Transforming facility operations at Microsoft with AI maps

Indoor building maps matter the moment accurate location data become important to solving an issue in facilities.

Imagine a facilities service technician responding to a high‑priority heating issue. The service ticket has the right building, floor and space, but no clear indication of where in the space the problem exactly is—this can be a particularly challenging problem when dealing with large spaces like we do here at Microsoft. Making matters more complicated, the technician might also need specialized schematics that are behind walls and ceilings.

In the past, it might have taken that technician a long time to get that necessary context.

Not anymore.

Thanks to a solution we created that is internal to Microsoft, our indoor maps are now always current. (And while this solution isn’t presently available to customers, we’re sharing our story around it in hopes that you can learn from our approach.)

With these maps available, the service ticket mentioned above now includes a visualization of the floor plan, which immediately indicates the correct room and highlights the faulty equipment’s exact location (if already on the floor plan).

The technician can now diagnose the problem in minutes. This can be done on their equipment, without the need to understand how to use specialized software or having knowledge of the building’s layout.

This shows the value of indoor maps when they work correctly. But our maps here at Microsoft didn’t always work this way.

A long-standing map gap

For years, facility service technicians at Microsoft could access floor plans that were stored in a central repository but needed specialized software to view them. Our floor plan files came from a wide variety of vendors, with different naming conventions and drawing standards.

Initially, we created indoor maps using this data for some buildings requiring a lot of manual work. As a result, updates to the maps were slow and expensive. As soon as a map slipped out of sync with reality, teams stopped relying on it.

“Enterprises have struggled for years to maintain accurate indoor maps. The heart of this struggle is ultimately standards that are applied inconsistently to the source material.”

Vishu Admal, program and product lead, AI indoor maps initiative, Microsoft Digital

These issues had a real impact on our day‑to‑day operations:

• Our facilities teams didn’t have the spatial context in their work order to understand exactly where problems were happening, because they didn’t have convenient access to detailed map layers that indicated the precise location of building elements (such as plumbing, or heating and cooling systems).
• Our security teams couldn’t easily overlay incident data on floor plans.
• Our IT teams couldn’t map device locations to the real world with confidence and relied on PDF version of maps.

“Enterprises have struggled for years to maintain accurate indoor maps,” says Vishu Admal, our program and product lead for our AI indoor maps initiative here in Microsoft Digital, the company’s IT organization. “The heart of the struggle is ultimately standards that are applied inconsistently to the source material.”

Recently, we’ve developed an intriguing solution: An AI‑driven mapping data pipeline—with out-of-the-box large language models (LLMs)—that recognizes patterns, identifies inconsistencies, and produces updated indoor maps every day, as the floor plans evolve and change.

Today, that data pipeline is keeping our indoor maps up to date for more than 500 buildings around the world. It supports the systems and teams that keep our campuses operating every day—facilities, space management, security, and IT.

And more importantly, this solution makes sure that when someone has a high-priority need for an indoor map, it’s completely accurate and current.

“The problem has always been tripping up on the varying quality and consistency of the AutoCAD files. This is especially true at enterprise scale, where drawings come from different firms and have different standards.”

I.M. Ndimubanzi, engineering manager, Microsoft Digital

From CAD to operational maps: Finding a solution

Some indoor mapping projects start with a simple assumption: The floor plan is already structured data.

Our project didn’t have that. What we had was computer-aided design (CAD) geometry, plus text, plus years of vendor variation.

Different architecture and construction partners drew buildings in different ways. Labels, layers, symbols, and even basic conventions (like how rooms were “closed” in a drawing) weren’t consistently followed. That inconsistency is what broke any attempts at automation.

“The problem has always been tripping up on the varying quality and consistency in the AutoCAD files,” says I.M. Ndimubanzi, an engineering manager in Microsoft Digital who is the technical lead for our indoor maps initiative. “This is especially true at enterprise scale, where drawings come from different firms and have different standards.”

So, we built a data pipeline that assumes the input will be messy, but it can still produce a reliable output, time and again.

Converting CAD geometry into render-ready maps

We split this work into three stages. In brief, these can be labeled as: parse, interpret, and serialize.

