What Is a Semantic Layer for Finance Data? Why ERP Data Alone Isn't One

Ask the controllerGlossaryControllershipControllership is the function responsible for accounting accuracy, the close, internal controls, and compliance. For controllership, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →, FP&AGlossaryFinancial planning and analysisFinancial planning and analysis is the finance function responsible for budgeting, forecasting, analysis, and decision support. For financial planning and analysis, the useful boundary is the driver, assumption, source data, owner, time period, scenario logic, and decision the model is meant to support.Open full definition → lead, and data lead the same question: How much revenue did we have in June? In one illustrative diagnostic, the ledger says $1.20 million, billing $1.31 million, and the CRM $1.44 million. All three can be correct.

A finance semantic layerGlossarySemantic finance layerSemantic finance layer is a business-friendly modeling layer that defines finance metrics, entities, and relationships once over underlying data sources. For semantic finance layer, a useful definition states a business-friendly modeling layer that defines finance metrics, entities, and relationships once over underlying data sources, the source data, owner, timing, evidence, and decision it supports before teams rely on it.Open full definition → defines and reuses one governed meaning across every report, spreadsheet, API, and AI system. It specifies the metric, grain, joins, period, entity scope, currency, exclusions, access, and owner. ERP data records transactions; it does not automatically settle cross-system meaning.

The $1.20 million is recognized revenue in a closed accounting periodGlossaryPosting periodPosting period is the defined time window to which transactions are posted, and the rules for opening or closing it. For posting period, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →. The $1.31 million is invoiced value. The $1.44 million is a run rateGlossaryRun rateRun rate is the practice of annualizing recent performance to estimate a full-year figure, and its limitations. For run rate, the useful boundary is whether the movement comes from customers, contracts, billing, cash timing, recognition rules, churn, expansion, pricing, or usage behavior.Open full definition → from $120,000 of monthly recurring revenueGlossaryMonthly Recurring RevenueMonthly recurring revenue is the normalized monthly value of committed recurring subscription revenue, usually tracked by new, expansion, contraction, and churn movements. For monthly recurring revenue, the useful boundary is whether the movement comes from customers, contracts, billing, cash timing, or recognition rules.Open full definition →. One noun, three defensible measures, and a meeting already heading sideways.

What is a semantic layer for finance data?

A semantic layerGlossarySemantic layerSemantic layer is a business-friendly abstraction above the warehouse that defines metrics, dimensions, joins, and access rules once so all downstream tools compute the same numbers. For semantic layer, a useful definition states a business-friendly abstraction above the warehouse that defines metrics, dimensions, joins, and access rules once so all downstream, who owns it, and which decision it supports.Open full definition → for finance data is an executable contract that translates physical records into business measures. It names what a number means, where it comes from, how it aggregates, which dimensions may filter it, who may see it, and which definition version applied at the time.

Microsoft describes a Power BI semantic model as a logical description of an analytical domain with metrics, business-friendly terminology, facts, and dimensions. dbt's current Semantic Layer documentation centers metric definitions above existing models and handles joins so downstream tools use the same logic. Snowflake's semantic views store business entities, relationships, dimensions, facts, and metrics as metadata in the database. Across all three products, reusable business definitions sit between physical data and the tools that consume it.

A glossary entry can explain a term. A semantic layer goes further: it can produce the same governed calculation when Excel, a dashboard, or an agent asks for it. Documentation without executable logic still leaves every consumer free to rebuild the number.

Why is an ERP not automatically a semantic layer?

An ERP is often an authoritative systemGlossarySystem of recordSystem of record is the application designated as the authoritative source for a given type of data, such as the general ledger for accounting or the CRM for customer records. For system of record, a useful definition states the application designated as the authoritative source for a given type of data, such as the general ledger for accounting or the.Open full definition → of record for financial transactions, books, posting periods, accounts, and reporting dimensions. That is indispensable. But a transaction system does not automatically settle how finance concepts should work across every consuming tool or source.

Oracle's own NetSuite analytics-source documentation describes inconsistent field names and exposure across some historical Saved Search and Report surfaces, while the newer Workbook data source aims to provide consistent fields inside Workbook. Reproducing an older search can still require different records, joins, or custom formulas. NetSuite includes analytics. Yet one database does not guarantee one reusable meaning across surfaces.

