Dominic Sartorio, VP of Product Marketing at Denodo – Interview Series

Dominic Sartorio is VP of Product Marketing at Denodo. Dominic has over 20 years of experience in the data management and governance market, having held various product and marketing leadership roles at Informatica, Protegrity, among other leading vendors.

Denodo is a global leader in data management, powering trustworthy AI agents and applications. The Denodo Platform, an award-winning logical data management solution, transforms enterprise data into reliable insights for AI, analytics, and self-service initiatives. Organizations worldwide use Denodo to deliver AI-ready, business-ready data in a fraction of the time compared to traditional data lakehouses, achieving up to 4x faster time-to-insight, 345% ROI, and 10x better performance. Based on insights from 850 enterprise leaders, Denodo’s AI Trust Gap Report reveals why many AI projects struggle to move beyond pilot stages and what organizations must do to build trustworthy, production-ready AI.

You’ve held senior leadership roles across companies like Informatica, Protegrity, Infoworks, and now Denodo, all focused on different layers of enterprise data infrastructure. How has your perspective on “trusted data” evolved as AI has shifted from analytics into autonomous and agentic systems?

Earlier in my career, trusted data was largely about accuracy, lineage, security, and giving analysts confidence in dashboards and reports. With agentic AI, the stakes are much higher because systems are not just interpreting data; they may be acting autonomously, triggering business workflows, or making decisions with real-world impact. That means trusted data now has to include live operational context, consistent business meaning, and guardrails enforced so one can have confidence that agents are acting correctly and safely.

Denodo’s AI Trust Gap Report found that 66% of organizations say AI data must be real-time or near real-time to be trustworthy. Why do you think so many enterprises are still struggling to deliver live operational data to AI systems?

Most enterprises were not architected for AI agents that need live situational awareness across many systems. Their data is spread across applications, clouds, warehouses, lakehouses, legacy systems, and other operational platforms. They may be copying this data to a central warehouse or data lake for analytics and BI, but this isn’t appropriate for AI agents that need live situational awareness.  Once data is copied, it is no longer live.  It is possible to stream in real-time, but this gets very expensive very fast. This is exactly where Denodo’s logical data management approach becomes important, because it gives AI systems governed access to live data without requiring enterprises to constantly copy and re-platform everything.

One of the more striking findings in the report is that enterprise AI initiatives now pull from hundreds of data sources, with some organizations accessing more than 1,000. How does that level of fragmentation change the way enterprises should think about AI architecture?

At that level of fragmentation, the architecture cannot depend on physically consolidating every source before AI can use it. Enterprises need an abstraction layer that can discover, integrate, govern, and deliver data across the distributed reality they already have. In my view, data architecture has to become more logical, metadata-driven, and semantic, so agents can find the right data in context without being tightly coupled to underlying systems.

The report argues that many AI failures are actually “data architecture failures” rather than model failures. Do you think the industry has spent too much time obsessing over models while underestimating the importance of data infrastructure?

Yes. Models matter, of course, but many failed AI projects do not fail because the model is incapable; they fail because the model is operating with incomplete, stale, inconsistent, or poorly governed data. The model worked great in the pilot, using a well-defined and curated data set, but once deployed in the “real world”, with its distributed messiness, the AI fails to produce trustworthy outcomes.  My experience has been that enterprises get much better AI outcomes when they treat the data layer as a first-class part of the AI architecture, not as an afterthought.

Denodo frequently talks about semantic consistency and the importance of a universal semantic layer. As AI agents begin making decisions autonomously, how critical does semantic alignment become for preventing incorrect actions or hallucinated business logic?

Semantic alignment becomes absolutely critical. If one system defines “customer,” “revenue,” “risk,” or “churn” differently from another, an AI agent can produce a technically plausible answer that is still wrong for the given business context. A universal semantic layer helps ensure that agents operate with consistent business meaning, not just raw data access.

Your AI & Big Data Expo session focused on moving from AI pilots to production. In your experience, what are the biggest reasons enterprises get stuck in the “pilot phase” and fail to scale AI into real operational systems?

Pilots often work because they are narrow, manually curated, and insulated from the full complexity of the enterprise. Production AI has to deal with live data from many sources, security, governance, performance, auditability, changing business rules, and integration into real workflows. Many organizations get stuck because they build an impressive demo, but not the governed data foundation needed to operate AI reliably at scale.

The report cites predictions that a significant percentage of agentic AI projects could be canceled over the next few years due to escalating costs, unclear value, or inadequate risk controls. Do you think the industry is entering a phase where enterprises will become far more selective about which AI projects survive?

Yes, and I think that is healthy. The first wave of AI experimentation was about possibility; the next wave will be about operational value, cost discipline, and trust. The projects that survive will be the ones tied to measurable business outcomes and supported by the right data, governance, and architecture.

Security and governance appear throughout the report as recurring themes, particularly around “guardrails” for agentic AI. How should organizations balance autonomous AI capabilities with the need for strict access control and auditability?

The key is not to treat governance as something bolted on after the AI system is built. Access control, policy enforcement, lineage, and auditability need to be embedded in the data access layer itself, so AI agents only see and use the data they are authorized to access. With Denodo, the same governance policies can be applied consistently across distributed sources, which is essential when AI is operating across hybrid and multi-cloud environments.

Denodo positions logical data management as a way to unify access across hybrid and multi-cloud environments without constantly moving data. As enterprises increasingly adopt retrieval-based AI architectures, do you see “zero-copy” or logical-first architectures becoming the long-term direction for enterprise AI?

Yes. Retrieval-based AI depends on getting the right data at the right time, not necessarily moving every data set into a single repository in advance. A logical-first, zero-copy approach is much better aligned with how enterprises actually operate: data remains distributed, but AI can access it through a governed, semantic, real-time layer. That is the direction I believe enterprise AI has to go.

Looking ahead over the next three to five years, what do you think will separate organizations that successfully operationalize trustworthy AI from those that remain stuck in experimentation mode?

The winners will be the organizations that recognize AI is not just a model strategy; it is a data strategy, a governance strategy, and an operating model strategy. They will invest in live data access, semantic consistency, reusable governance, and architectures that can span the full enterprise. Those that continue building isolated pilots on fragmented or stale data will struggle to move beyond experimentation.

Thank you for the great interview, readers who wish to learn more should visit Denodo or download Denodo’s AI Trust Gap Report