AI

Retrieval-augmented generation (RAG) isn't new. But governed RAG is what sets this solution apart. In a typical RAG setup, the LLM uses retrieved content to formulate an answer, but there's little control over how it uses that content or what it outputs. For risk-related queries, that's unacceptable. FM needed every answer to be traceable back to an approved source, with zero tolerance for invented data.

The governed layer enforces three constraints. First, answers are generated exclusively from the retrieved, authorized corpus. If the confidence score falls below a threshold, the system refuses to answer rather than guess. Second, every response includes inline citations that link back to the exact document and passage, so an engineer can click through and verify. Third, role-based access controls (RBAC) ensure that the search index respects document-level permissions. A junior engineer won't see a confidential failure analysis report simply because the LLM can retrieve it; the retrieval step itself applies security trimming so only entitled users get context.

This approach leans heavily on Azure AI Search's integration with Microsoft Entra ID (formerly Azure Active Directory) and Azure Policy. FM can define governance rules that restrict which indexes a user can query, enforce encryption in transit and at rest, and log every interaction for compliance audits. The platform also integrates with Microsoft Purview for data classification, automatically labeling sensitive content and preventing it from leaking into the wrong RAG response.

Why Spyglass MTG and Microsoft

Spyglass MTG, a consultancy specializing in Microsoft cloud solutions for engineering and manufacturing, built the customized knowledge platform on top of FM's existing Azure footprint. They designed the indexing pipeline to handle a heterogenous mix of file formats—PDF, Word, CAD annotations, SharePoint lists—and orchestrated the AI enrichment with Azure Functions and Logic Apps. The result is a fully serverless architecture that scales automatically as FM's document library grows.

Microsoft's profile of the deployment, published in early June 2026, highlights the partnership as a blueprint for industrial enterprises that have been hesitant to adopt LLMs due to precision and security concerns. By using Azure AI Search's built-in security model and limiting the generative component to only synthesize approved content, FM sidesteps the validity nightmares that plagued early chatbots.

Real-World Impact on Engineer Productivity

FM reports that the platform has slashed time-to-answer for complex risk queries from an average of 45 minutes to under 90 seconds. Field engineers now perform ad hoc risk assessments on tablets while standing next to a boiler or a turbine, rather than calling back to the office and waiting for a specialist to pore over documents. In pilot tests, 92% of generated answers were rated as “fully accurate” by senior engineers, with the remaining 8% flagged for minor omissions rather than incorrect facts—a stark contrast to the 30% error rate seen in ungoverned LLM deployments.

More importantly, the platform changes how risk knowledge accumulates. Every query and its resulting answer get logged along with user feedback. FM's knowledge managers can spot which documents are most frequently accessed, which generate the most confusion, and which are ripe for revision. The system becomes a living tool, not a static archive.

Broader Enterprise AI Signals

For Windows and Microsoft cloud enthusiasts, FM's story is more than a case study. It demonstrates how the Azure ecosystem brings together data, AI, and governance in a way that's difficult to replicate with a patchwork of point solutions. Azure AI Search acts as the retrieval backbone, Azure OpenAI Service provides the LLM, and Microsoft Entra ID plus Purview handle identity and compliance. The whole stack runs on Windows Server and Linux workloads, but the end-user experience lives on any modern browser or Windows app via Power Apps or custom front-ends.

This architecture also underscores a key trend: the line between enterprise search and generative AI is vanishing. Organizations that already use Azure Cognitive Search (the predecessor to Azure AI Search) can upgrade to vector search and RAG capabilities without rebuilding their entire information architecture. FM's path from a legacy SharePoint search to an AI-powered risk advisor took months, not years, because the underlying infrastructure was already in place.

While the technology is impressive, FM's real innovation is governance-first design. Many AI projects stall because security teams are brought in after the fact. Here, legal and compliance stakeholders shaped the RAG guardrails from day one. They defined acceptable answer formats, mandated that the LLM could not use public web data, and required an ephemeral conversation model where user queries aren't stored for future training. That model aligns with Microsoft's responsible AI principles and avoids cross-tenant data contamination.

FM also opted for a dedicated Azure AI Search instance rather than a shared resource, giving them full network isolation and private endpoints. This is a critical detail for industries that must comply with NIST, ISO 27001, or sector-specific regulations. Engineers asking about risk parameters for a chemical plant can't have their queries traverse the open internet. Everything stays within FM's virtual network, right up to the point where the LLM generates a response.

What This Means for Windows Environments

FM's platform is accessed via a Windows-based mobile app that integrates with the company's existing single sign-on. It uses Windows authentication natively, meaning no additional credential vaults for IT to manage. The app caches frequently used document snippets locally on the device, encrypted with BitLocker, so engineers working in remote areas with spotty connectivity can still get answers offline. That hybrid online-offline capability is increasingly important as field service expands into IoT-connected equipment where real-time data might intermittently feed the knowledge base.

For organizations running Windows 11 Pro on ruggedized tablets, the platform provides a glimpse of how AI can be embedded directly into the OS experience. Microsoft has been teasing Windows Copilot for enterprise, and services like Azure AI Search provide the backend intelligence to make those assistants useful for specialized verticals. FM's deployment is essentially a vertical-specific Copilot, tuned not for email summarization but for life-saving engineering decisions.

Looking Ahead

FM and Spyglass MTG plan to expand the platform with multimodal retrieval—allowing engineers to upload a photo of a damaged component and get instant risk guidance based on visual similarity to known failure modes. Azure AI Search's vector capabilities already support image embeddings, and with Microsoft's ongoing investment in multimodal models, the roadmap is clear. Future iterations may also incorporate streaming telemetry from IoT sensors, so the RAG engine can reason over both historical documentation and live equipment data.

As governed RAG matures, expect more enterprises to follow FM's lead. Microsoft has been clear that Azure AI Search is a cornerstone of its enterprise AI strategy, and the early June 2026 profile serves as a public benchmark for what's achievable when retrieval meets governance. For FM's field engineers, the immediate payoff is simple: they get home safer, faster, and with confidence that every decision is backed by the best available evidence.