Boards and executives increasingly expect AI to accelerate analysis and support faster decisions, while investment across the AI market continues to climb. In its September 2026 forecast, Gartner projected worldwide AI spending of nearly $2.7 trillion for 2026. But the push toward AI-driven decisions runs into a much older question: which numbers should those decisions rest on?
What counts as revenue? Who qualifies as an active customer? When has a customer churned? Which definition of margin should guide a pricing decision? Ask finance, sales, and operations, and the answers may differ, often for good reason.
This issue isn’t necessarily bad data. Different teams use different definitions because they serve different purposes. Problems emerge when an AI system is expected to reason across these definitions without knowing which one applies. AI doesn’t make those disagreements disappear. It makes them harder to ignore.
The Metric Trust Crisis Predates AI
Metric inconsistency isn’t new. Business units built separate definitions for metrics like gross profit and churn long before AI entered the picture. A salesperson may count a customer at contract signature, while finance may count a billable customer only once invoicing begins. Both definitions are legitimate. They were simply built for different jobs.
"AI did not create the metrics consistency problem. It exposed it," said Chris Lynch, CEO of AtScale, a semantic layer platform for AI and connected tools. "For years, companies got away with different numbers in dashboards, spreadsheets, and reports because humans could reconcile the gaps. Agents won't get the benefit of the doubt."
Over time, the gaps widen. Different BI tools encode different calculations. Acquisitions bring overlapping systems and competing definitions. A definition changes in one department and never gets updated elsewhere. Gradually, metric drift becomes a structural issue rather than an occasional discrepancy.
AI Turns Metric Disagreement Into an Automation Problem
Traditional analytics generally keeps a human in the loop. When a dashboard shows two conflicting numbers, someone asks why, and investigates the discrepancy before making a decision.
AI changes that dynamic. AI systems and agents are increasingly expected to:
Answer business questions directly
Compare performance across teams or periods
Recommend actions based on that comparison
Generate forecasts from historical data
Trigger downstream workflows with less manual review
Each of those tasks requires the system to select a definition, whether or not the organization has agreed on one.
The result can be an answer that is technically defensible but contextually wrong for the question being asked. That isn’t necessarily a hallucination. The AI may be working with perfectly valid data. The problem is that the organization has multiple versions of what that data means.
AI Is Forcing Enterprises to Confront Their Semantic Debt
Semantic debt is the accumulated cost of allowing business definitions and logic to fragment across an organization. It builds quietly. A metric gets defined once, then redefined by another team solving a different problem. Calculations live inside spreadsheets or dashboard formulas that no one documents, and knowledge about which number to trust exists only in the heads of a few analysts.
The result is duplicate metrics, conflicting definitions, and business logic embedded separately across BI tools. Humans have absorbed that debt for years by relying on context and informal knowledge. AI systems may lack access to that informal knowledge, making unresolved ambiguity more visible. Although they can sometimes flag conflicting definitions or request clarification, they still need explicit business context to apply the appropriate metric.
“When a finance system and a marketing platform use different logic to calculate revenue, a human analyst catches the discrepancy,” says Lynch. “An AI agent doesn't. It returns a confidently wrong answer.”
As AI becomes embedded in daily workflows, semantic ambiguity is no longer just an analytics inconvenience. It becomes a critical operational risk.
Shared Meaning Becomes Infrastructure
Addressing semantic debt requires a way to define metrics, calculations, and business terminology once. These definitions, along with the underlying business logic and relationships, can then be referenced consistently across systems. That is where a semantic layer enters the conversation. It provides a centralized, governed foundation for defining and managing business logic, helping support consistent interpretation of data across AI and BI systems.
Rather than requiring every dashboard, application, or AI agent to interpret raw enterprise data independently, a semantic layer provides a governed representation of business meaning that multiple systems reference. When the logic lives in one place instead of being rebuilt inside each tool, revenue can be calculated the same way whether the request comes from a BI dashboard or an AI agent.
This distinction is important because a semantic layer doesn’t decide what "active customer" or "margin" should mean, and it doesn’t resolve organizational disagreements on its own. Humans still make that call. What it can do is encode those decisions so they don’t need to be reinterpreted or re-litigated each time a new system asks the same question.
A semantic layer can improve consistency, but its effectiveness depends on accurate definitions, maintained mappings and ongoing testing; it does not eliminate AI errors.
Companies such as AtScale are extending semantic-layer architectures into AI environments, reflecting a broader push to make governed business definitions available not only to AI applications and agents, but also to BI tools.
Governance Starts Before the AI Makes a Decision
AI governance conversations often center on model controls, human oversight, and monitoring. Those safeguards remain important. But they can implicitly assume that the metrics feeding an AI system are consistent and appropriately defined.
The US National Institute of Standards and Technology’s AI Risk Management Framework similarly links trustworthy AI to data, system design and human oversight. Shared definitions are one part of that wider governance responsibility.
“I see a convergence of data, analytics, and AI rapidly approaching,” Lynch says. “Security, trust, and governance will be equally important in this convergence.”
Consider an AI system that recommends a price change based on margin. Useful governance questions arise well before anyone reviews the output. Which definition of margin was used? What data supported it? Which calculation logic was applied?
Data lineage, the record of where a value came from and how it was produced, can help answer those questions. One question still requires human judgment: was that definition appropriate for this decision? Tracing and evaluating the definitions behind an AI-generated recommendation turns metric governance into part of AI governance.
AI Can't Settle a Debate the Business Hasn't Settled
AI models may become faster, more capable, and more autonomous. But increasing model capability doesn’t automatically resolve disagreements embedded in enterprise data. Even highly capable models inherit whatever ambiguity exists in the metrics they interpret.
AI can calculate a metric, explain it, and increasingly act on it. It can help compare possible definitions, but responsibility for deciding what that metric should mean for the business remains with the organization. The enterprises best positioned to automate decisions are the ones that first do the less glamorous work of agreeing on what the numbers mean, then making those definitions explicit, governed, and reusable across the systems that depend on them.
Platforms like AtScale's universal semantic layer provide one architectural approach for making those agreed-upon definitions available consistently across analytics and AI systems. But the harder work happens earlier. Before enterprises hand decisions to AI, they may need to finish a conversation that has nothing to do with AI at all, but rather with agreeing on what the numbers mean.