Our first impression of Microsoft 365 Copilot
As part of the early access program, we have been able to test various features of Microsoft 365 Copilot. Below are some of our findings:
Qualitative data, qualitative output
Microsoft 365 Copilot uses the accessible data within your Microsoft 365 tenant: your emails, presentations, chats, files, etc. If you ask Copilot to summarize a report you created a few weeks ago, it can access that content and summarize it for you. This becomes more difficult when there are multiple possible answers in your data. These can be duplicates of certain files, which contain both outdated and current information, or emails that contain contradictory information. Poor data quality can lead to unwanted results and can thus hinder adoption in the long term.
It is therefore useful to critically examine the quality of your data, and to ensure that outdated information is kept out of or removed from your Copilot database.
Make sure your access policies are set correctly
Another thing to consider is that Copilot only takes into account the data you have access to. This means that when you enter a prompt in Copilot where you are looking for sales reports, for example, you will only get results that you have access to. This is important to consider if you don't want your users to accidentally come across sensitive information, and you want to ensure that your prompt results are as relevant as possible. We also recommend following the Zero Trust principles to ensure that your data is safe and managed properly.
Explore and identify different use cases
Finally, we have noticed that Microsoft 365 Copilot varies in usability depending on the use cases and the specific user type or 'persona'. Microsoft 365 Copilot excels at quickly summarizing content, generating concepts, and providing insights into data. For niche applications or a limited dataset, Copilot is often not the best option. Therefore, we are also actively testing the capabilities of AzureOpenAI to meet those needs.
It is important in any case to explore different use scenarios and to identify which personas are most suitable for this. In this way, you get better Copilot results and strengthen the overall adoption of the tool.