Improved sense of time to Copilot in Teams

Those who have been using Copilot in Teams actively on chats and channels, have very likely hit this issue when Copilot doesn’t seem to understand when asked about messages in a specific time frame that was further in history. For example, asking Copilot to summarize key action items from last week worked without issues, but asking that for something that was more than a month ago provided mixed results as it sometimes seemed to work and sometimes didn’t.

In fact, Copilot’s “sense of time” in chats and channels was limited to the last 30 days of data in Copilot in Teams chats and channels.

Before this rollout, when users asked Copilot about a long time period (such as “Summarize the key action items from March”), Copilot was limited to the last 30 days of data.

Check out M365 Admin Center Message center, with message id MC801583 and public roadmap.

Message center states also other improvements Copilot in Teams is getting: Improvements to how Copilot in Teams understands and responds to time-related queries in chat and channels. With this rollout, Copilot will provide more precise and relevant responses to prompts such as “yesterday,” “last week,” and “last month.”

How is this change going to be seen in action when using Copilot in Teams? Looking at Message center’s photos there will be two key changes:

Improved sense of time

Simplified timestamp for indicating the timeframe

This update should roll out during this month (August 2024). Targeted release tenants may already have it or get it soon and general availability will be rolling during mid-August to the end of August. This will only affect people who has Copilot for Microsoft 365 -license. The update will be affecting all clients: Desktop, Mac, New Teams Web version and Teams mobile apps.

I needed to try this one early summer in the demo environment to see the impact of this change. Using Copilot side panel for channel conversation thread in Teams.

The chats and channels Copilot provided an answer that included data from December – January, but it the end date was at the end of May. It was at the end of May, when the thread had the last messages. The answer contained only information during December-January messages, so Copilot understood the timeframe already. The reason for this may be that the conversation thread is not too long and thus can be included easily to the language model. However, the result Copilot provided wasn’t fully complete.

Doing the same mid-August, we can see the timestamp has been simplified. This time the result includes all messages and their key tasks.