Finance said revenue was down.

Marketing said revenue was up.

Product said it was basically flat.

Three teams, three dashboards, three confident stories. Nobody was fabricating numbers. Nobody was incompetent. They were all measuring different versions of "revenue" and arguing as if the disagreement were computational instead of definitional.

That is consistency.

Consistency asks whether the same concept is represented and calculated the same way across systems, reports, and time. If one team uses gross revenue, another excludes refunds, and a third reports recognized revenue after a delay, the numbers can all be internally correct and still collectively useless.

This is one reason the LLM era should make analysts stricter, not looser. AI can generate queries fast. It can also generate three perfectly coherent queries for three subtly different definitions and leave a team arguing over outputs that were never aligned at the input level.

"Revenue" is not a metric until someone defines it

People speak about metrics as if their names contain their meaning.

They do not.

Revenue might mean:

  • Gross order value before discounts
  • Net collected amount after discounts
  • Recognized revenue after accounting rules
  • Revenue excluding tax and shipping
  • Revenue excluding refunded orders

All of these can appear in a dashboard labeled simply Revenue.

One sharp opinion here: most dashboard disputes are not data disputes. They are product-definition failures wearing the costume of analytics.

Gross, net, recognized, refunded, booked, and imaginary revenue

Businesses accumulate metric variants because each function has a legitimate local need. Finance wants one view. Growth wants another. Product wants a faster proxy. The problem starts when local definitions escape their context and become organizational assumptions.

Then everyone says "revenue" while quietly meaning different things.

The same pattern appears in conversion, active users, churn, CAC, retention, and even customer counts. Filters vary. Time windows vary. Entity scopes vary. Refund handling varies. Internal traffic exclusions vary. Small differences compound into meetings that feel political but are actually structural.

The silent violence of inconsistent filters

Consistency problems are often not dramatic. They are quiet.

One dashboard excludes test orders. Another does not.

One uses local timezone. Another uses UTC.

One counts only completed purchases. Another includes pending authorization.

One filters to attributed sessions. Another includes all sessions.

Individually, these can seem minor. Together, they create a stable environment for endless disagreement. Every team gets a dashboard that confirms its worldview because each dashboard is built from a slightly different universe.

Metric definitions should be products, not tribal knowledge

Many organizations treat metric definitions as documentation debt. That is backwards. A definition is part of the analytical product.

If a KPI drives decisions, then its definition should be versioned, owned, reviewed, and discoverable. Not hidden in one analyst's notebook. Not preserved only in the memory of the person who built the original Looker tile two years ago.

Single source of truth is a nice slogan, but it is less magical than people think. A warehouse can centralize data while still centralizing ambiguity. Truth does not emerge from storage location. It emerges from explicit agreement.

How analysts usually miss it

They assume shared terminology implies shared logic.

That assumption is fatal in cross-functional work.

Analysts also inherit old business logic without checking whether it still matches current practice. A revenue model that was sensible before subscriptions, refunds, or multi-currency expansion may become misleading later while retaining its authoritative aura.

Consistency failures are especially hard to spot because each individual report can look polished and stable. The contradiction appears only when compared with another report built under different assumptions.

A practical metric contract

If a metric matters, define it like something real.

Metric: Revenue

Business definition: Net collected value from completed orders, excluding tax, shipping, and fully refunded transactions.

Grain: One row per settled order.

Source tables: orders, payments, refunds

Inclusions: Completed and settled payments within reporting window.

Exclusions: Test orders, pending payments, canceled orders, internal employee purchases.

Timezone: UTC for storage, converted to business timezone in reporting layer.

Owner: Finance Analytics

Change policy: Any definition change requires version note and stakeholder signoff.

That is not bureaucracy for its own sake. It is the minimum structure required to stop organizations from debating words that were never stabilized.

What consistency breaks in business terms

Inconsistent metrics slow decisions because every planning conversation becomes a reconciliation exercise. Leaders stop trusting dashboards. Teams defend their own numbers instead of solving the business problem. Confidence shifts from evidence to political credibility.

The worst part is that everyone can be acting in good faith while the system keeps producing conflict.

Suppose the teams finally align on the definition. There is still another failure mode waiting: the right number may arrive too late to matter. That is timeliness.