In my previous article, I shared the first observation from testing Copilot in the Power Apps Studio experience. Today, we will continue to explore this feature. However, we will switch to a different environment. In this environment, we have a specific Opt-in card in the Power Platform Admin Center. This card allows us to move data across regions and to enable or disable Bing Search.
I used the same prompts as in my previous article, where I performed my tests in a US environment, and created an app. I tried to use DALL-E again and asked Copilot to help me summarize the last email I received. However, as you can see from the results shown in the image above, neither of the actions was successful.
I tried again with my prompt on writing an introductory paragraph, but I got a different result this time. Unlike the US-side, I got an error message saying „Sorry, something went wrong…“. Was it a technical issue? I decided to try again. No, the same message appeared. I tested it with different environments in the EU, but the outcome was always the same. What about the VanArsdel-thing that I got when using a US environment? I decided to check that too.
Same as in the US environment, when performing a search in the environment, I don´t receive a result, unless I switch the toggle to include web results. The results then include the documentation from learn.microsoft.com. Why is this happening? Is Copilot using a different training model in the EU? Is it because the data doesn’t leave the EU boundary and Bing Search is not used, even though it is active? Let me try a different challenge to verify this.
Switching back to the US environment and ask Copilot for some help with licensing questions. To my surprise, it gave me a relevant answer and allowed me to explore more details based on the previous answer. However, unlike Copilot for Microsoft 365, Copilot in Power Apps did not show me the sources of the information it used. I could only guess that it either learned from Power Apps docs or used Bing Search to find information from Power Apps docs. Next, I wanted to test Copilot in the EU environment and see how it performed there.
Okay, this is a surprising outcome. Copilot cannot assist with this task in any of the EU scenarios. I verified this in other EU environments as well, to rule out any technical glitches on my end, but I got the same results. It seems that Copilot cannot access any Power Apps documentation in the EU environment. Either Bing Search does not index them, or is not permitted to. And it seems that they were not used in a different way to train the model or provide grounding, because otherwise I would expect Copilot to give an answer close to the US-one.
The test results may vary depending on your situation, but they show the importance of testing Copilot experiences and documenting the risk assessment. I also suggest that you read this document before creating any internal documentation, as it may help you to refer to this resource and avoid updating your risk assessment docs every time Copilot improves.
That concludes my tests and observation on Copilot in Power Apps experience for building apps through conversation, but that doesn´t mean it´s the end of this series. So stay tuned for more, until then…
The Power Platform Talks 