1. Parse CAD input into machine-usable signals

We start by extracting raw geometry and text from CAD using open-source parsing libraries. That gives us the basic data we can feed into downstream steps without forcing every file to look identical first.

2. Use AI for interpretation and hygiene

The hardest part of the work isn’t reading the CAD file. It’s interpreting what the drawing means when dealing with variations in room names, abbreviations, and other conventions (which may differ by vendor, region, or even building).

This is where we use AI-driven large language models to transform the extracted CAD signals into structured data.

Instead of manually cleaning and translating each file, we use AI models to ingest CAD drawings directly and interpret what the data represents. Walls become walls, rooms become rooms. Doors, elevators, and fixtures are identified as distinct, usable elements rather than raw line work.

That same approach helps solve a long‑standing data hygiene issue: inconsistent naming. For example: across the portfolio, the same type of space can appear as “Conference Room,” “Conf. Rm.,” “MPR,” or “Multi‑Purpose Room.”

The AI helps normalize those variations into standardized space categories, turning messy labels into consistent, structured data that can be reused across systems.

3. Serialize to GeoJSON with proven tooling

Once AI produces a structured representation, we convert the data into GeoJSON—a popular spatial data exchange and rendering format—using open-source tooling.

GeoJSON gives us a clean, reliable data source for our mapping tools. This keeps the final output consistent and predictable, which is critical for rendering at scale and integrating into other applications.

Note that this design is intentional: AI does the interpretation, while deterministic tooling does the formatting. This separation is what makes the pipeline stable.

“As long as they can add this SDK in their application or interface, they can connect to our databases. It gives them access to our map library.”

Amr Dawood, senior software engineer, Microsoft Digital

Creating an SDK that makes maps usable everywhere

A mapping pipeline is only valuable if other teams can use the results without becoming mapping experts. That’s why we paired the mapping pipeline with a software development kit (SDK) that makes indoor maps embeddable inside operational tools.

“As long as they can add this SDK in their application or interface, they can connect to our databases,” says Amr Dawood, a senior software engineer in Microsoft Digital. “It gives them access to our map library. They can use predefined functions to choose the buildings, the layers they want to render, and how they want to display and order those layers.”

We built this SDK so product teams can treat the new indoor maps like any other UI component:

• Drop it into a web app and connect to our map storage without building custom integration
• Choose buildings and floors using built-in selectors and navigation patterns
• Toggle layers to show only what matters for the scenario, including specialized operational layers
• Overlay operational data on top of the floor plan, so teams can visualize work in spatial context, not just in tables
• Make the map interactive by adding pins, polygons, and other spatial annotations directly in the app experience

Under the hood, the SDK is built on MapLibre, and open-source toolset for interactive maps and geospatial visualization. It gives the team a mature rendering foundation without locking them into a bespoke mapping stack.

We also built the SDK for “plug and play” adoption. That means providing examples, tutorials, and guidance so teams can embed maps quickly and consistently, instead of reinventing the same integration patterns across multiple apps.

This is the part of the solution that turns the pipeline into a platform. It’s how we move from “we have maps” to “any team can build with maps.”

Turning floor plans into user interfaces

We’re currently integrating indoor maps directly into several of our facilities processes and applications. As we do, we’ve noticed an important change: Employees stop treating the floor plan as reference material and start treating it as the user interface for detailed building information.

“Managing a physical space through tables and charts only gets you so far. It’s much more powerful when that information is visualized through a floor plan.”

Harris Thamby, integration lead, Microsoft Digital

Take our LiveCampus app, for example.

Live Campus is an internal app used by the Microsoft Facilities team to aggregate all operational information about Microsoft buildings into a single, comprehensive view. It simplifies facilities management by integrating various data points and presenting them on a visual floor plan.

Historically, building data lived in tables, tickets, and dashboards scattered across multiple systems. None of it was spatially oriented by default. If something broke, you read a text description, then tried to figure out the location.

With our new indoor map solution, location comes first, and it’s bringing life to Live Campus.

