Two figures for the same measure in one meeting

The monthly review opens and the figure on the sales slide is 4.2 million while the figure finance brought is 3.9 million. Both teams prepared properly. Both can show their working. The next forty minutes go on the gap, the decision the meeting existed to make slides to next week, and two analysts leave with an action to reconcile.

I have sat in that meeting on both sides of the table. The pattern repeats. Somebody says the data is wrong, somebody else says the report is wrong, and a third person suggests moving reporting into a different tool. None of those statements can be tested yet, because nobody in the room has established what the two figures were counting.

The deferred decision is the expensive part. A pricing change waits a month. A headcount request waits a month. The reconciliation gets done by the two analysts who are the only people who understand it, and the same gap appears next quarter because nothing was written down. A metric two people cannot reproduce has a habit of coming back.

Almost every disagreement is timing, population or attribution

Before anyone touches a report, work out which of three things is producing the difference, because the fix differs for each. Timing covers the as-of date, the day the period closes, and whether corrections posted after the close get pulled back into the earlier period or left where they landed. Two teams running identical logic a day apart against a moving table will differ, and neither of them has made a mistake.

Population covers who is in and who is out. Cancelled orders, internal transfers between legal entities, test records left behind by a training session, employees on unpaid leave, accounts flagged as duplicates but not yet merged. Each of those is a fair exclusion for one team and a fair inclusion for another, and the choice almost never appears in the report title.

Attribution covers which dimension a record counts against when more than one applies. An order placed by a customer in one region, fulfilled from another, sold by a rep who reports into a third. A requisition opened by one manager and filled under a different cost centre. Sales counts on the owner, finance counts on the entity that books it, and both totals are complete.

Arguing about tooling while the disagreement is definitional burns a quarter. Moving the same two definitions into one warehouse gives you two columns instead of two reports. Work out which of the three you have first, then say it out loud, because the sentence 'we differ on population' closes an argument that 'the numbers are wrong' can run for weeks.

Get both definitions written down first

The first move costs nothing. Ask each team to write its definition in three or four sentences on one page, covering the three properties above. No screenshots, no query, no link to a report. Words.

Often that ends it. One side sits down to write and finds they cannot state the rule they have been defending, because the logic lives in a filter somebody set two years ago and nobody has read since. A team that discovers this on its own arrives at the next meeting wanting what you want. Sending them a reconciliation instead produces a defence.

Keep the two pages side by side and unedited. Resist writing one combined definition at this stage, because the differences are the material you need and a merged draft hides them. Analytics questions belong in the requirement for the same reason, which is that the wording is the specification.

Reconcile at the record level, then decide

Pick one closed period and one specific record set. Last month, one region, one legal entity. Have both teams produce their own figure for exactly that scope and hand over the record identifiers behind it rather than the total.

Find the records that appear on one list and not the other, then trace five or six of them individually. Tracing at the record level finds the cause. A reconciliation at the total level tells you the gap is 300 thousand, which nobody can act on. A reconciliation at the record level tells you that eleven cancelled orders sit inside one figure and outside the other, and it closes the investigation in an afternoon.

Then somebody chooses. It has to be somebody with the authority to tell a department that its number is changing, because two reasonable definitions never resolve themselves through further discussion. Analysts cannot settle it either, since neither department will accept the other's analyst as the referee. Who gets to define revenue is a question about authority rather than logic, and the same holds for headcount, pipeline and cost per case.

Give the agreed definition a home and one implementation

Most of these efforts fail after the agreement rather than during it. The decision gets made in a meeting, everyone nods, and nine months later the two figures diverge again because the agreement lived in a chat thread. The definition needs a written home that somebody owns, and the owner is a named person rather than a team.

Then implement it once and reference it everywhere. One calculation in the shared model, one view in the data layer, and every report reads that instead of rebuilding the logic locally. Reimplementing the same definition in nine reports gives you nine chances to drift, and the drift arrives quietly. It surfaces the first time somebody compares two dashboards that carry decisions.

The written home stays short. For each agreed measure, this is what I want on the page.

  • The measure name, in the words the business says out loud.
  • The named owner, and who stands in when they are away.
  • The as-of rule, the close date, and how late corrections are handled.
  • The population, written as what is included and what is deliberately excluded.
  • The attribution rule for records that could count in more than one place.
  • The one place the calculation is implemented, and the reports that read it.
  • The date of the last change and who approved it.

Different definitions get different names

Some divergence is legitimate. Finance needs a booked figure on a statutory close and sales needs a committed figure as of last night. Forcing one definition onto both would stop somebody doing their job properly.

Give them different names, and never let two measures share one. This is the most useful discipline available here, and the whole cost of it is a naming convention. Two things called revenue will be compared by anyone who sees them next to each other, and whoever built the slide gets asked to explain a gap that was designed in. Booked revenue and committed pipeline do not get compared by accident, because the names carry the difference.

Record the definition that lost alongside the one that won, with a line on where each applies. The same discipline runs through an ERP master data governance starter, where one owner and one written rule per field achieve more than any amount of cleanup.

Definitions change, so run the change through the same door

Businesses reorganise, entities get sold, a new product line changes what counts as an order. The definition will need to change, and that change needs the process that produced the original agreement, including the owner's decision and a dated entry. A quiet edit to the calculation is how a hard-won agreement dies.

Add one recurring check. Twice a year, take each agreed measure and confirm that the reports claiming to use it actually read the shared implementation. Somebody will have built a local copy under deadline pressure, meant to come back to it, and moved on. Finding that in a review costs an hour. Finding it in a board meeting costs the definition.

Start with the measure your last two leadership meetings argued about. Ask both departments for a written definition by Friday, then pick one closed month and ask for the record identifiers behind each figure. You will know which of the three properties you are fighting about before the written definitions land.