Transforming the data foundation behind our datacenter expansion at Microsoft

The team building our datacenters here at Microsoft needed to get organized.

Microsoft Cloud Operations + Innovation (CO+I), our organization that plans, builds, and operates the datacenters behind Microsoft Azure, built OneMDM UniVerse to give us a centralized master data platform to manage the materials, suppliers, equipment, and locations behind our datacenter buildout.

“Everyone is focused on building really good agents and really good workflows. But all of that begins to fail at scale if you don’t have the foundation set up. Master data management is absolutely that foundation.”

Dikul Singh, principal product manager, Microsoft CO+I

Built on Microsoft Dynamics 365, the new platform has a name that addresses our challenge and our solution.

OneMDM, the first part of the name, is a single master data management system that replaces many of our disparate sources of data. UniVerse, the second part of the name, is the ambition layered on top, because our goal was never to master one data set in isolation. We needed to hold the full hierarchy of our datacenter in a single place, from the cloud it serves down to the individual co-location room.

“Everyone is focused on building really good agents and really good workflows,” says Dikul Singh, a principal product manager in CO+I. “But all of that begins to fail at scale if you don’t have the foundation set up. Master data management is absolutely that foundation.”

100 systems, 100 versions of the truth

As our Azure business grew, CO+I grew with it. More than 100 different data systems emerged across planning, supply chain, construction, and operations functions, each with its own copy and its own definitions of the data required to maintain and build new datacenters.

“Everybody knew we had to connect everything together, but when you’re running at breakneck speed, it’s very difficult to put things together and standardize.”

Praveen Nair, principal software engineering lead, CO+I

Explosive growth meant our teams across CO+I were pursuing one goal via separate routes, and their data diverged in the same way.

“Everybody knew we had to connect everything together, but when you’re running at breakneck speed, it’s very difficult to put things together and standardize,” says Praveen Nair, a principal software engineering lead in CO+I.

Location became the clearest illustration of divergent data. One team would cite the physical address, another the Azure region, another the planning metro, another a position in the org hierarchy. Same datacenter, four different answers.

“Employees would ask, ‘Who’s the owner for this data? Who’s the steward?’ There was no governance and no data quality at all in what people were consuming.”

Kavitha Subramaniam, senior software engineer, CO+I

Inconsistent definitions were not the only problem. Even when teams agreed on what a record should say, no one was accountable for keeping it consistent.

“Employees would ask, ‘Who’s the owner for this data? Who’s the steward?’” says Kavitha Subramaniam, a senior software engineer in CO+I. “There was no governance and no data quality at all in what people were consuming.”

“With Excel files, you don’t know who the data owners are, and if you’re aggregating across a whole bunch of them, it becomes very challenging.”

Karen Albrecht, director of asset inventory management systems, CO+I

On the operations side, the gaps were filled by spreadsheets across roughly 350 campuses and more than 4,000 colocations.

“With Excel files, you don’t know who the data owners are, and if you’re aggregating across a whole bunch of them, it becomes very challenging,” says Karen Albrecht, director of asset inventory management systems in CO+I.

For years, people filled the gaps by hand, reconciling systems record by record. At this size, that manual effort could no longer keep up. The answer had to be a system, not more people.

Building a platform that teams would actually use

We built OneMDM UniVerse on Microsoft Dynamics 365, which serves as the platform’s business application layer. This meant the team did not have to build the fundamentals itself.

“There are industry-standard solutions on the market, but they didn’t quite fit the way we’re structured. They would have required a lot of customization, which would become harder to manage and maintain over time.”

Venkatesh Muthiah, principal software engineering manager, CO+I

Security and permissions came built-in, along with data storage and configurable process flows the team could shape to match how each domain approves a record. Microsoft Dataverse sits underneath, making mastered data readable by other services without a custom pipeline for each one.

That freed the team to focus on the master data problem itself. The packaged products on the market couldn’t do that part for them.

“There are industry-standard solutions on the market, but they didn’t quite fit the way we’re structured,” says Venkatesh Muthiah, principal software engineering manager in CO+I. “They would have required a lot of customization, which would become harder to manage and maintain over time.”

Three principles shaped the platform: scalability, reusability, and governance. Data enters the product through APIs, integrations, or the user interface. These patterns were built once and then reused as each new system came online.

