We continue our Copilot Journey. This time, we will cover Copilot Pane within Power BI Desktop. As a reminder, this article is a part of the Copilot series that I want to cover on my blog:
What are we going to cover?
Copilot provides us with Report Authoring capabilities. Initially I wanted to cover it within Power BI Service article, however, I took a chance to create two shorter articles. On top of that, Microsoft announced new Copilot Pane within Power BI Desktop. When it comes to Report Authoring, I am old fashioned, so, I prefer to use Power BI Desktop for this task. We are going to test the Copilot, and, as we did in previous article, we will look at the Capacity Utilization at the end.
In previous article I talked about setting up Copilot to work within Power BI Desktop. Make sure to catch up and let’s start.
Important disclaimer
Remember that Copilot is still in Preview, and a lot of the final result depends on the available data. Microsoft described the way to prepare Semantic Model ot make sure, that we are going to squeeze maximum value from Copilot. So, we will see what Copilot can currently do, but let’s not be too judgmental. We are here mainly to check how much it impacts the Capacity Utilization. For my tests, I am using P2 Capacity (F128).
How to access Copilot Pane?
Main option is to use a relatively big button in the Home ribbon:
Alternatively, you may also make it available as one of the panes, using Pane Manager:
If it’s your first-time using Copilot, you will be asked to complete the setup (provide workspace details, etc.). If you ‘ve done it already, Copilot Pane will appear.
First interaction
As described in
announcement from Microsoft, there are currently three options to interact with Copilot. Two of them have their own “buttons” in Copilot Pane. One is kind of hidden –
Summarize the model. Let’s type that prompt and send to Copilot:
The output is ok (ish). It’s short, accurate, but is it really useful for the Power BI Developer? I can hardly imagine a situation, where I start to work with specific data product, and my knowledge about it is so low, that I would benefit from a model description like the one presented. On the other hand, we could use this feature to quickly generate description of Semantic Model, which we could use maybe in some sort of “Data Catalog” for data consumers.
Report creation
We are already warmed up, let’s start creating a report using Copilot. You’ve seen already what the available options are. Let’s select “Suggest content for the report” and see what Copilot has to offer. As a reminder, I am using a Retail data sample from Fabric Pipelines (described in previous article). Copilot’s output was the following:
As you can see, Copilot has a pretty decent idea of the report to create. As expected, first suggested page is an overview of sales performance. However, I must admit I am surprised with next two: “Customer Buying Patterns” and “Sock Item Analysis”. If you are happy with page suggestion, you may click on “+ Create” button. If you want to edit the prompt (request/question to Copilot), click on “pencil”. This will copy the original prompt to chat, where you can edit it before sending to Copilot:
I will start simple, and I will just hit “+ Create”. Here is the generated output:
Couple of things to highlight:
- Design is very clean
- Objects are well positioned and distributed
- As two main values for Card Visuals, Copilot used the two measures we’ve been calculating in previous article
Color theme used seems to be the same we see during the demos of Copilot, let’s make it first thing I will try to change. I tried asking to change color theme, but this is not possible (at least with current palette of commands). I tried slightly different approach and asked about the colors used in report, but result was the same:
Let’s leave it for now and try other customizations. I tried to modify the content of the report, ask to change the chart type, or the data attribute used to slice the data. Both requests failed:
Seems that once the report layout is built, Copilot in current stage is not capable of modifying it. I will try to provide some suggestions for the page creation. This also turned out to be impossible:
At least, by having a look at the output from Copilot, we can assume, that my request was actually understood by the Copilot. Answer is not at all generic. It clearly alludes to my request. Before giving up, let’s not forget about the other option we saw – “Create a page that shows“. This time Copilot prepared the report page that meets provided requirements:
I will use the “Undo” option to get rid of the page and try once again, but this time I will mention a chart types as well as color theme. Sadly, it didn’t work:
I will try once again, but without the color part:
This time it almost worked. Page was created and Copilot used Country information instead of the Region within bottom charts. Still, I asked for pie charts and got doughnuts and area charts instead of line charts. Let’s see if the “Create Page” option allows for some modifications when page is created. I asked to update chart types and failed. Copilot can’t currently add other objects like navigation to different page. This is probably still too advanced feature for Copilot at the moment:
We know now that we can’t go wild with Copilot yet. Let’s quickly ask Copilot to generate the remaining pages that you could as a part of “Suggestions” and assess the Capacity Utilization, which is the main point of this exercise.
