Revenue Definition Ambiguity Data Agents

Revenue has more definitions than any other business term, and AI agents without explicit glossary entries for each will produce contradictory numbers. Gross vs net, booked vs recognized, GAAP vs non-GAAP, recurring vs one-time — each is a distinct SQL template and each needs its own owner.

The single most common question a data agent gets is about revenue. It is also the question most likely to produce wrong answers because revenue has more valid definitions than any other metric. This guide covers the variants that exist and how to encode them in a glossary. Related: churn definition for AI data agents and AI for data infrastructure.

Variants You Will Encounter

• •Gross revenue — total billed before refunds or discounts
• •Net revenue — gross minus refunds, chargebacks, discounts
• •Booked revenue — contract value on signature date
• •Recognized revenue — GAAP revenue spread over delivery period
• •Collected revenue — cash actually received
• •Recurring revenue — subscription only, excludes one-time
• •Committed revenue — contracted but not yet delivered

Why the Difference Matters

A company with $100M in annual contracts has booked $100M, recognizes some fraction per month under GAAP, collects whatever customers actually paid, and reports different numbers depending on who is asking. Finance reports recognized; sales reports booked; cash flow reports collected. An agent that picks the wrong one produces a defensible-looking number that is still wrong for the audience.

Glossary Structure

Each variant gets its own glossary entry with a SQL template, an owner, a changelog, and a test suite. The templates point at specific source tables and columns, respect fiscal calendars, and apply the correct filters for the definition. When a new variant becomes relevant, someone owns its creation and maintenance.

The entries are code, not Confluence. They live in the same repo as your dbt models, get tested in CI, and get versioned so changes are auditable. When finance updates revenue recognition policy, the glossary entry gets a PR, tests run, reviewers approve, and every downstream agent sees the new definition immediately.

Disambiguation at Query Time

When a user asks about revenue, the agent checks the glossary and finds seven entries. If the scope (finance team) or the question phrasing (quarterly revenue for earnings) implies a specific variant, the agent uses it and surfaces its choice in the answer. If ambiguous, the agent asks.

Scoping is the most powerful signal. A user from finance almost always means recognized revenue; a user from sales almost always means booked; a user from ops almost always means collected. Default based on scope, ask when scope is missing or weird.

Refunds and Edge Cases

Refunds are the most common edge case. Gross revenue includes them; net revenue excludes them; recognized revenue treats them as reversals. The glossary entry must document refund treatment explicitly. Same for discounts, credits, tax, and currency conversion.

Currency conversion is particularly tricky. Some companies report in USD at contract date rate; others at reporting period end rate; others at trailing 30-day average. The glossary has to pick one and stick to it, or report multiple numbers with the conversion method surfaced.

Testing Revenue Definitions

Every revenue definition must have a test: run the template against a known period and verify the output matches a trusted dashboard. When the template changes, the test catches regressions. When the warehouse changes upstream, the test still catches regressions. Without tests, revenue definitions drift silently and trust erodes fast.

Common Mistakes

The biggest mistake is a single revenue entry in the glossary. The second is not testing templates against dashboards. The third is hardcoding currency rules without documenting them. The fourth is letting agents convert between variants silently — monthly booked to annual recognized is not a valid conversion and should fail loudly.

Data Workers builds the glossary agent to treat every revenue variant as a separate entry with its own owner, template, tests, and scope. Agents pick the right one per question, surface the choice in the answer, and ask when ambiguous. To see it on your warehouse, book a demo.

What To Do When Finance Changes Policy

Finance policy changes routinely. New revenue recognition rules ship every few quarters. New subsidiaries get added. Currency conversion methods change. Each change has to flow through the glossary entries, the SQL templates, and the tests. If any link in the chain is missed, the agent starts producing numbers that no longer match the official reports.

The fix is a change-management process owned by finance but implemented in code. When finance changes policy, a pull request updates the relevant glossary entries, the SQL templates, and the tests. Reviewers from finance and data engineering both approve. Once merged, every downstream agent picks up the new definitions on the next context refresh.

This process turns definition changes from silent drift into auditable events. Auditors can trace every change back to its approval. Users see a changelog of what changed and when. Data Workers versions every glossary entry with full history so rollback and audit are trivial.

The Earnings-Prep Use Case

The highest-value use of a well-curated revenue glossary is earnings prep. Public company finance teams assemble hundreds of numbers every quarter for the earnings release and 10-Q. Manual assembly is slow and error-prone; agent-powered assembly using glossary-grounded templates is fast and auditable. Every number in the release traces back to a template that is tested and versioned.

The auditability is what makes this practical for public companies. SOX compliance requires a paper trail for every material number. The glossary plus CI tests plus agent traces produce exactly that paper trail. Auditors reviewing quarterly numbers can follow every one back to its source in minutes.

Data Workers builds for this use case with full audit trails, version history, and reproducibility. Teams using it for earnings prep save dozens of hours per quarter and catch errors earlier. The ROI on the glossary investment shows up every three months like clockwork.

Revenue is not one number — it is seven. Put each variant in the glossary with its own owner and tests, and your agents stop contradicting the earnings report.