Governance was designed with each domain rather than imposed on it. We set up a data governance council that named owners and stewards for each domain and defined the approval workflows so that every domain used the same process. Rogue data dumps that previously went through without approval now underwent review, first through an AI screen and then with a human review as a final check.

The team also used AI to help with ingestion. Exact-match validation had existed before, but it was imperfect—when two pieces of data appeared to be similar but were not exact duplicates, the AI review wouldn’t catch it.

“Now, the moment I try to create the record, even before governance kicks off, we check for duplicates and run a similarity check,” Subramaniam says. “Then it tells us, ‘Dublin DC already exists. Do you still want to continue?’”

Configurable quality rules run continuously, with dashboards showing business users which records are failing and why. Some are authored by the business; others are AI-generated.

Getting the team aligned came down to agreeing to use the same language.

“We had so many ways to label a location, and we still do, but OneMDM was able to connect all of them together,” Singh says. “All of our leaders speak about Azure regions, but we speak about physical locations. Now, whether it’s capital funding, planning, or procurement, there’s no cross-talking.”

The immediate relief was evident.

“Every decision we make downstream is only as good as the data underneath it. Master data isn’t glamorous work, but it is the difference between growing fast and growing well. Sponsoring OneMDM was about giving our teams a common foundation so they could stop reconciling numbers and start acting on them.”

Sameer Tarey, partner director, Cloud Infrastructure Engineering

Teams that had spent their time reconciling competing versions of the same record got that time back, and coordination that once required a translator became routine. Now the connections held, because someone owned them.

For Sameer Tarey, partner director of Cloud Infrastructure Engineering, that alignment was the objective.

“Every decision we make downstream is only as good as the data underneath it,” Tarey says. “Master data isn’t glamorous work, but it is the difference between growing fast and growing well. Sponsoring OneMDM was about giving our teams a common foundation so they could stop reconciling numbers and start acting on them.”

Standardized equipment records and bills of materials now feed our integrated business planning engine, which tells planners what to buy, where, and when.

In operations, Albrecht’s team went from updating inventory records one at a time in Excel to updating 100% of the fleet in less than two days.

“Moving CO+I business data out of Excel and into modern data mastering platforms is transformative because it turns fragmented tracking into a trusted operating system for decisions,” Albrecht says. “It gives leaders governed, real-time visibility across teams, reduces manual reconciliation, and creates the data foundation we need to scale automation, AI, and measurable business outcomes across global datacenter operations.”

The adoption challenge

Adoption turned out just as challenging as building the platform. Every team that needed to adopt OneMDM already had a working system, a backlog, and deadlines of its own. Asking them to stop, clean up years of accumulated records, and re-point their integrations meant asking them to spend today’s capacity on a payoff that may not manifest for more than a year.

The team’s first instinct was to standardize by decree. If everyone adopted one definition of region, the mismatches would disappear. But being told to abandon their established systems for something new landed badly.

“I took a demo system, created sample materials, showed the entire flow to each group, and asked, ‘Is this what you want to see?’ Going from theoretical alignment to practical realization is what worked.”

Nirmala Kummathi, supply chain data manager, Microsoft Global Supply Chain

Only after working through the objection with several teams did the pattern become clear: This was not resistance; it was how each group actually ran its business. The fix was not a single definition imposed from above, but a common level in which everyone could follow the same map.

Nirmala Kummathi, supply chain data manager in Microsoft Global Supply Chain, describes a three-legged approach for solving this challenge, one that encompassed technology, process, and people. Through trial and error, Kummathi realized that agreeing in principle never worked as well as doing a demo and having a conversation face to face.

“I took a demo system, created sample materials, showed the entire flow to each group, and asked, ‘Is this what you want to see?’” Kummathi says. “Going from theoretical alignment to practical realization is what worked.”

What’s next in master data management

Adoption of OneMDM UniVerse continues across our organization, domain by domain. The foundation is already enabling us to accomplish things that weren’t practical before: predictive maintenance signals, conversational access to equipment data, and agent-based workflows built on data the organization can finally trust.

“Just because I get excited by data doesn’t mean everybody does. You have to figure out what’s in it for them,” Singh says. “If you want someone to build a ship, you don’t get them excited about building a ship; you teach them to long for the sea. I was selling OneMDM, and everybody said it was boring. So we started selling everything it could enable instead, and people got on board.”