Gustavo Pilatti explains why Data and AI Governance fail when ownership, trust, and data foundations are weak in the age where AI agents are rapidly scaling.
During a recent conversation on The Executive Outlook, Gustavo Pilatti began with three simple questions that reveal the real health of Data and AI Governance inside any organization.
Do the numbers on the executive board match the numbers on the operational dashboards? Can leaders trace their most important KPIs from the source all the way to the CEO’s screen? And do people actually trust the dashboards they use every day?
Over fifteen years of working with data in different types of organizations, Gustavo says the answer is almost always the same.
No. No. And no.
Those answers explain why so many companies struggle with data, analytics, and now artificial intelligence. The problem is rarely the next tool. It is rarely the next model. It is rarely the next platform. The deeper problem is that the foundation underneath those investments is still not trusted.
Gustavo holds a PhD in data science with a focus on platform economics. He has built data functions from the ground up, including growing a data team from one person to 35 people in four years inside one of Brazil’s fastest-growing food delivery platforms. He has seen every major data wave arrive with urgency. Big data. Cloud platforms. Machine learning. Streaming data. MLOps. Generative AI. Agentic AI.
Each wave came faster than the previous one. Each promised transformation. Yet many organizations remained stuck with the same unresolved question.
Can the business actually trust the data it is using?
That is where Gustavo’s view becomes important for every executive. While everyone is chasing the next wave, the foundation is still leaking.
Gustavo is clear about why Data and AI Governance fail in many organizations. It is not only a technology problem. It is not only a budget problem. It is not even a strategy problem.
It is a perception problem.
For years, governance has been framed as something boring. Something owned by legal teams. Something connected to compliance forms, policies, and approval of workflows. Something that slows people down instead of helping them move with confidence.
That framing has damaged how leaders think about governance. It has allowed organizations to treat it as a reporting exercise instead of a business capability.
Gustavo has seen the same cycle repeat several times in his career. Boards and senior leaders often want to check the next buzzword card. At one point, the card was hiring data scientists. Then it was appointing a Chief Data Officer. Then it became the first AI company. Today, the card has AI agents in production.
The problem is that the announcement often comes before the foundation exists.
The same low-quality data that people do not trust in dashboards is now being used to feed probabilistic AI systems. The same unclear ownership that created confusion in reporting is now being carried into agentic AI workflows. The same weak controls that made KPIs unreliable are now affecting automated decisions.
When that happens, AI does not solve the problem. It scales the problem.
The most powerful idea Gustavo offers is simple. Data and AI Governance is not the brake. It is the seatbelt that lets the organization drive faster.
This is the executive reframe most companies need.
A strong data governance strategy does not mean creating long policy documents that nobody reads. It does not mean building a compliance theatre that looks impressive but changes nothing in daily operations. It means creating a short, clear set of rules that people can understand, follow, and believe in.
A practical Data and AI governance framework begins with clarity. What data moves from one system to another? How often does it move? In what format? Who owns each critical dataset? What happens when the data fails to check? Can the organization trace the journey of a KPI from its original source to the executive dashboard?
These are not abstract governance questions. They are business survival questions.
The strongest data governance framework examples often come from highly regulated environments, especially in Europe. Regulators may ask which data was used for a decision, where that data came from, who made the decision, and which customers were affected. Organizations that can answer these questions are not simply compliant. They are more prepared to compete.
This is where data governance best practices become more than a regulatory discipline. They become a foundation for speed, trust, and AI readiness.
A clear data governance strategy and roadmap give leaders confidence that the numbers they see are reliable, the systems they use are traceable, and the decisions being automated are accountable.
Without that, every AI initiative is built on uncertainty.
When Gustavo helped grow data function from one person to 35 people, the hardest part was not technology.
The clouds were available. The tools were improving. The company was growing. The technical side had its challenges, but it was not the real bottleneck.
The real challenge was humans.
The first challenge was creating genuine buys from the business. A data team does not generate value in isolation. Even the most accurate model has no impact if another team does not trust it, use it, and act on it.
This is one of Gustavo’s strongest leadership insights. Data teams are not revenue engines by themselves. Their success depends on how well they help other teams make better decisions. Trust is not built through a beautiful dashboard or a polished presentation. It is built one project at a time through concrete results.
The breakthrough came when the data team stopped operating like a support function and started sharing responsibility for business outcomes. The team’s goals were connected to the company’s goals, including orders, revenue, and retention.
That changed the relationship between data and business.
The team was no longer running reports. It had skin in the game. It became a partner in outcomes.
For leaders building a data governance strategy, this point matters deeply. Governance cannot live only in policy documents. It must live in incentives, ownership, and the daily choices teams make together.
The second challenge was hiring. Gustavo believes technical skills are important, but they are easier to test than character, collaboration, and judgment. His hiring filter is what he calls the airport test. If he missed a connecting flight with this person and had to spend 8 hours at the airport, would that person make the experience better or worse?