Instead of navigating through multiple systems to gather information, Facilities employees can access everything they need in one place. The floor plan serves as the primary interface, allowing users to overlay different types of information, such as facility tickets, service issues, and role-based employee data.

This visual approach helps facility managers quickly identify and address problems, improving operational efficiency.

“Everything is aggregated and presented as a building view,” says Harris Thamby, who leads integration work for Live Campus in Microsoft Digital. “But managing a physical space through tables and charts only gets you so far. It’s much more powerful when that information is visualized through a floor plan.”

Live Campus uses the latest AI‑generated maps through our SDK. That means facilities teams always see the current layout, not a snapshot from months ago.

And because the map is up-to-date, teams can trust what they’re seeing.

We’re also making big changes to FacilityLink, our internal implementation of Dynamics 365 Field Service. The new indoor maps solution is becoming the center of the technician experience.

“Our facilities teams will use the same maps in Dynamics 365 Field Service. They can turn layers on and off to see exactly what they need. Where are the tickets? Where are people sitting? What areas are impacted?”

Sonaly Choudary, program manager, Microsoft Digital

As the integration progresses, technicians will be able to open a work order and see the associated indoor map alongside the request, either in FacilityLink on the web or through Microsoft Dynamics 365 Field Service Mobile on a phone or tablet.

From the same screen where they read the details of the issue, they can also visualize the exact floor, room, and surrounding spatial context of the problem. This will reduce the need to switch between systems or return to a desk to look up drawings, helping technicians diagnose issues faster and with more confidence while they are already on site.

That spatial context—available on a mobile device instantly—changes how work gets prioritized.

“Our facilities teams will use the same maps in Dynamics 365 Field Service,” says Sonaly Choudary, a program manager with Microsoft Digital. “They can turn layers on and off to see exactly what they need. Where are the tickets? Where are people sitting? What areas are impacted?”

“When a technician responds to a work order, every minute matters. Historically, they had to jump between systems to find drawings, interpret layouts, and understand what was behind walls or ceilings. We’re eliminating that friction and giving technicians the spatial context they need to diagnose and fix issues faster.”

Michelle Schaefer, principal program manager, Microsoft Digital

Instead of scanning lists of open issues, facilities managers and technicians can see clusters of problems on a single floor.

They can spot patterns and determine when a single issue is affecting multiple teams or when a problem is isolated.

“When a technician responds to a work order, every minute matters,” says Michelle Schaefer, a principal program manager in Microsoft Digital. “Historically, they had to jump between systems to find drawings, interpret layouts, and understand what was behind walls or ceilings. By embedding AI‑generated indoor maps directly into FacilityLink, we’re eliminating that friction and giving technicians the spatial context they need to diagnose and fix issues faster.”

The result?

It’s faster and easier for facilities teams to understand where an issue is, what assets are involved, and how to act—without leaving the system they already depend on. Indoor maps become a practical extension of FacilityLink, embedding spatial awareness directly into day‑to‑day facility operations. If a service ticket is opened, it’s no longer just text; it’s a pin on the map.

Outcomes and what’s next

As our maps become reliable and embedded into daily workflows, the solution stops being a mapping project. It becomes a platform with the potential for impact beyond facilities operations.

The most immediate outcome has been scale. We’re moving from selectively supporting a limited set of buildings to supporting the entire real estate portfolio. Maps will be onboarded, updated, and maintained automatically, without the manual effort that had slowed previous approaches.

That automation changed the economics.

Instead of paying for one‑off conversions or ongoing vendor updates, the mapping pipeline runs continuously. As layouts change, the maps are automatically updated. That consistency is what allows downstream systems to depend on the data.

We’re continuing to refine the mapping pipeline as models improve and standards evolve. The SDK can be expanded to support more scenarios and platforms. And additional layers and integrations will unlock new operational use cases across facilities, IT, and security.

As teams across Microsoft have learned more about AI indoor maps, the excitement and adoption potential keeps growing. When spatial data is accurate, current, and reusable, teams stop asking whether they can visualize a problem and start asking what else they can do with it.

That’s the real outcome of this work: Not just better maps, but better decisions, built on a shared, trusted source of spatial truth.