Blog · AI and self-service analytics
Use AI-assisted questions in Power BI with an independent check on the answer. Work from a sales export to signed revenue totals, reconcile the report and verify the explanation.
Artificial intelligence with Power BI is most useful when a business question ends in a result you can reconcile to records. Define the sales measure, ask the question, then check the exported figures against signed transaction detail and an agreed control total. Keep the explanation only after those checks pass.
Microsoft describes Copilot for Power BI as supporting questions about data, report summaries and other assisted analysis. Its documentation distinguishes available experiences from previews and lists capacity, administrator and regional requirements. Check those requirements with your report owner rather than assuming every Power BI user sees the same capabilities (Microsoft's Copilot overview).
For a sales review, a useful question is: “Compare January and February 2026 net revenue in USD by region. Include posted credit memos and show the dollar and percentage change.” That gives the answer a measure, a time boundary, a currency and a grouping.
The broader distinction between explanation and calculation belongs to AI analytics. Here, the task is narrower: determine whether this particular answer agrees with the sales records.
Covirage is not affiliated with Microsoft, the provider of Power BI, Copilot and Excel.
Before asking, write down what net revenue includes. For this example it means posted invoice revenue plus signed credit memos, excluding sales tax and pass-through freight. The posting date assigns each record to its month. The agreed monthly totals come from the sales ledger under that same definition.
Record the report's visible filters and the date of the export. A region filter, a limited customer selection or an incomplete month can change a correct answer without changing its wording. Ask the report owner about any scope you cannot see.
Power BI exports can contain summarized data or underlying data, subject to permissions and report settings. Microsoft explains that summarized exports contain the columns and measures used by the visual; underlying exports provide additional detail where supported (Microsoft's export guidance). Keep the report export and the source transaction detail separate during reconciliation.
A summarized export checks what the report presented. The invoice and credit records check whether that presentation answers your business question. If you cannot obtain the necessary detail, describe the result as reconciled to the available summary rather than fully traced to transactions.
Consider a US industrial supplier. Its sales detail has these columns in A through E, with headers in row 1 and records in rows 2 through 7. Amounts are signed USD values; the period labels refer to January and February 2026.
| Record | Period | Region | Type | Amount (USD) |
|---|---|---|---|---|
| J-101 | January | East | Invoice | 40,000 |
| J-102 | January | West | Invoice | 30,000 |
| J-C01 | January | East | Credit | -2,000 |
| F-101 | February | East | Invoice | 46,000 |
| F-102 | February | West | Invoice | 36,000 |
| F-C01 | February | East | Credit | -4,000 |
Put East in H2 and West in H3. Use I for January, J for February, K for the change and L for the percentage change. In Excel, enter the January calculation in I2:
=SUMIFS($E$2:$E$7,$B$2:$B$7,"January",$C$2:$C$7,H2)
Enter the February calculation in J2:
=SUMIFS($E$2:$E$7,$B$2:$B$7,"February",$C$2:$C$7,H2)
Calculate the dollar change in K2:
=J2-I2
Calculate the percentage change in L2:
=IF(I2=0,NA(),K2/I2)
Fill down for West. Microsoft documents SUMIFS as adding values that meet multiple criteria. The period and region criteria here must match the exported labels exactly.
| Region | January net revenue (USD) | February net revenue (USD) | Change (USD) | Change |
|---|---|---|---|---|
| East | 38,000 | 42,000 | 4,000 | 10.5% |
| West | 30,000 | 36,000 | 6,000 | 20.0% |
| Total | 68,000 | 78,000 | 10,000 | 14.7% |
Percentages are rounded to one decimal place after calculation. East is 40,000 - 2,000 = 38,000 in January and 46,000 - 4,000 = 42,000 in February. West's increase is 36,000 - 30,000 = 6,000. The company increase is 10,000 / 68,000 = 14.7% after rounding.
Calculate the total percentage from summed revenue. In row 4, use =SUM(I2:I3) for January, =SUM(J2:J3) for February, =J4-I4 for the change and =K4/I4 for the percentage. Do not average the region percentages.
Suppose the report export contains gross invoice revenue while the question asks for net revenue. The reconciliation is explicit:
| Reconciliation | January (USD) | February (USD) |
|---|---|---|
| Gross invoices in report export | 70,000 | 82,000 |
| Less credits from detail | -2,000 | -4,000 |
| Recomputed net revenue | 68,000 | 78,000 |
| Agreed ledger net revenue | 68,000 | 78,000 |
| Unexplained difference | 0 | 0 |
The gross invoice increase is 82,000 - 70,000 = $12,000, or 17.1% after rounding. Net revenue increased by $10,000, or 14.7%. Both calculations can be correct, but only the second answers the specified question.
The next check is the explanation. “West contributed the larger dollar increase” follows from $6,000 versus $4,000. “West improved because a promotion worked” does not follow from these records. Check promotion evidence before accepting that cause.
For a reusable wording checklist, see how to ask an analytics assistant a question it can answer. For the question-to-calculation mechanism, see natural language analytics.
Keep the reviewed question, the scope, the export and the reconciliation together for the meeting. If the real decision is which reporting approach fits your team, the separate Power BI alternatives guide covers that choice.
Bring the exported sales detail with signed credits, period and region fields, plus the agreed totals for comparison. Covirage maps the columns; its tools compute and check the figures, and the external AI model explains the result with every figure traced to its rows. Create a free Covirage account, which takes about a minute, then set up your business and bring a sample of your own export.
See self-service analytics for the supporting business workflow.
Microsoft documents Copilot experiences that answer questions about data and summarize reports. Availability depends on your organization's supported capacity, configuration and permissions. Start with a specific measure, period and grouping, then check the result against the relevant export.
Use summarized data to reproduce the figures shown in a visual. Use underlying detail, when available and permitted, to investigate the records behind the total. A summary alone cannot prove whether a credit, duplicate or excluded transaction was handled correctly.
The figures may use different filters, dates or definitions. Gross invoices differ from net revenue when credits are present. Check the report scope and export options before comparing totals, then reconcile every remaining difference to identified records.
It proves the stated calculation, provided the source rows and definitions are correct. It does not prove why customers bought more or less. Separate an observed contribution to the change from a causal explanation that needs additional business evidence.