If the answer is unclear, he does not hire the person.
That may sound simple, but it reflects a larger truth. Best practices in data governance are carried out by people, not documents. If the people responsible for governance cannot work across teams, build trust, and speak the language of the business, the framework will fail no matter how well it is designed.
The most urgent question in AI and data governance today is one that Gustavo believes many companies are still avoiding.
Who owns the data that AI agents consume, create, and act on?
This question matters because AI agents do not behave like traditional software. Traditional systems store data, process instructions, and produce predictable outputs. AI agents interact, infer, remember patterns, make decisions, and generate new intelligence from the data they touch.
That creates a new ownership problem.
When an AI agent interacts with thousands of customers, it creates something beyond raw data. It creates memory. It creates behavioral patterns. It creates derivative intelligence. It may produce decisions or recommendations that influence real business outcomes.
So, who owns that intelligence?
Is it the customer who provided the original data? Is it the company that deployed the agent? Is it the AI platform provider that sees the prompts, usage patterns, and telemetry? Is it the regulator that controls what can and cannot be done with certain categories of data?
Gustavo frames four possible ownership layers.
The first is the principal, which could be the customer, employee, partner, or company that originally owns the data. This is the traditional ownership model most organizations understand.
The second is the agent, who may not be a legal person but still creates operational intelligence through interaction, memory, and decisions.
The third is the platform, which could be the cloud provider, LLM provider, or orchestration layer. Even when contracts say the data belongs to the client, the platform may still see prompts, patterns, telemetry, and usage behavior.
The fourth is the regulator, especially in regions where certain data rights are nontransferable. In those cases, even if the user, company, and platform agree on something, the regulator can still restrict what is allowed.
Most organizations are only thinking about the first layer. They are still using a mental model from the period before agentic AI. That model assumes that data has one clear owner, and that ownership is mostly about storage.
But AI agents do not just store it. They accumulate. They infer. They recombine. They decide.
That is why the single owner's assumption breaks down. The real owner of data or agent output is the party that captures its economic and decisional benefit, not necessarily the one who physically holds or stores it.
Gartner has predicted that a round 40% of agentic AI projects may be cancelled by 2027 because of rising costs, unclear business value, poor risk controls, and weak governance. Gustavo is not surprised by that prediction. In fact, he believes the real number could be higher.
His concern is not only that some projects will be cancelled. His bigger concern is that many weak projects will survive and enter in production.
That is where the real risk begins.
Many companies are launching agentic AI projects because the board is asking them; competitors are announcing them, or the market is rewarding the appearance of innovation. The agent becomes the answer before the organization has clearly defined the question.
This is one of the biggest mistakes leaders can make.
If a team cannot explain the business problem in one clear sentence, the project is already at risk. Token usage is not an outcome. The number of agents in production is not an outcome. The number of use cases launched is not an outcome.
A real outcome is a business problem solved, a cost reduced, a process improved, a customer experience strengthened, or a decision made with greater confidence.
Without that clarity, companies end up chasing vanity metrics instead of value.
Gustavo also warns that public conversation around AI agents is too optimistic. In real deployments, agents often fail to complete assigned tasks at very high rates, depending on the benchmark and context. Even more concerning, agents are still weak at identifying their own mistakes.
That means companies cannot rely on agents to know when they are wrong.
Human oversight, observability, ownership, and governance are not optional. They are the conditions that make safe scaling possible.
This is why Data and AI Governance matters so much in the agentic AI era. Garbage in, garbage out is no longer the full problem. With AI agents, garbage can move across systems, compound across interactions, and create decisions that affect real customers.
Across the full conversation, Gustavo does not describe governance as a technical discipline or a regulatory obligation. He describes it as the difference between organizations that can truly benefit from AI and organizations that will pay the price of deploying it too quickly on weak foundations.
His message to executives is direct.
Stop chasing buzzword cards. Stop measuring AI progress by the number of agents launched. Stop treating governance as paperwork. Start asking whether your teams trust the data, whether your KPIs can be traced, whether every critical dataset has an owner, and whether your AI systems can be audited when something goes wrong.
A strong data governance strategy and roadmap are not about slowing innovation. It is about making innovation safer, faster, and more valuable.
The future of AI will not belong to the organizations with the most pilots. It will belong to organizations that can connect trustworthy data, clear ownership, business outcomes, and human accountability.
That is where Data and AI Governance become boardroom priorities.
And in 2026, every leader in building AI agents needs to ask the question Gustavo keeps bringing back to the table.
Who owns the intelligence your agents are creating?
That question may decide whether AI becomes a competitive advantage or the next expensive mistake.
Want to hear more conversations with leaders shaping data governance, AI strategy, and enterprise transformation? Explore more on The Executive Outlook.
Listen in for business strategies and interviews with business experts to help you build a business and a life that you love.
About Us
