September 8, 2025

There are a multitude of use cases for artificial intelligence with an attractive outlook. In future you will talk to a general AI which rapidly comes up with the right answer and which takes the right action. Under the hood, reasoning takes place, on demand for further details are acquired and agents work with other agents to solve your problem.

Right now, however, we are not yet at this point - even though month by month we see improvements. This means that, through good prompting, you can and need to support Copilot or other AI application to choose the right tool.

Getting Answers Based on Organizational Knowledge

A top use case is to get answers based on Graph connected knowledge or by using other RAG applications, such as our RheinInsights Retrieval Suite.

In this case, you combine a hint where to look for information together with the actual question so that Copilot queries organizational knowledge:

Search in our organizational graph sources for answers to the question how do you book a train track?

Will very likely bring up an answer based on the configured graph sources or based on M365 knowledge. If not, you might even need to tell in which source to look. If you leave out the first part, in particular in a newly started chat dialog, for instance

How do you book a train track?

chances are extremely high that you only get a general answer based on public knowledge which might not be helpful for the task at hand.

What Is the Alternative to Prompting?

For Copilot and other AI tools it is practically impossible to compute all possible solution paths to an input. There are timeliness constraints as well as constraints on computational resources. Therefore, as we saw, prompting is key to get the right answers.

If you want to however skip prompting and to always receive great answers questions which should be answered based on your organizational knowledge, you could integrate your search as a bot.
This is a traditional approach and comes close to offering a specialized interface or an enterprise search as a separate application.

Within this approach, your bot is directly integrated in Teams or Slack and there is no intent recognition in place on Microsoft’s side. Namely, the intent recognition takes place within the bot, your bot. This means that for instance an integration of the Retrieval Suite as Bot application provides right away a grounded answer for

How do you book a train track?

based on your organizational knowledge.

So in a nutshell, if you are thinking of tailored applications, such as an AI powered contract search or an IT knowledge or case deflection bots, then it might make sense to leave the tool selection to the user (i.e., choosing the right bot in Slack or Teams) so that users get more immediate answers to their questions.