AI and Power BI: practical uses for ChatGPT and Claude

Where AI fits into the work

Building a Power BI report involves more than choosing visuals. Source files need checks, measures need definitions, calculations need to be written and results need reconciliation. ChatGPT and Claude can assist with those tasks, including when Copilot is not enabled in Power BI.

A small sample table, column descriptions or a specific calculation is often enough to start. The examples below are illustrative tasks, not tests using client data. Sharing a file or code in an ordinary chat does not establish a live connection to Power BI or automatically update a report.

ChatGPT: inspect a source file

Suppose a sales CSV contains mixed date formats, several spellings of a branch name and blank amounts. ChatGPT can help identify questionable records and propose processing rules.

Example request: “Inspect this sales file. Flag mixed date formats, blank amounts and possible duplicates. Use the document number plus line number as the candidate duplicate key. Do not delete anything yet. Return a separate list of flagged records and explain each issue.”

The first output is a review list. A blank amount does not automatically become zero, and a similar row is not automatically a duplicate to remove. Once confirmed, the cleaning rule can be implemented in Power Query so subsequent files follow the same process.

ChatGPT: draft a DAX calculation

DAX is Power BI's calculation language. A useful starting request describes the existing measures and the exact rule for a new calculation. “Write a margin formula” leaves too much unspecified.

Example request: “The model has [Net Sales], which deducts returns, and [COGS], the cost of goods sold. Draft a gross margin percentage measure using these two measures. Return BLANK when sales are zero. Explain the calculation and list test cases.”

The intended logic is (sales − cost of goods sold) / sales. With fictional sales of 1,000 and costs of 700, the result should be 30%. Then check zero sales and the combined result for multiple branches. Overall margin is total profit divided by total sales; simply averaging branch percentages gives a different result.

Claude: review Power Query logic

Claude can review Power Query M code, explain its steps and help identify questionable transformations. State the conditions that any revision must preserve.

Example request: “Review this M code. Product codes must stay as text, including leading zeros. The expected date format is day/month/year. Unknown branches should go into a separate review list. First explain the current steps, then propose a revision and identify assumptions that need clarification.”

Try the proposed change in a copy of the Power BI Desktop file. Compare row counts, distinct products and sales totals before and after the change. A convincing explanation of the code does not establish that it runs correctly.

Claude: reconcile two report exports

Another practical task is comparing two aggregated exports for the same period. For example, an ERP export and a Power BI export may show different totals. Claude can help break the difference down by branch and product group.

Example request: “Compare these two CSV files. First check the period, amount columns and level of aggregation. Then join them on branch and product-group codes. Show amount differences and unmatched records on both sides. Separately list differences that the supplied data cannot explain.”

Claude's file-handling capabilities can support this comparison and the creation of a reconciliation table. An analyst still checks the join keys and control totals. Reconciling two files is not a complete causal analysis of sales performance.

Move the result into a repeatable workflow

The examples assign tasks to ChatGPT and Claude to illustrate workflows, not to claim one is more accurate than the other. Similar tasks can be tried in both. Choose based on the output, available features and the team's working environment.

Provide more than a file: explain the columns, business rules and a known example. After checking the result, document the approved logic—what the measure calculates, what it excludes and who checks it.

Use fictional or appropriately anonymized samples for a trial. Before sharing actual work files, check the organization's approved environment and data-use settings. Available features and file-processing conditions may differ by account.

Where Copilot fits

Copilot offers assistance within Power BI itself, including report summaries and help creating pages. Compared with discussing code in a separate chat, this makes the existing data model and Power BI environment central to the setup.

Microsoft currently requires a suitable paid Fabric F2+ or Power BI Premium P1+ capacity, relevant administrator settings and a supported region. A Pro or Premium Per User licence alone is not sufficient. Evaluate Copilot against the particular workflow the team needs.

A short plan for the first trial

Start with one recurring task: checking a file, drafting one DAX measure or reconciling two reports. Define the correct result and how to verify it before beginning. Move generated code or text into the working report only after checking it.

To discuss a project with DataStudio, start with a description of the current report, definitions of its main measures and a task the team repeats regularly. Our services and the official sources used are linked below. Technical information was checked on 16 September 2026.

• Limit shared data to the context the task actually requires.
• Specify the period, calculation rule and required output format.
• Preserve the original and test changes on a copy.
• Check totals, missing values and edge cases.
• Record the approved logic and its owner.