Even inside a single ERP, the consuming surface matters. NetSuite's dataset-linking guidance shows separate datasets joined through common fields such as sales representative and accounting period. Oracle warns that one-to-many links can duplicate results, common keys need compatible values and data types, and aggregation occurs at the visualization layer. An error in grain or cardinality changes the reported amount, so finance should review those rules as carefully as an account mapping.

A data warehouseGlossaryData warehouseData warehouse is a central, structured, schema-on-write store optimized for analytical queries and reporting across data consolidated from many source systems. For data warehouse, a useful definition states a central, structured, schema-on-write store optimized for analytical queries and reporting across data consolidated from many source systems, the source data, owner, timing, evidence, and decision it supports before teams rely on.Open full definition → centralizes prepared tables and history; it does not by itself establish reusable finance meaning. Microsoft recommends a warehouse and ETL process before a Power BI semantic model when source preparation becomes complex. The warehouse governs where prepared data lives. The semantic layer governs what a reusable business number means.

Why can the same finance question produce three correct numbers?

The same finance question can produce several correct numbers when systems measure different events, use different clocks, or apply different recognition rules. In the illustrative June diagnostic, the ledger measures posted recognized revenue, billing measures invoices, and the CRM annualizes recurring revenueGlossaryRecurring revenueRecurring revenue is revenue a business can reasonably expect to continue at a predictable cadence, and how it differs from one-time and variable revenue. For recurring revenue, the useful boundary is whether the movement comes from customers, contracts, billing, cash timing, or recognition rules.Open full definition →. The intended decision determines which measure belongs in the answer.

The semantic layer does not choose one universal winner. It gives each measure a precise name, formula, owner, and valid use. If the official question is 'recognized revenue for the June management P&LGlossaryIncome statementIncome statement is the statement reporting revenue, expenses, and profit over a period of time. For income statement, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →,' the layer selects the posted GL measure, the June posting period, approved entities and books, and the reporting-currency rule. If the question is 'current ARRGlossaryAnnual Recurring RevenueAnnual recurring revenue is the annualized value of committed recurring subscription revenue, excluding one-time fees and non-recurring services. For annual recurring revenue, the useful boundary is whether the movement comes from customers, contracts, billing, cash timing, recognition rules, churn, expansion, pricing, or usage behavior.Open full definition →,' it uses a different contract.

Currency alone can change the answer. NetSuite's consolidated exchange-rate documentation assigns Average rates by default to income-statement accounts, Current rates to most balance-sheet accounts, and Historical rates to equity accounts. Oracle's reporting guidance explains that the selected subsidiary context determines reporting currencyGlossaryReporting currencyReporting currency is the currency in which consolidated group financial statements are presented, into which each entity is translated. For reporting currency, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →. A field labeled amount is not a complete metric until the currency basis and entity scope are known.

Time rules can also change a finance answer. Finance may need posting period, transaction date, service period, invoice date, CRM close date, or cash date. A semantic contract declares the clock instead of letting the dashboard designer pick the nearest date column.

What belongs in a finance metric contract?

A finance metric contract should specify the metric question, owner, sources, grain, joins, formula, scope, time, entity, currency, dimensions, access, lineage, tests, refresh behavior, version, and valid uses. The 20-field Finance Metric Contract below is an editorial framework, not an industry standard. Every blank field forces a downstream analyst, report, or model to guess.

Start with revenue if it causes debate. Ask finance and data teams to complete the contract independently, then compare account scope, period, entity hierarchyGlossaryEntity hierarchyEntity hierarchy is the ownership structure and thresholds that determine whether an investment is consolidated, equity-method, or held at cost/fair value. For entity hierarchy, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →, currency basis, eliminations, and exclusions. Every disagreement is a missing semantic decision. No benchmark is required. The blank cells are the finding.

How do you build a finance semantic layer without boiling the ledger?

Build the first finance semantic layer around one recurring decision, not the entire chart of accountsGlossaryChart of accountsChart of accounts is the structured, numbered list of every account a company uses to record transactions. For chart of accounts, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →. A bounded metric exposes the hard parts while the blast radius remains small.

