Skip to content

Data

Why departments report different revenue numbers, and how to fix it

The meeting should be about the decision, not about whose number is right. Write each metric definition once and make every report read it.

Computing America. Applications, integrations, data analysis and AI/ML engineering.

Send the three figures
2 min readComputing America
Three printed bar charts on a conference table, the same bars at different heights on each.

In short

  • Have three people write down your most-quoted metric, independently. The differences are your project scope.
  • The uncomfortable step is the diff: collect every existing implementation of a metric and compare them before building anything.
  • Test a definition against a month finance has already closed. If the metric doesn't equal the closed figure, the definition is wrong or the books are.
  • Route every consumer through the new definition and delete the old implementations, or the new one is a fourth number.
  • Delete dashboards unopened for a full reporting cycle, and put the governed number inside the screen where the decision is made.

Departments often report different revenue because each one's queries define it differently: the definition of “revenue” lives in the WHERE clauses of many queries, written by different analysts with different assumptions about refunds, intercompany transfers and recognition timing.

A new BI tool pointed at the same queries doesn't help, because it inherits the same conflicting definitions.

Write each definition once, in code

A metric definition is business logic, so it gets what business logic gets: review, version control and tests, in a semantic layer, the service that holds metric definitions and answers every consumer's queries. Copies drift without anyone being careless, because each owner improves their own copy.

Fig. 1
  1. Group: Today: delete once replaced

    Saved query

    Its own refund rule: number one

  2. Group: Today: delete once replaced

    Spreadsheet formula

    Its own intercompany rule: number two

  3. Group: Today: delete once replaced

    Dashboard filter

    Its own recognition timing: number three

  4. Group: With one definition

    Net revenue definition

    Filters, joins, edge cases, one owner

    • To Semantic layer
  5. Group: With one definition

    Semantic layer

    One reviewed, tested commit

    • To Every consumer
  6. Group: With one definition

    Every consumer

    BI tool, API, finance export, spreadsheet

Three copies of a definition drift apart; kept once in a semantic layer, every consumer reads the same reviewed commit.

SourceStructure described in this article, not a measurement.

When the definition changes, it changes in one commit, dated, with the reason attached.

Regulation G asks for a reconciliation, not one definition

Regulation G governs what public companies disclose to investors, and it doesn't apply to your internal reports. Its logic transfers anyway. A public company that discloses a figure on its own definition must present the most directly comparable standard figure beside it and reconcile the two. Apply the same rule inside: any number that departs from the governed definition gets a label saying what it departs from.

The sequence that works

  1. 1.Inventory the metrics used in decisions, not every number on every dashboard.
  2. 2.Find the conflicts. Collect every existing implementation of each metric, export each query's filters, and compare them side by side. This step is uncomfortable, and it is where the value is.
  3. 3.Name one owner per metric. A named person arbitrates the definition.
  4. 4.Implement it once, in version control, in a semantic layer such as dbt's MetricFlow (opens in a new tab), Cube (opens in a new tab) or LookML (opens in a new tab). Test the edge cases the conflicts revealed. The most useful test: the metric, computed for a closed period, equals the figure finance reported for it.
  5. 5.Route every consumer through it and delete the old implementations.

Delete more dashboards than you build

Unused dashboards break silently and cost maintenance. They also let someone find a number that supports a position without anyone noticing which definition it used. Instrument usage, publish the list, and delete anything unopened for a full reporting cycle, with a way to restore it.

Put the number where the decision happens

A governed metric on a dashboard is available. The same metric inside the purchasing screen, when someone decides how much to order, is operational. Reverse ETL (pushing warehouse data back into operational tools), an embedded panel or a figure in the transactional system will each do it. Pick the one your team can maintain, and ask which screen your biggest recurring decision is made in.

Sources

  1. 1.17 CFR § 244.100 (Regulation G): disclosure of non-GAAP financial measures (opens in a new tab), U.S. Government Publishing Office,
  2. 2.About MetricFlow (opens in a new tab), dbt Labs
  3. 3.Cube documentation: introduction (opens in a new tab), Cube
  4. 4.What is LookML? (opens in a new tab), Google Cloud

Next step

Send us the three numbers.

The same month as finance, sales and operations each report it, with the figures changed if you would rather. We will tell you where the definitions diverge.

Reply
A person replies, not a sequence: within one business day, from someone who would be on the engagement.

Packages

Start with a package

Four packages, each with its price on the page. Pick one and buy it online.

Billing