How to extract automatic insights with Copilot in Power BI

João Barros 29 de July de 2026 4 min read

This tutorial shows how to use Copilot in Power BI to extract automatic insights from a report: identify outliers, obtain explanations for variations and suggest measures. It is useful for turning raw data into actionable recommendations without writing complex DAX.

Prerequisites

Account with access to Power BI and Copilot enabled.
Report in Power BI Desktop or in the service with at least one fact table (sales, transactions) and dimensions (date, product, region).
Data model with well-named columns and defined relationships.

Step 1: Prepare the report for Copilot

Copilot works better with clear columns and tables. Check field names, add descriptions and ensure date columns are of type Date. This helps Copilot interpret questions correctly such as "why did sales fall in April".

Step 2: Open Copilot and ask for a high-level summary

In Power BI Desktop or in the service, open the Copilot pane. Start by asking for a general summary so Copilot can create context about the report. Use a simple and direct prompt.

Resumo geral: dá-me os principais drivers de receita no último ano e três insights accionáveis.

Explanation: this request forces Copilot to analyze trends and highlight main drivers (products, regions, channels).

Step 3: Ask Copilot to identify outliers

After the summary, ask Copilot to find outlying values by month or by product. This is useful to detect data errors or specific campaigns that affected results.

Identifica outliers nas vendas mensais do último ano e mostra os 5 meses e produtos com desvios superiores a 2 desvios padrão.

Explanation: by mentioning criteria (2 standard deviations) Copilot applies statistical logic to justify the outliers — if the model cannot calculate directly, it will ask for clarification or generate a suggested DAX measure.

Step 4: Generate a suggested DAX measure to explain variations

If Copilot indicates the variation is due to price or quantity, ask for a DAX measure to decompose the variation. Copilot often suggests ready-to-paste code for the model.

Cria uma medida DAX que calcule a variação percentual ano a ano de SalesAmount e outra para decompor em PriceEffect e VolumeEffect.

Minimum example of a measure that Copilot may suggest (place in your model):

YoY Sales % =
VAR Current = SUM('Sales'[SalesAmount])
VAR Prior = CALCULATE(SUM('Sales'[SalesAmount]), SAMEPERIODLASTYEAR('Date'[Date]))
RETURN
IF(ISBLANK(Prior), BLANK(), (Current - Prior) / Prior)

Step 5: Request causal explanations with evidence

Ask Copilot to explain why a segment fell or rose, and to present supporting filters and visualizations (e.g., a table with the top 10 products). This helps validate the conclusion.

Explica por que as vendas na região Norte caíram em Março. Fornece três hipóteses e mostra uma tabela com os 10 principais produtos afectados.

Explanation: Copilot should respond with hypotheses (price change, promotion, stockouts) and propose filters/visuals that confirm each hypothesis.

Step 6: Turn recommendations into actions in the report

When Copilot suggests filters, visuals or measures, apply them in the report: create a new page or insert the suggested visuals. Ask Copilot to generate the visual code (if applicable) or step-by-step instructions.

Cria um visual sugerido para comparar PriceEffect vs VolumeEffect por mês e adiciona um filtro para Categoria de produto = "Electrónica".

Explanation: follow Copilot's instructions to add the fields to the visual and configure the slicer for the category. Test different granularities (month, quarter).

Verify the outcome

Confirm you obtained: (1) summary with revenue drivers; (2) list of outliers with criteria; (3) DAX measures created (e.g., YoY Sales %); (4) causal explanations with a supporting table; (5) new visual or page with the applied recommendations. Validate the numbers by comparing with manual calculations in pivot tables or simple DAX.

Conclusion

By using Copilot in Power BI to extract automatic insights you save time and get action ideas without writing much code. Next steps: try more specific prompts (by channel, customer or campaign) and save the measures/visuals generated. Tip: always validate Copilot's hypotheses with a sample of the data — what was the most surprising insight you found?