Active Governance Starts With the Decision

When I speak with customers about data governance, they almost never begin with governance. They begin with a decision they are trying to make, and with the small pile of doubts sitting underneath it. Can I trust this number? Why did it change last month? Am I allowed to combine it with the customer data from another team? If I share the result, does something break? And increasingly, can an AI application or an agent act on this safely when nobody is watching the interaction?

That is why a number of Gartner’s 2026 Data & Analytics trends caught my attention. The individual technologies matter, but what struck me was the theme running underneath them. As data moves faster and AI participates more directly in business decisions, governance has to move closer to the decision too. It cannot live only in documentation written months earlier and somewhere upstream of the moment anything is actually used.

This article is my view of what that shift looks like from the customer and partner side. The foundations organizations have already built still matter. What changes is where governance and business context need to operate, and what has to travel with data when a person or an agent picks it up.

When I speak with customers about data governance, they rarely begin by talking about governance.

They talk about decisions.

They want to understand why a number changed between one reporting cycle and the next. They want to know whether the information in front of them can be trusted well enough to act on. They want to bring together information that lives in different parts of the organization, usually because the question they are answering does not respect internal boundaries. They want to respond to an operational problem before it becomes an expensive one. They want to know whether they are permitted to use a particular field, and whether the answer they produce can be shared. More and more often, they also want to know whether an AI system can consume that information and act on it responsibly.

Governance is present in every one of those conversations. It is simply not the word people use. They describe it as trust, or permission, or explanation, or confidence.

So when I read Gartner’s Top Trends for Data and Analytics for 2026, what interested me was not another prediction that AI will become more important. Everyone expects that. What interested me was how much of Gartner’s attention has moved toward decisions, governance, context and operational control. That is a noticeably different emphasis, and it mirrors what customers have been telling us in their own language for some time.

One of the Gartner trends is framed as reducing AI agent risk with decision governance. Gartner’s argument is that AI agents are increasingly participating in decisions across the organization, from strategic choices through to tactical and operational ones, and that organizations therefore need to make the logic of those decisions explicit rather than implied. Gartner predicts that by 2029, explicitly modeled business decisions will be five times more trusted and 80% faster than ungoverned decisions.

What I find interesting in that prediction is the pairing of speed and trust. Those two things are often treated as opposites inside enterprises. Governance is assumed to slow things down, and speed is assumed to require some quiet loosening of control. Gartner’s point suggests the opposite relationship when decisions are modeled properly. Organizations want AI to accelerate decisions, but a faster decision only creates value if the information, context and controls behind it can be trusted. Speed without trust simply produces the wrong answer sooner.

This is where Data Tiles thinking about Decision-Driven organizations comes in. Cameron Price has written extensively about moving from Data-Driven to Decision-Driven. Working with customers and partners, I see the practical reason for that shift every day.

The difference is easier to see in the questions each approach asks first. A Data-Driven approach usually begins with the data estate. What data do we have? Where does it live? How do we integrate it? Where should we store it? What dashboard should we build on top of it? Those are legitimate and often necessary questions, and the platforms built to answer them are genuinely valuable foundations.

A Decision-Driven approach begins somewhere else. It starts by asking what decision we are trying to improve, and then works backward from that decision to the information, context and controls required to support it.

Cameron describes Active Governance as governance operating at the moment data is accessed, combined, shared or consumed. From the customer side, the significance of that idea becomes very practical very quickly.

Think about what somebody actually does when they sit down with information. They ask a short sequence of questions, usually without noticing they are asking them.

Can I access it? Can I understand it? Can I combine it with this other information? Am I permitted to see this particular field? Can I share the result? Can my AI agent consume it? Can I trust the answer?

Every one of those is a governance question, and every one of them is asked at the point of use. That is why I think the point of consumption is becoming the real test of governance.

Customers do not experience governance through the number of policies that have been written, or the percentage of the estate that has been cataloged. Those things matter, and the organizations that have invested in them are in a better position than those that have not. But nobody feels a policy document. People feel governance when they try to use data and something either works smoothly or quietly stops them without explaining why.

