When Power BI Reports Disagree: Two People, Two Numbers, One Meeting
Finance says revenue was 2.4. Sales says 2.6. Both pulled from the same system, both are confident, and the next twenty minutes are spent on arithmetic instead of the decision. This is not a data problem. Here is what it actually is.
01
You will recognise this
Somebody puts a slide up. Somebody else says "that is not the number I have." The room goes quiet in the particular way it does when everyone privately suspects their own figure might be the wrong one.
What follows is never a resolution. It is an action item: two people will take it offline and reconcile. They will spend a day on it. They will find the cause, quietly agree a fudge, and it will happen again next quarter with a different metric.
Nobody built a wrong report. Each one is internally consistent and defensible. They just answer slightly different questions while using identical words.
The tell
If your organisation has a person whose informal job is to explain why two reports disagree, you have found the workaround. That person is a control, and they are the only one you have.
02
Where the difference actually comes from
In almost every case it is one of five things, and none of them are errors.
| The quiet decision | What one person assumed and the other did not |
|---|---|
| Which date | Order date, invoice date, payment date or delivery date. All defensible, all different months. |
| What is excluded | Cancellations, internal transactions, test accounts, GST, freight, the one client everyone treats as a special case. |
| When it was pulled | Tuesday's extract against Thursday's. Late entries and credits landed in between. |
| Which grain | One counts orders, the other counts order lines. The same data, counted at two different levels. |
| Whose hierarchy | Region by customer address or by account manager. Reorganisations make both true and neither comparable. |
Notice that not one of these is fixed by better visualisation. They are all decisions made silently at the moment somebody built their own copy of the data.
03
What actually fixes it
One model, used by everybody, where the definition is written down once as a measure rather than re-derived by each person who needs the number.
In Power BI this is the difference between a calculated column and a measure, and it is the single concept that separates people who fight with Power BI from people who find it obvious.
A column computes a value for every row, once, when the data refreshes. A measure computes when somebody looks, using whatever filters they have applied. Write Total Revenue as a measure and it is correct by month, by region, by salesperson and in total, because it is one definition being evaluated in different contexts rather than five definitions that happen to agree.
That is what stops the disagreement. Finance and sales can slice it differently. They cannot calculate it differently.
The part people skip, and should not
A shared model needs a proper date table, marked as the date table, with one row per day covering your whole range. Almost every "the totals do not match by month" problem traces back to its absence. It takes ten minutes to add and it prevents an entire genre of argument.
04
The half nobody wants to do
Here is the uncomfortable part. Building the model is the easy half. The hard half is that somebody has to decide which definition wins, and that is not a technical decision.
When finance and sales have used different revenue definitions for six years, choosing one means somebody's historical numbers change. That conversation is genuinely awkward and it is why so many organisations quietly prefer four dashboards to one.
What makes it manageable is doing it in the right order:
- Write the definitions down before you build anything. One line each, in words, agreed by the people who use them. "Revenue is invoiced amount excluding GST, on invoice date, excluding intercompany."
- Accept that some metrics need two. Sales pipeline revenue and recognised revenue are genuinely different things. Give them different names and stop pretending they should reconcile.
- Name an owner per metric. Not a committee. One person who decides when the definition changes.
- Then build it once, publish it to a shared workspace, and have every report connect to that model rather than importing its own copy.
Power BI supports this with shared semantic models and endorsement, so a model can be marked as promoted or certified and users can see which one is the real one. That feature is worth nothing until step one is done.
05
What it honestly takes to learn
- One day to load data, build relationships and produce a report people can use. This is a genuinely fast start and it surprises people.
- A second day for the model itself: star schema, a real date table, measures instead of columns, and why a flat table imported from Excel causes most beginner problems.
- A third for DAX, where filter context stops being mysterious. Most people hit this wall about three weeks in, when a total is right and the subtotals are not.
The honest warning: Power BI lets you build something that looks finished on day one and is structurally wrong. Nothing tells you. It works until the day someone asks a question you did not anticipate, and then it quietly gives the wrong answer. That is the specific failure a course prevents and self teaching usually does not.
Power BI Beginner gets you producing. Power BI Intermediate is where the modelling lives, and it is the one that fixes the problem in this article. Power BI DAX is for when the subtotals stop behaving.
Try this in ten minutes
The definition audit
You do not need to open Power BI for this, and it is more revealing than anything you could build.
- Pick the metric that appears in the most meetings. Revenue, utilisation, active customers, whatever yours is.
- Message three people who use it, separately, so they cannot see each other's answers.
- Ask exactly this: "In one sentence, how is this calculated? Which date does it use, and what is excluded?"
- Put the three replies side by side.
If they match, you are in better shape than most and this article is not urgent. If they do not, you have just found the cause of every reconciliation meeting for the last two years, and you have it in writing, which is exactly what you need to get the fix funded.
Build the capability
Related Nexacu courses
Instructor led, one day each, face to face across Australia or live online.
Power BI Intermediate
1 day | $595
Relationships, star schema, date tables and measures. The direct answer to this article.
Power BI DAX
1 day | $595
Filter context, time intelligence, and why the total is right when the subtotals are not.
If the disagreement spans several teams, corporate training run on your own model gets everyone to the same definition in the same room, which is usually the actual bottleneck.
Common questions
Frequently asked questions
Do this next
Run the definition audit on your most argued about metric. It takes ten minutes, it costs nothing, and the three replies sitting side by side make the case for the fix better than any business case you could write.
The Workaround is a fortnightly series on the manual processes organisations build because nobody showed them the feature. Previously: two files open side by side. Next: one person knows how the model works, and what happens when they resign.