The AI-ready finance data checklist is the prerequisite for source scope, periods, mappings, dimensions, and ownership. A semantic layer cannot repair an incorrect posting, a missing subsidiary, or an account mapping nobody will approve. With explicit tests and failure rules, it can make those gaps fail loudly.

Keep the first test boring. Remove a mapping. Feed a duplicate accounting line. Change the reporting subsidiary. Reopen the period with a late journal. A governed metricGlossaryCanonical metricCanonical metric is a single governed definition of a business metric that serves as the authoritative way it is calculated everywhere it appears. For canonical metric, a useful definition states a single governed definition of a business metric that serves as the authoritative way it is calculated everywhere, who owns it, and which decision it supports.Open full definition → should stop, identify the affected result, and name the owner. A number that survives only the happy path is a demo.

How is a semantic layer different from an ontology?

A semantic layer makes measures executable across data and tools. An ontologyGlossaryOntologyOntology is a formal model of the concepts, entities, and relationships in a domain, including hierarchies and constraints that support inference, distinct from the analytic semantic layer. For ontology, a useful definition states a formal model of the concepts, entities, and relationships in a domain, including hierarchies and constraints that support, who owns it, and which decision it supports.Open full definition → models business concepts and their relationships beyond a single metric: customers sign contracts, contracts govern invoices, invoices create receivablesGlossaryAccounts receivableAccounts receivable represents amounts customers owe a company for goods and services delivered but not yet collected. For accounts receivable, the useful boundary is the source cash view, timing horizon, owner, liquidity exposure, and operating decision before payment timing, runway, or financing options change.Open full definition →, and subsidiaries own the transactions. The two can overlap, but they solve different parts of the meaning problem.

The Financial Industry Business Ontology formalizes financial concepts and relationships in machine-readable OWL. FIBO describes itself as a financial-services ontology; it is not presented as an FP&A metric catalog. For a mid-market finance team, the practical distinction is simpler: the ontology is the map of what exists and connects; the semantic layer is the governed calculation and query interface that uses the map.

Pluvo's published ontology essay goes deeper on the entity-and-relationship layer. Here the boundary is narrower: how a finance number gets one reusable definition.

What can AI not do, and what must stay human?

AI should not hold authority to choose the official definition of revenue, approve a new account mapping, accept a broken tie-out, set FX policy, or waive a late journal as immaterial. Those are finance decisions with owners and consequences.

An AI system can query an approved semantic layer, assemble source evidence, explain a governed result, and flag a contract failure. The calculation still belongs in deterministic logic. The two-system control pattern keeps language generation away from the arithmetic and preserves a source-to-answer path.

Human review should focus on changes in meaning: a new account, revised entity structure, exception to the recognition rule, altered materialityGlossaryMaterialityMateriality is the threshold above which an omission or misstatement could change a user's economic decisions. For materiality, the important details are the accounting period, source evidence, reviewer, materiality threshold, and control purpose that make the treatment auditable during close, reporting, and later review.Open full definition →, or a source failure. Automated tests can clear stable definitions and routine source checks, while exceptions route to the people authorized to decide them.

How does Pluvo use finance semantics?

The 20-field Finance Metric Contract above is Pluvo's editorial framework. Pluvo's current ontology models entities, relationships, and definitions from connected finance and operating systems. Definitions are versioned and governed; the history remains when a definition changes. Lineage connects a reported figure to source records, definitions, and transformations. Those are product claims documented on Pluvo's public pages, not proof that every implementation begins clean. Finance still reviews the definitions and the source data must still reconcile.

For practical field notes on Finance EngineeringGlossaryFinance engineeringFinance engineering is the discipline of designing finance workflows, data models, automation, controls, and AI interfaces as an integrated operating system. For finance engineering, the useful boundary is the data it uses, the tools it can call, the approvals it needs, the review standard, and the finance decision it may influence before the output is trusted or automated.Open full definition →, subscribe to the Finance Engineering Newsletter.

The next time three systems answer revenue, do not ask which one is lying. Ask which question each was built to answer.

About the author

Vanessa Galarneau

CFO & COO

LinkedIn →