That is a high bar, and it is not reached by writing more rules. It is reached by making sure the controls and the meaning are close enough to the data that they can be applied in the moment rather than assumed in advance.

This becomes even more important once AI becomes a consumer of enterprise data rather than simply another application sitting on top of it. Gartner’s attention on AI governance, on agentic data management and on data streaming for agents describes the same world I hear customers describing, where data is being consumed continuously by systems rather than periodically by people. Gartner’s discussion of GraphRAG and the growing importance of semantics supports the same point. Retrieval works better when the relationships and meaning between things are explicit, not inferred.

I keep coming back to one consequence of that. An AI agent cannot be expected to rely on organizational folklore.

Humans compensate for gaps in enterprise information constantly, and we barely notice we are doing it. Someone knows which revenue definition applies to this particular report and which one the finance team uses. Someone knows which of the three similar sources is the authoritative one. Someone knows why a field is sensitive and which customers it should never be shown against. Someone knows that a calculation has an exception for one region. Someone knows what a number actually means in the business, as opposed to what its column name suggests.

Much of that knowledge has never been formally encoded anywhere. It lives in experience, in habit and in the people who have been around long enough to remember. AI cannot draw on that. It cannot ask the colleague two desks away why the figure looks strange this quarter.

This challenge is becoming increasingly visible across the industry. McKinsey, writing about the foundations required for agentic AI at scale, describes the semantic layer as a way of turning enterprise data into machine-readable business knowledge. The underlying problem is important: without shared business meaning, agents can encounter incomplete or conflicting interpretations of the same information.

So that knowledge has to become usable with the data. Business context must travel. Definitions must travel. Permissions must travel. Lineage must travel. Sensitivity must travel. Governance must travel.

If context and governance have to travel, something has to carry them. That is the practical reason data products matter, and it is a more useful reason than the architectural one.

A well-designed data product brings together the data required for a particular business need along with everything needed to use it responsibly. That includes its business purpose, the context around it, the meaning of the measures inside it, who owns it, how it is governed, and what use of it is permitted. It is the difference between handing someone a table and handing them something they can act on.

Instead of saying “here is some data,” the organization can say “here is trusted information designed to support this decision.”

Reuse is where this becomes genuinely valuable. Establishing shared understanding of what a measure means, which source is authoritative and which controls apply is real work. It involves the business, the data team and often the risk or privacy function. Having done that work once, an organization should not have to reconstruct it every time a different person, application or agent needs the same information. That still happens repeatedly across enterprises, and it is one of the quieter reasons data teams can feel permanently behind.

McKinsey has made a related argument about reuse, describing scalable data products as assets designed to support multiple business needs rather than being repeatedly rebuilt around individual use cases. From the customer side, I would take that one step further. Reuse should not depend on moving everything into the same place first. The data an organization needs may continue to live across different platforms and systems. What needs to become reusable is the trusted business understanding around it: what the data means, how it can be used, which controls apply and the decisions it is designed to support.

That is the principle behind Latttice. It works with the data and technology investments an organization already has, allowing business-led data products to bring together the information needed for a decision with the context and governance required to use it confidently. The underlying data can remain across the platforms and systems where it already lives. Latttice is not another destination everything has to be moved into. It provides a way to make trusted, governed business understanding reusable across people, applications and AI.

That distinction matters. Organizations have already invested heavily in data platforms, warehouses, lakehouses, catalogs, governance technologies and integration. The opportunity is not necessarily to move everything again. It is to make those investments more useful at the point where somebody, or increasingly something, needs to make a decision.

Enterprise data programs have a habit of beginning with enormous ambitions. Govern everything. Catalog everything. Clean everything. Move everything to the new platform. Transform the operating model. Then, somewhere on the far side of all that, deliver business value.

I understand why programs are shaped that way, and the intent behind them is usually right. But from where I sit, working with customers through the reality of delivery, the sequence tends to be backward. Value proves the approach, and value comes from decisions.

So I would start with one decision that matters. It does not need to be strategic. It might be an everyday operational decision that people make dozens of times a week. It might be a regulatory decision where the cost of getting it wrong is obvious. It might be an executive decision, a customer decision, a supply chain decision, or the first AI agent use case somebody wants to put in front of real users.

