Before I Trust an AI-Built Excel Dashboard, I Check These Five Things
![]() |
| A fictional travel budget makes the checks easy to follow: €290 for accommodation, €92 for transport and €58 for meals. Original chart; not an Excel screenshot or a Copilot-generated result. |
I like a good dashboard. A few clear numbers, a chart that answers an actual question, and suddenly a spreadsheet becomes much easier to understand. What makes me nervous is how quickly a polished layout can make those numbers feel trustworthy.
Microsoft Copilot makes building an Excel dashboard more approachable: describe the result, let it organise the presentation, then refine it. That is a useful role for AI. But when the subject is my spending, a beautiful chart with the wrong total is still the wrong total.
My rule is simple: before I trust the dashboard, I check the calculation, the filters and what happens when the data changes. Here is the process I recommend, using a small fictional travel budget rather than anyone’s private accounts.
First, make sure the tool fits the job
Microsoft’s guide to AI data visualisation describes creating charts and dashboards through Copilot in Excel. Editing requires an eligible Microsoft 365 subscription; Microsoft lists Personal or Family with an AI credits plan, Premium, or a commercial Microsoft 365 Copilot subscription. The exact features available can depend on the account and organisation settings.
I check that before paying for anything. If I already have access, a small dashboard is a sensible experiment. If all I need is a monthly total and three category sums, ordinary Excel formulas may already finish the job. The useful comparison is the time needed to produce and check the finished result.
1. Give the numbers a clear meaning
For this example, the budget is €500. Every amount is in euros, expenses are positive, and refunds are negative. There are seven transactions:
| Date | Category | Transaction | Amount |
|---|---|---|---|
| 3 October | Transport | Train | €48 |
| 3 October | Accommodation | Hotel | €160 |
| 4 October | Meals | Food | €32 |
| 12 October | Transport | Train | €44 |
| 12 October | Accommodation | Hotel | €150 |
| 13 October | Meals | Food | €26 |
| 13 October | Accommodation | Hotel refund | −€20 |
That definition matters. “Spending” could mean everything booked, everything paid, or the net amount after refunds. For sales, the same question applies to cancelled orders, returns and tax. I want the definition settled before asking AI to choose a headline number.
I keep one header row, consistent category names and actual numeric amounts. I also keep the original data intact and work on a copy. A prompt cannot reliably rescue a workbook whose meaning I have not decided.
2. Ask for a dashboard I can inspect
My starting prompt is deliberately specific:
Create a travel-spending dashboard on a separate sheet using this transaction table. All amounts are EUR. Treat refunds as negative expenses. Show net spending, the remaining amount from a €500 budget, and spending by category. Add a category filter. Explain which figures respond to the filter and which remain whole-budget totals. Keep the source data unchanged and explain how new rows will be included.
Then I identify what Copilot has produced: an editable chart, a PivotTable, a dedicated dashboard sheet or a static visual. Those are different things to maintain. Microsoft’s data-analysis guidance specifically warns that some returned tables or image visualisations are static and cannot be refreshed.
I also want a plain explanation of the source range and calculations. If I cannot trace a number back to the transactions, the dashboard is not ready.
3. Check the total independently
The seven amounts above add up to €440, leaving €60. Accommodation contributes €290 after the refund, transport €92 and meals €58. Those category totals also add to €440.
I verify that with the source rows and a simple independent calculation. Asking the same AI whether its answer is correct is useful as an extra check, but it does not replace doing the arithmetic separately.
For a sales dashboard, I check both revenue and the number of included orders. If average order value is displayed, its numerator and denominator must describe the same set of orders. A correct division can still answer the wrong question.
4. Make the filters earn their place
Filtering this example to Transport should show €92 across two transactions. Clearing that filter should restore the €440 total and all seven rows. Selecting Meals should produce €58.
The remaining-budget figure needs particular care. If it stays at €60 while a category is selected, that can be perfectly sensible—but it must say that it is the remaining overall budget. A filter that changes some charts while leaving other figures untouched becomes confusing unless the labels explain the difference.
I also check combinations, such as a date range plus a category, and a selection with no matching transactions. Zero, blank and an error message are not interchangeable.
5. Add a row before calling it finished
Here is the most useful final check: add a fictional €15 meal on 20 October to the test copy. The expected total becomes €455, meals become €73 and the remaining budget becomes €45. There should now be eight transactions.
If those figures do not change, I check the source range, the filter state and the refresh behaviour. Microsoft’s PivotTable refresh instructions explain the available controls; automatic refresh depends on the version and feature availability. I do not assume that one automatically updating dashboard means every AI-generated output will behave that way.
After changing the layout or asking Copilot to restyle the page, I repeat the checks. Visual changes should not quietly alter what is being counted.
My verdict: useful assistance, with a checking habit
The appeal of this workflow is real. AI can help me get past the blank sheet, suggest a useful layout and reduce formatting work. My contribution is deciding what the numbers mean and making sure the result still behaves correctly next month.
That fits the approach behind the AI skills I think are worth learning: a clear brief, a result I can inspect and a repeatable way to check it.
For a one-off calculation, I may not need a dashboard at all. For a recurring report, the time spent checking totals, filters and new rows is part of building something useful. That is the version of AI-assisted productivity I am interested in.
.png)
Comments
Post a Comment