AI systems that draw on enterprise data need to know which version of a record to use. Most companies keep separate copies of the same records in sales, finance, and operations, and over time those copies stop matching. Tracing where each copy came from is becoming a core step in getting data ready for AI.
Nandakumar Sivaraman is Senior Vice President and Chief Architect of Enterprise Data at Bridgenext, a digital engineering and consulting firm that helps enterprises build data and AI platforms. He has spent more than a decade at the company, where he helps lead enterprise architecture and data engineering work for clients, and he previously held solution architecture roles at Mphasis, Hitachi Consulting, and Nalco. In his view, how well AI performs depends on how carefully a company manages its data first.
"Without governance, lineage, and process re-engineering, it's very difficult to say that we are data mature," says Nandakumar. A data-mature organization can trace where its data came from and confirm it's reliable. Lineage is what makes that tracing possible. It records every step data passes through, including how each department reshapes it for its own work. That record matters most in fields such as financial services, healthcare, and supply chain.
Know who's responsible
A dashboard shows what data a company has. Knowing who owns that data and whether it can be trusted takes more work, especially once several departments keep their own copies. "Data visibility is the ability to know what data exists, where it resides, how it flows, who owns it, and how trustworthy it is, so that it can bring value across the enterprise," notes Nandakumar.
Some teams plan their data strategy without accounting for who owns the data or how reliable it is. The data can look fine until the company uses it to make decisions. "We may believe the data is handy, but when we get into the market and start using it for customer opportunities, the decisions may not get the right traction," he says.
He takes clients through the work in a set order. First, teams find the data and learn where it's stored, where it came from, and how people use it. Next, they study it and check that it's accurate. Then they apply governance, which means setting rules for who can use the data and how. People and AI systems act on the data only after those steps.
Ownership matters when data moves between departments. If finance and sales each change the same customer record, each version needs a clear owner. That way, when an AI system produces a result, the team knows who to go to about the data behind it.
Without governance, lineage, and process re-engineering, it's very difficult to say that we are data mature.
Nandakumar SivaramanSenior Vice President & Chief Architect (Enterprise Data), BridgenextMore systems, more handoffs
Some enterprises are keeping older business software in place because replacing it can cost millions and take years. The data those systems already hold can answer many of the business's questions now. In Nandakumar's client work, that can mean pulling data from 18 or more systems into one central platform. This gets the business answers sooner than a full replacement would, and it also means data changes hands more times before it reaches an AI system.
Each stage along the way can change the data. Ingestion brings it in from source systems, processing cleans and reshapes it, and orchestration moves it between tools. Governance needs to cover every stage, including the database. "If any of these layers is missed on governance, there's a risk to whether the data is trustworthy. To avoid that, we always bring governance to the top layer," he warns.
Problems can start at ingestion. Some teams pull in every dataset they can reach without asking whether they need it, and each extra dataset is one more version to track. That's why his sequence starts with discovery. Deciding what data a team needs before bringing it in keeps the data clean through the rest of the process.
This work moves faster than it once did. In his experience, governance that used to take six or seven months can now be in place within a few weeks. Business leaders see results sooner, and data teams can prepare the data before AI systems start using it.
Data built to share
Much of the focus in enterprise data is on real-time information, and some decisions do need data from the last few minutes. Nandakumar sees value in historical data too. It shows how data looked before it changed, and it gives AI models enough volume to find patterns and estimate what comes next. That history is only useful if it has been governed and tracked from the start.
More enterprises are starting to package their most useful datasets as services. A team prepares a governed dataset once, and every team and application that needs it can use the same version. Some of his clients run on more than one cloud provider and want the freedom to switch without rebuilding. Running these services on Kubernetes, the open-source system for running software across cloud and on-premises environments, lets the same package work wherever it's deployed.
For enterprises that want their AI systems working from consistent data, that means fewer versions to reconcile and one place to set the rules. "If a data entity gives great value to an organization, and customers see that it's reusable across their ecosystem, why not? It can become data as a service, and that concept is completely in line with the Kubernetes model," he adds.
