I used to think metric definition work was mainly about cleaning up dashboards.

Different reports showing different numbers. Different teams using slightly different definitions. Different leaders asking, “Why does this number not match that number?” It felt like a classic analytics hygiene problem, important, sometimes painful, but mostly about reporting consistency.

But as I have started thinking more seriously about AI agents sitting on top of enterprise data, I am realizing that definition work is much more strategic than that.

It is how the business teaches AI what the business means.

That is a pretty big shift.

The semantic layer is usually discussed as a technical architecture problem. And to be fair, there is a very real technical component. Data needs to be modeled well. Relationships need to be defined. Metrics need to be calculated consistently. The warehouse, transformation layer, BI layer, and AI layer all need to connect in a thoughtful way.

But in enterprise AI, I am increasingly convinced that the harder challenge may be organizational.

Who decides what important business concepts mean? Where are definitions allowed to vary? When is local customization legitimate, and when does it create unnecessary fragmentation? How do those meanings evolve as the business changes?

That, in my view, is the role of the Business Semantic Council.

Take something as simple as pipeline.

One team may define pipeline as every open opportunity. Another may only count sales-qualified opportunities. Finance may care most about forecastable pipeline. A business unit leader may exclude renewals to understand new-logo demand. A channel team may want partner-sourced opportunities included. A sales leader may want current territory assignment, while another analysis may require historical territory assignment.

None of these definitions are necessarily wrong.

But if an AI agent answers, “Why is pipeline down in this segment?” without knowing which version of pipeline applies, it can give a confident answer to the wrong question.

That is the risk.

In the dashboard, insights, and analytics era, ambiguity due to inconsistent definitions created friction. Meetings slowed down. Numbers had to be reconciled. Analytics teams explained the differences. Sometimes the issue got fixed. Sometimes it became tribal knowledge.

In the AI era, ambiguity does not always look like confusion. Sometimes it shows up as confidence.

AI agents will not just retrieve reports. They will answer questions, explain trends, summarize performance, recommend actions, and eventually trigger workflows. If the meaning underneath is unclear, AI does not magically resolve the ambiguity. It may simply make the ambiguity sound more polished.

Most organizations already have pockets of strong business meaning. The problem is that this meaning is scattered across people, reports, spreadsheets, SQL logic, meeting notes, and “the way we have always done it.”

AI cannot reliably reason over tribal knowledge unless the enterprise first decides what needs to become institutional knowledge.

That is why I like the idea of a Business Semantic Council.

In my current role, I am introducing a practical version of this through what we are calling a KPI Council, which may confirm that branding is not my strongest competency. But the more I work through it, the more I realize the broader idea is not really KPI governance. It is business semantics governance.

Not a dashboard committee. Not a reporting clean-up group. Not a theoretical governance body that produces a glossary nobody uses.

A Business Semantic Council is a cross-functional operating model for deciding, documenting, and evolving business meaning.

At a practical level, it should answer five questions:

1. What does this business concept actually mean?

2. Where does the enterprise need one standard definition?

3. Where is local variation legitimate?

4. Who owns the approved meaning?

5. How does that meaning get encoded into dashboards, analytics products, semantic layers, and AI agents?

The third question may be the most important.

The council does not need to boil the ocean. In fact, it probably should not. The goal is not to standardize every business definition across every part of the company. That is usually unrealistic, and sometimes not even desirable.

Different business units may need different definitions because their motions are genuinely different. A high-velocity SMB motion may not define pipeline the same way as an enterprise motion. A partner channel may not measure activity the same way as a direct sales team. A renewal motion may not treat opportunity stages the same way as a new-logo motion.

The goal is not forced sameness. The goal is governed clarity.

Sometimes the council’s job is to pick one enterprise definition. Other times, its job is to make variation intentional, documented, owned, and usable by the systems that depend on it.

That work cannot be owned by technology alone.

A technical semantic layer should not invent business meaning. It should encode business meaning.

That means the business has to show up first.

For me, this is the part that feels both obvious and under-discussed. We talk a lot about models, copilots, agents, warehouses, vector databases, governance, and security. All of that matters. But there is a more basic question sitting underneath many enterprise AI ambitions:

Have we made the business understandable enough for AI to reason over it?

If the answer is no, adoption gets harder. Users start asking, “Where did that answer come from?” or “Why is this different from my report?” Once trust is lost, even a technically impressive solution becomes difficult to scale.

This is why I believe the Business Semantic Council may become an important operating model for enterprise AI adoption.

Its job is not to make AI exciting. Its job is to make AI usable with less friction.

And because business meaning is not static, this cannot be a one-time exercise. Companies reorganize. Products change. Sales motions evolve. Segments are redefined. Channels mature. Incentive plans change.

Every time the business resolves an ambiguity, approves a definition, clarifies an exception, or validates a common analytical question, that learning should not disappear into a meeting. It should improve the system. It should make the next dashboard better, the next insight more trustworthy, and the next AI answer more grounded.

Enterprise AI adoption will not become easier just because the technology gets better.

It will become easier when the business becomes more legible to the technology.

And that starts before the semantic layer is built.

It starts with the semantic council.