Beyond the Semantic Layer

Oct 23, 2026

Your data has a shared vocabulary. Do your decisions?

Enterprise AI is getting a shared vocabulary. Strategic initiatives also need shared logic: an encoded model of how the firm decides, authored by the practitioners who hold that judgment. That model turns expertise into an asset the firm can reuse, govern and scale.

The semantic layer arrives

In March 2026, Gartner predicted that by 2030 universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity. It told data and analytics leaders to budget for them as non-negotiable. The market had already moved. Microsoft introduced Fabric IQ, with an ontology at its center, in November 2025. Snowflake, Salesforce, dbt Labs and partners launched the Open Semantic Interchange to standardize business definitions across tools. Databricks built business semantics into its data catalog in April 2026, with a blunt warning: "fragmented definitions do not just create confusion, they scale it."

The logic holds. An AI agent that does not know what "revenue" or "active customer" means in your company will produce confident errors at machine speed. A semantic layer fixes the nouns.

Strategic work runs on verbs

Strategic initiatives rarely fail on the nouns. An acquisition, a transformation program or a joint venture goes wrong in the logic that connects its decisions: why a synergy target was set, which assumptions it rests on, what the chosen integration approach rules out, which workstreams a change in strategy should reach. A metric definition captures none of that.

Palantir, whose ontology is the most mature in the market, has moved its own framing in this direction. Its chief architect describes the ontology as designed "to represent the decisions in an enterprise, not simply the data."

The research points the same way. In a 2023 benchmark, GPT-4 answered enterprise questions correctly 16 percent of the time over raw database tables, and 54 percent of the time over a structured graph of the same data. Later studies add nuance: graph structure helps most on questions that connect several facts, and helps little on simple lookups. Strategic questions are almost always the first kind. Which integration risks touch which synergy assumptions, and who owns each one? That is a chain of reasoning, and a system that cannot hold the chain cannot answer the question.

Who authors the model

Here the two approaches part ways. An enterprise data ontology is built by modeling the organization's data, which means integration work and engineers before it returns value. That suits operations, where the data is the work.

In strategic work, the scarce input is judgment. It belongs to the experienced practitioner who knows that a capability acquisition should be integrated lightly, or that a cost program needs a different gate than a growth bet. That judgment can be encoded directly. A methodology model captures the categories an organization uses to frame an initiative, the rules that say which combinations of strategy and execution approach work, the deliverables each situation calls for, and the reference material that should ground every recommendation. The people who hold the expertise author it, and it delivers value from the first initiative it frames, with no data integration project in front of it.

Authorship carries risk. A wrong model can be worse than none, because it makes errors consistent. The governance questions are practical: who can change the model, how changes are reviewed and versioned, and whether every recommendation traces back to the rule or reference that produced it.

Built for change

There is a second reason strategic work needs a methodology model. Strategic initiatives exist to change the business. Every acquisition brings in new products, customers and operating assumptions. Every transformation redraws accountabilities. A model frozen at one point in time goes stale at exactly the moment it is needed most.

The value of reconfiguring quickly is already visible. At Palantir's AIPCon in September 2026, defense company Anduril described compressing post-merger integration from 12 to 24 months to about three. That is a customer claim at a vendor event, and it is unaudited. The direction still matters: in strategic work, speed increasingly comes from how fast the shared model can absorb change.

What this means for expertise

For leaders, an encoded methodology changes four things.

Time to value. Value starts with the first initiative the model frames. No multi-quarter data program has to come first.
Expertise that reaches past the expert. The firm's best judgment is applied on every initiative, including the ones its most senior people never see.
Consistency that adapts. Rules respond to the type of initiative in front of the team, so standardization does not collapse into a lowest-common-denominator playbook.
Methodology as an asset. For advisory firms and internal strategy functions alike, the method becomes versioned, governed and reusable. It can be improved, audited and licensed, and its value no longer rises and falls with billable hours.

Domain expertise, launch-ready

Supramodal is built around this idea. A practitioner's methodology becomes the platform's configuration: authored without code, versioned and governed. The platform supplies everything that stays constant from one domain to the next, including persistent records of decisions and their rationale, AI agents that work within the methodology, and governance and audit. Launching a new domain solution is a configuration event that practitioners lead.

Our first domain solution, purpose-built AI for M&A, is deployed. The same platform carries transformation programs, portfolio decisions and partnerships, because strategic decisions share one pattern even when their content differs completely. Each needs its rationale captured, its stakeholders aligned, its work scoped, its governance enforced and its outcomes measured.

The semantic layer gives AI a vocabulary. Methodology gives it judgment. Firms that encode how they decide will get more from every model that follows.

Sources

‹ Meaning Before Intelligence

Decision Debt ›