Operating Models for Enterprise AI

As AI becomes embedded in core business processes, executive conversations are shifting toward durability. For technical leaders, the  question is no longer whether AI can deliver value, but whether the organization is structured to support it over time.

I sat down with Dael Williamson, Chief Technology Officer for EMEA, to explore what actually distinguishes companies that are serious about AI from those that are simply experimenting.

What emerged was an operating-model view of enterprise AI, highlighting where structure and ownership matter more than technology choices.

The signals that indicate an organization is serious about AI

Catherine Brown: When you walk into an organization that is investing seriously in AI, what do you look for first in their operating model and platform choices?

Dael Williamson: Most organizations today recognize that AI is real and that it matters. What varies is how clearly that commitment shows up in the operating model.

The first thing I look at is ownership. Who owns data and who owns AI, and how close that ownership sits to the CEO. If data and AI are owned directly by or close to the CEO, that signals a high level of strategic importance. More often, ownership sits several layers down, and in many cases data and AI are owned by entirely different groups.

When AI is structurally distant from data, the result tends to be static use cases and fragmented experiences. But the world that organizations operate in is dynamic. Traffic changes. Markets shift. Supply chains fluctuate. If data and AI are separated, it becomes very difficult to respond to that reality in real time.

The second thing I look for is whether the organization has an inventory of its data assets. While financial and physical assets are well documented, many organizations are still maturing how they catalog and understand data assets. In many cases, organizations don’t fully know what data they have, where it lives, or how valuable it might be.

The third signal is how broadly the organization defines data. Many still think of data primarily as structured tables or logs. But images, email, collaboration tools (documents, spreadsheets), and code all contain rich operational insight. Organizations that expand their definition of data tend to unlock far more value over time.

Why proximity between data and AI matters

Catherine: A lot of what you’ve described comes back to how closely data and AI are connected. Why does that proximity matter so much in practice?

Dael: When data and AI operate on the same foundation, organizations can support more dynamic use cases. When they’re separated, AI tends to rely on slower, more static inputs.

Traditional governance and catalog tools are very effective at managing structured data, but they struggle with unstructured, fast-changing sources. That’s one reason why expanding the scope of data governance is difficult, and why comprehensive data inventories are still rare.

If you’re trying to solve problems like liquidity modeling, credit risk, or supply chain resilience, you need AI working directly with timely, continuously updated data. Otherwise, decision-making is always delayed, and insights arrive after the moment they’re most useful.

Catherine: How are leading companies structuring the relationship between central teams and the business?

Dael: The leader responsible for data and AI needs a seat at the executive table, and they need a deep understanding of how these systems actually work. AI behaves differently from traditional software, and organizations benefit when leadership reflects that reality.

When it comes to tooling, leading companies resist the temptation to rely exclusively on AI features embedded across dozens of SaaS tools. While those tools can improve individual productivity, they rarely help teams work cohesively across functions. Over time, that approach tends to reinforce existing inconsistencies in definitions, metrics, and processes.

At the same time, these organizations are rethinking the build-versus-buy equation. They don’t aim to build everything internally, but they also avoid excessive lock-in. Portability, transparency, and control over data and AI assets are increasingly important.

Winning organizations also manage AI initiatives as a portfolio. Not every project succeeds. Some need to be paused. Others warrant additional investment. Treating AI as a portfolio of bets, rather than a linear roadmap, allows organizations to adapt as technology and business conditions evolve.

What the enterprise AI operating model looks like in three years

Catherine: Looking ahead, how do you expect enterprise AI operating models to change over the next three years?

Dael: Most organizations will still be in some stage of transformation, but one of the biggest shifts will be a narrowing of the traditional gap between IT and the business. Business teams will become more technically fluent, and technical teams will become more closely aligned with business outcomes. That shift is already underway, and it will continue.

As a result, IT organizations are likely to change in size and shape. Historically, IT has focused on risk management, governance, and operational complexity. AI is increasingly effective in those areas, particularly in cybersecurity, IT support, and compliance.

When organizations also reduce legacy complexity and move away from siloed vendor ecosystems, operating models begin to change more fundamentally. Teams become less defined by the systems they use and more by the outcomes they deliver.

Over time, this may lead to leaner organizations or the creation of entirely new units focused on new forms of value creation. Exactly how that plays out will differ by company.

How skills and roles evolve in an AI-driven enterprise

Catherine: That kind of operating model shift has major implications for talent. For our closing question, how do you see skills and roles evolving?

Dael: Many IT organizations will continue to shrink, largely because so much enterprise technology is still built on decades-old systems that are costly to maintain. At the same time, the software development lifecycle is changing. Tasks that once consumed most of the effort, such as manual coding, are increasingly AI-assisted. More time is now spent on evaluation, behavioral testing, guardrails, and ongoing monitoring.

That change brings business and technical teams closer together. Business teams become more involved in defining and validating behavior. Technical teams focus more on outcomes, reliability, and governance. New roles are emerging around observability, orchestration, and system oversight. These roles often blend technical, operational, and organizational skills, and they don’t always come from traditional engineering backgrounds.

Management itself is also evolving. As AI takes on more administrative work, management shifts back toward analysis, judgment, and improvement of how work flows. Critical thinking becomes essential. People who are comfortable experimenting, learning, and adapting will do well. And analytical and scientific mindsets will be increasingly valuable as organizations navigate this transition.

Closing Thoughts

Enterprise AI readiness is ultimately an operating model decision. Leaders who are making sustained progress have clear executive ownership of data and AI, treat data as a known and governed asset, and ensure AI works directly with timely, shared data rather than through fragmented handoffs. They manage AI initiatives as a portfolio, not a pipeline, with discipline around where to invest, pause, or stop. And they organize teams around evaluation, oversight, and outcomes rather than tools or projects. The organizations that succeed will not be those that predict the future of AI most accurately, but those built to adapt as it changes.

To learn more about building an effective operating model, download the Databricks AI Maturity Model.