Customer Buying Patterns
Interestingly, Copilot didn’t use the Invoiced Amount here at all, even though it was that important to include it on main page. On the other hand, you can see here properly identified key data attributes for this type of analysis: Customer, Buying Group and Category.
Stock Item Analysis
Here, we see Total Invoiced Amount back again. Rather boring and unidimensional analysis. Important data attributes (related to stock items) like Buying Package and Brand are not included. On the bright side, I am really happy to see Profit and Quantity values included in analysis. In future iterations of Copilot, I would like to see features like applying consistent color coding for each section.
Employee Sales Contribution
The least interesting creation out of all the generated pages. Are charts at the bottom are not easy to read with Employee as a dimension.
Feature summary
If you thought about the Copilot as help to create advanced report functionalities faster – this is not the case here. It works well for DAX Query View, but Copilot Pane (at least for now) is capable of supporting Power BI beginners. Could be useful to go through what Copilot has to offer in terms of Suggestions – maybe it will include some charts that you didn’t think about including. Additionally, I must say Copilot comes up with very decent naming for Report Pages. This is very often overlooked aspect of report design. To summarize current capabilities of Copilot, let’s review again available options:
- Summarize model – provides short description of semantic model.
- Suggest content for this report – generates list of pages that Copilot suggests including in your report. Prompts to Copilot are pre-generated, but you can edit them before sending. Currently, Copilot can’t accept detailed requirements regarding content of generated pages, specifically types of charts or dimensions included. You can’t update color schema either.
- Create a page that shows… – option where we less rely on Copilot Suggestions, but we know what we want to include. We can provide chart types (almost, it doesn’t work very well), as well as dimensions included. Color schema can’t be changed, and report page can’t really be updated once it is generated.
As you can see, capabilities are very limited at the time being. But let’s not forget this is still a Preview Feature. That’s why it is important to keep the interest in this feature on adequate level to added value it provides. It’s ok to allow small group of people to use the Copilot, and actively test capabilities. Enabling product on this stage to the entire organization will result in sending too many prompts, generating higher impact on Capacity consumption.
How much does it cost?
As it was done in previous article, it’s time to check the Capacity consumption:
All of our interactions with Copilot generated impact on Capacity equal to 0.2% of Base Capacity (P2 Capacity, 0.4% of P1). Again, as in previous article, it’s hard to say if it’s a lot. Having 20 sessions like mine, will consume 4% of Base Capacity (P2 Capacity, 8% of P1). Adding it to 20 developers running similar session with DAX Query View as we did last time, it will sum up to 13% of Base Capacity. Let’s not forget, that I am using P2 Capacity for tests. Translating it to standard P1 Capacity (F64) requires multiplying Base Capacity consumption by 2. In this case, we don’t talk about 13% but 26% percent of Base P1 Capacity… If we look at the Copilot consumption from perspective of individual session, it’s not scarry at all. When we try to make it more realistic and scale it up to large organization size, it looks more and more concerning.
Conclusion
I covered another important aspect of Copilot usage, and its impact on Capacity consumption. We can argue if my example of 26% of P1 Capacity consumption is an edge case scenario or not. In my opinion it’s very realistic, but I present it more for you to draw your own conclusion by looking at the cost of a single session and trying to translate it to the reality within your organization.
In the next article, we will cover Power BI Service consumption.
For now, thank you for reading, and see you in next article 🙂
Great job!
huge dose of knowledge!