It’s time for last article in the series. It is already longer than initially anticipated, and I didn’t even cover all the Copilot features available today in Power BI, not to mention Fabric. As usual, let’s look at what was covered so far:
- Power BI Copilot – Introduction
- Copilot in Power BI Desktop
- Copilot in Power BI Service
- Conclusion – current article
A lot of testing was done. What I will try to do in this article is to summarize all the findings and calculate a single outcome of what you could expect when Copilot is enabled in your organization.
Test summary
- Report generation (developer):
- Single session – 0.4 %
- DAX Query View (developer):
- Single session – 0.92 %
- Dataflows Gen 2 (developer):
- Single session – 1.26 %
- Report Summary (consumer):
- Single session – 0.86 %
The average cost of a developer session is around 0.86 %, coincidentally the same as a consumer session. This allows us to disregard the user type mix in these calculations. Whether there are 10 developers and 30 consumers or the reverse, it doesn’t significantly impact the overall cost. Let’s round this to 0.9 % and scale it for the given number of users, assuming daily consumption.
Is 30 users per day a realistic number, combining both developers and consumers? I believe it is. This results in a total Capacity Consumption impact of 27 %. If your current Capacity workload is between 70-80 %, adding another 27 % would push it above 100% for a 24-hour period. This triggers the highest level of Capacity Throttling, completely blocking interactive and background operations. In simple terms, users won’t be able to view any reports, and all data processing jobs will stop.
Dipping the toe
My thoughts
My opinion may be influenced by my initially low expectations, which left me more impressed than disappointed. However, I struggle to definitively say that Copilot provides significant added value today. It seems to benefit advanced users more than Power BI rookies. One could argue whether experienced Power BI developers need Copilot at all and if beginners can ask precise enough questions for Copilot to be truly helpful.
A crucial point, as noted during my test of the Report Summary feature, is the potential for increasing data literacy in your company. Report consumers could use Copilot to learn how to understand the data, eventually reducing the need for Copilot to interpret results.
Whether you love or are skeptical of Copilot, it’s likely here to stay. Demand and expectations are high. As a Tenant Admin, you may find yourself swimming upstream, but let’s hope enabling Copilot will ultimately be worth the effort and the cost.
Few ideas
To mitigate the risk of high capacity impact, it could be useful to enable isolation of Copilot workloads to a dedicated Capacity. Currently, for Power BI Desktop Copilot, you can assign a workspace backed by Premium Capacity (F64). Even an empty Capacity can be used to track Copilot costs. Extending this to all Copilot workloads would reduce the risk of Capacity Throttling. You can support this idea here:
In my tests, I focused on Power BI-related Copilot (except for Dataflows Gen 2 in Data Factory). However, Copilot exists for nearly all experiences. Having more granular control over which Copilots are enabled would be beneficial. If you agree, vote here:
Granular control for different Copilots in Fabric
Last one is ability to turn off the Copilot for specific reports. Microsoft have published guidelines on how to prepare your Semantic Model to work with Copilot. It means that it requires a certain groundwork to be done before Copilot could be enabled for end users. Similar control is available today for Q&A feature, and it would be great to have the same for Copilot. If you agree, please, vote for idea here:
Ability to turn off Copilot ona single Power BI Report level.
Conclusion
A lot of testing was done, but I hope it will help you to make the right decision regarding Copilot, and you will choose slow and responsible adoption over the uncontrolled hype. If you have any questions, feel free to reach out to me in comments below or email.
For now, thank you for staying till the very end, and hope to see you again in next article 🙂
