Now we can interact with Copilot through natural language to get data analysis and results, you can refer to the following principles and tips when asking questions.
1. Question description should be clear and concise
1.1 Try to keep questions as unambiguous as possible, otherwise they may return answers with different possibilities
If you want to know the value of a measure, you need to bring in the type of aggregation. For example, avoid questions such as "coffee sales" or "maximum sales of product categories" and instead ask "total coffee sales" or "which product category has the highest total sales". The total sales of coffee", "which product category has the highest total sales".
1.2 The same question may be asked differently with different data.
For example, if the question "What product category sells the best?" is asked in the data, and there are both sales and total sales, the question will be answered differently. If there are both sales and sales fields in the data, then Copilot may interpret it as sales or sales. It is recommended that this be changed to "What product category has the highest total sales?" .
1.3 Avoid unattainable data requests
For example, in the question "Show sales by month in 2016 using a map", there is no geographic information in this question, so it can't be shown on a map, Copilot can't answer it, and it may end up returning other graphs.
1.4 Avoid vague descriptions of intent
For example, "I would like to know the data of products with an average total cost between 40 and 55", it is suggested to change it to "Which products have an average total cost between 40 and 55".
2. Copilot Limitations
At present, Copilot dialogue supports single dimension and single measure, multi-dimension and multi-measure, sort, filter, year-on-year, month-on-month, simple dynamic calculation, time level, time nesting, topN, etc., and does not support subquery, automatic creation of calulated column, attribution analysis and other scenarios. |
