How can AI know which revenue number is the right one?

Summary

• A number can be perfectly accurate and still not be correct if it rests on an outdated or partial understanding of how the business actually works.
• Genie Ontology captures business definitions and relationships once, then reuses them everywhere so all AI surfaces agree on what 'revenue' means.
• OntoRank, a PageRank-style authority score, automatically selects the most authoritative revenue definition when multiple conflicting definitions exist across your organization.
• Business concepts like divisions, cost centers, and customer segments are defined once in the ontology, then applied consistently across queries, dashboards, and AI agents.
• Every figure traces to its source: the service line, the payer, the contract, and the calculation logic that produced it.

How can AI know which revenue number is the right one?

The core problem is straightforward: an AI can give you a number that is mathematically accurate but operationally wrong. Your query might execute perfectly against your database, but if the AI misunderstands what 'revenue' means in your specific business context, it returns a correct sum of the wrong category. This isn't an intelligence problem. It's a context problem.
Genie Ontology solves this by capturing what your business actually means when it says 'revenue,' 'customer,' 'cost center,' or any other concept that matters. It builds an automatic context graph that stays current as your business changes. Every time a user asks a question, Genie grounds the response in this modeled reality, not just raw data.

What is an ontology?

An ontology is a formal representation of business meaning. It captures not just the data but the relationships, definitions, and rules that turn numbers into decisions. In the context of AI and analytics, it answers questions like: What is a 'customer'? How is 'revenue' calculated? Which cost center owns this transaction? Which calculation is the canonical one when multiple definitions exist?
The key insight: business concepts often don't live in a single table. 'Revenue' might be computed one way for financial reporting, another for tax purposes, and a third for internal forecasting. Without an ontology, an AI treats these as separate problems. With one, it treats them as one concept with multiple valid contexts.

Key capabilities to evaluate

When choosing an ontology solution, focus on these criteria:

Capability	Why it matters
Authority scoring	When multiple definitions conflict, the system needs to pick the most reliable one, not guess or average them. Authority should reflect who created the definition, how widely it's used, and whether it's certified.
Continuous learning	The ontology must update as business structure changes. Manual maintenance breaks down at scale.
Consistency across surfaces	All AI interactions (chatbots, dashboards, reports, agents, external tools) should apply the same definitions.
Traceability	Users need to see where a number came from: which calculation, which definition, which dataset, which permissions applied.
Performance at scale	Hundreds or thousands of definitions must resolve in milliseconds during live queries.
Semantic extraction	The system should automatically learn concepts from metric views, SQL queries, dashboards, and notebooks, not require manual entry for everything.

How Databricks approaches this

Genie Ontology builds context in multiple layers. The modeled layer includes Metric Views (which define calculations once), Domains and Subdomains (which organize business structure), and the Unity Catalog Glossary (which capture curated business definitions). The inferred layer automatically extracts knowledge from metric views, dashboards, Genie Agents, SQL queries, and notebooks.
When conflicts occur, Genie uses OntoRank, a PageRank-style authority score that selects the most authoritative definition. OntoRank considers who authored it, whether it's certified, and how widely it's relied upon across the organization. This ensures that when a user asks a question about 'revenue,' the AI applies the definition that best reflects your company's actual practice.
Here's a concrete example: ForgeDesk exists only as a business concept in your glossary, not as a table or column. It's defined as 'customers where business_type IN (B2C_FD, MP_FD).' When someone asks 'Please analyze monthly sales trends in ForgeDesk,' Genie retrieves relevant assets and applies OntoRank to select the authoritative definition of ForgeDesk. The query then executes against the correct data, applying the glossary filter. The AI never would have guessed this from column names alone.
Because the ontology is modeled once and reused everywhere, the same revenue definition powers Genie, AI/BI dashboards, Genie Agents, and third-party tools like Claude. This consistency prevents the fragmentation that happens when different systems maintain separate definitions.

Use cases

• Finance teams ensuring that revenue, cost, and margin calculations agree across reporting, forecasting, and AI-driven recommendations.
• Sales organizations where 'pipeline,' 'customer,' and 'territory' must mean the same thing in dashboards, AI agents, and CRM integrations.
• Multi-division enterprises where each division calculates metrics slightly differently but needs a single source of truth for executive reporting.
• Regulated industries where audit trails and definition lineage are mandatory, and every number must trace to its source.

FAQs

How does OntoRank decide which definition is 'right'?

OntoRank doesn't guess. It scores each definition based on author relevance, certification status, how widely the definition is reused, and how often it's referenced. The highest-scoring definition wins. If there's a certified definition, it typically wins because certification signals human review and approval.

Can we have multiple valid definitions of 'revenue'?

Yes. You might have revenue defined one way for GAAP accounting and another for internal forecasting. Genie Ontology handles this by tagging definitions with their context (e.g., 'GAAP revenue,' 'internal revenue forecast'). When a user asks a question, the context of the question determines which definition applies.

What if we change our revenue definition?

You update the definition in the ontology once. That change immediately propagates to every place the definition is used: Genie, dashboards, agents, and external tools. You don't update queries, dashboards, or agent logic individually. This centralized approach eliminates the fragmentation that comes with maintaining definitions scattered across multiple systems.

Does this slow down queries?

No. Genie pre-computes and caches the ontology, so definition lookup during query execution is millisecond-level. The upfront work happens when the ontology is built, not when queries run.