Then work backward. Ask what must be known to make that decision well. Identify what data is required and what business context has to come with it. Determine which governance must apply, and where. Create the reusable data product that carries all of it. Allow people, applications and AI to consume it appropriately. Then measure whether the decision actually improved, and feed what you learn back into the product.

Taking one decision all the way through to a trusted outcome can reveal things about practical readiness that months of abstract planning never will.

Strip away the vocabulary and the questions customers ask are consistent. Can I trust this decision? Can I understand how we reached it? Can I explain it to someone who was not in the room? Can I reproduce it in three months when somebody challenges it? Can the appropriate controls apply without slowing everybody down?

And then there is the question that has appeared in the last two years and now sits behind almost every AI conversation I have. Can I trust an AI system to act on our data when I am not sitting there supervising every interaction?

That question changes governance in a way the earlier ones did not. A person works in sessions and asks for help when something looks wrong. Agents operate continuously. They may reach across systems that were never designed to be read together. They may combine sources in ways nobody anticipated. They may reason across multiple sources and draw conclusions that are harder to trace back to a single query or system. They may initiate actions rather than simply produce an answer.

This is also why the governance conversation is moving beyond policy alone. The World Economic Forum's 2026 playbook for trusted AI agent adoption focuses on the conditions under which agents are authorized to act and the need for that authority to remain enforceable as systems operate and evolve. That matters because once an AI system can act, governance has to influence what happens during the interaction, not simply describe what should have happened beforehand.

The same principle appears in NIST's AI Risk Management Framework from a risk-management perspective. Trustworthiness does not stop when an AI system is deployed; NIST considers it across the AI lifecycle, including deployment, use and evaluation, while its Generative AI Profile extends that thinking into areas including provenance and ongoing monitoring.

Governance written for periodic human access is not sufficient for that. It increasingly has to become part of the interaction itself, applied when the information is requested rather than assumed to have been handled somewhere earlier.

When I look at what is changing across enterprise data and AI, I see a much larger change in the relationship between data, governance, AI and decisions.

Data is moving faster. AI is consuming it faster. Agents are beginning to participate in decisions and actions. And the distance between accessing information and acting on it is getting smaller.

That changes what governance has to do.

Governance cannot simply describe what should happen somewhere upstream. It increasingly needs to be present when the data is actually used.

The context needs to travel with the data.

The permissions need to travel with the data.

The business meaning needs to travel with the data.

The evidence needs to travel with the data.

And the governance needs to travel with the data.

That is why Gartner’s 2026 trends resonated with me. They point toward many of the same pressures customers are already describing from the other direction.

And the more decisions we entrust to AI, the less we can rely on people filling in the gaps between what our systems know and what our business actually means.

That is also why I believe the shift from Data-Driven to Decision-Driven matters.

Do not start with the agent.

Do not even start with the data.

Start with a much more ordinary question. What decision are we trying to improve?

Then work backward from there.

The following research and publications informed the thinking behind this article and provide additional perspective on decision governance, Active Governance and trusted data products.

1. Gartner. “Gartner Identifies the Top Trends for Data and Analytics.” 16 June 2026. Available at: Gartner Identifies the Top Trends for Data and Analytics
2. Gartner. “Signature Series: Top Trends in Data and Analytics for 2026.” Gartner Data & Analytics Summit, Sydney.
3. McKinsey & Company. “Building the foundations for agentic AI at scale.” Available at: mckinsey.com
4. McKinsey & Company. “The missing data link: Five practical lessons to scale your data products.” Available at: mckinsey.com
5. World Economic Forum. “AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling.” 2026. Available at: weforum.org
6. National Institute of Standards and Technology. “Artificial Intelligence Risk Management Framework (AI RMF).” Available at: nist.gov
7. National Institute of Standards and Technology. “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.” Available at: nist.gov
8. Cameron Price, Data Tiles. Active Data Governance — Part 3: From Knowing Data Exists to Using It
9. Cameron Price, Data Tiles. From Data-Driven to Decision-Driven: Why Data Products Are Becoming the New Operating Model for Data