Shared Data Ownership Turns Governance from an IT Afterthought Into an AI Prerequisite

As companies experiment with enterprise AI, many run into basic data problems they thought they had already solved. Where organizations have treated data governance as a siloed IT task rather than an enterprise operating model, that decision is now catching up with them: conflicting metrics, compounding technical debt, and AI investments that stall before they can deliver. Closing that gap means treating governance as a rigorous software engineering discipline, with the same ownership, standards, and accountability applied to any other critical system.

José Siles is a Data Engineer in the Analytics, Data & Integration division at Nestlé, with prior engineering roles at IAG, Inditex, and Amazon. At Inditex, he worked across a Snowflake-based data warehouse maintaining more than 150 data entities built on Kimball modelling principles. He also writes The Data Path, a weekly newsletter reaching over 10,000 data professionals. He sees the same pattern across every organization he has worked in: data is never treated with the discipline of a software product, and the technical debt compounds from there.

"The amount of technical debt these teams are generating is insane," Siles says. "For the same metric, like how many chocolates we sold today, marketing will report one number and finance will report another. You end up going to the CEO without knowing if you sold a million or 1.2 million today." The breakdown is not a reporting problem but a coordination problem, rooted in teams that build and maintain data in isolation with no shared standard for metrics.

To reduce technical debt, many organizations start by modernizing their infrastructure, and the architecture decision matters enormously. At Nestlé, Siles's team is migrating legacy systems including SAP, SharePoint, and scattered Excel files into a unified enterprise data cloud repository that unifies data across the organization. More than a quarter of organizations lose upwards of five million dollars annually, to poor data quality. Siles explains that the same controls that keep software deployments predictable, including shared repositories, code review, and standard pipelines, can prevent these discrepancies in data work as well. Without a culture of integrated data analytics and AI governance, even the best platforms get bypassed.

• Excel purgatory: The harder challenge is behavioral. For many teams, the platform choice matters less than whether everyone actually uses it. Spreadsheets continue to circulate because no one has enforced a single system of record, leaving data without lineage or any shared accountability for what it means. "The important thing is for people to adopt the technology and stop passing Excel spreadsheets back and forth. Sharing spreadsheets across teams with no lineage or tracking is a massive problem," notes Siles.

• One source to rule: The fix is not complicated in principle: consolidate everything into one place and make sure everyone uses it. "If you can use just a single tool to integrate all your data, it's a game changer," says Siles. "It might be a bit more expensive upfront, but at least everything is in one place."

The infrastructure question connects directly to AI readiness. Executive demand for AI returns routinely outpaces the operational reality. A large language model rarely untangles fragmented workflows or standardizes inconsistent data on its own, and strategies that ignore improving the quality of unstructured data and aligning fragmented cross-functional metrics tend to compound the problem. Proper governance, supported by integrating AI data platforms with underlying architecture, lays the groundwork to simplify enterprise AI adoption. Done right, it also supports responsible AI deployment and master data management at scale.

• Cart before the horse: "If you don't do proper data governance, your AI strategy won't work, because AI is only as good as your data," Siles explains. "It doesn't matter which LLM you apply on top of it. You are not going to automate anything." The instinct to reach for AI before the data foundation is in place is understandable, but it guarantees the outcome Siles describes: more spending, less output.

• Paper trails: The same logic extends to documentation, which Siles treats as a first-class engineering obligation rather than an afterthought. "I know no one likes documentation, but especially in the AI era, it's super important. If you don't have documentation, your AI cannot learn from your processes, your database, or your codebase," he adds.

Governance doesn't fail because data teams aren't doing their jobs. It fails because the rest of the organization never took ownership of it. In practice, that gap shows up in one of the most avoidable ways possible where teams build data products in isolation, unaware that another team has already built the same thing. Siles recommends one simple corrective habit before any new build begins: check whether it already exists.

• Leave Tom alone: The payoff is practical in both directions. If the work is already done, ask for access. If something similar is in progress, find out how the other team approached it and build on that foundation instead. "It's not just Tom from the data team doing it all," Siles notes. "It's about everyone documenting everything. Before we build something new that adds more technical debt, we have to realize another team might already be building something similar."

• Ctrl+C, Ctrl+V: The alternative is a pattern Siles has seen in every organization he has worked in. "If you work in silos where everyone is doing their own thing, you might end up with three identical data products built in different ways by three different teams," he says. The answer is a shared effort to know where data lives, why it exists, and who is responsible for it.

The deeper problem is one of incentives. Because intense business demands dictate sprint priorities, managers often reward visible, short-term output while leaving foundational data hygiene in the background. Consequently, unmanaged infrastructure becomes a growing financial drain, with unused database objects consuming thousands of dollars a month. Organizations frequently look to automate governance workflows to help control these costs. The teams that manage it well treat governance time as non-negotiable, not optional. Siles says the pattern is consistent: managers celebrate shipping tables, not documenting processes. "If you report that you shipped two tables into production for your data scientists this week, your manager will be happy," he says. "But if you spend the whole sprint documenting your processes, your manager will likely dismiss the effort and say they do not care. We have to help managers adapt."

"The best teams I know don't let the business dictate everything," Siles concludes. "They take what the business says, and that makes up 80% of the workload. The other 20% is dedicated to reducing technical debt. These teams are ultimately more productive."