Copilot back on the map after a false start

A session of about 35 minutes within a company-wide AI information session, to explain Copilot to the whole organisation. The underlying question was harder: an earlier speaker had painted Copilot in decidedly negative terms, so the session had to do more than explain; it had to correct that picture, with a clear view of what the included and the paid version can each do. The tech lead also had three technical questions: about connecting agents to their own systems, about an existing document extraction agent, and about how to get the business to come up with use cases itself.

The agenda was chosen together with the tech lead from a full overview of possible topics, on time and relevance. Positioning was central: not starting with technology, but with shadow AI and data leak risks, so that Copilot becomes the safe, managed alternative, followed by a factual comparison with ChatGPT. For the live demo we chose one continuous use case on the client's own policy documents: from the policies supplied we selected two comparable files and wrote out the full flow in a script of ten prompts, from summarising and comparing policies, through market data via web search and an analysis in Excel with charts, to a personalised email. The technical questions were followed by research into the connection options and into what has to be answered before an implementation can start. Six months later came an overview of all Copilot changes since, based solely on official Microsoft documentation and reduced to what matters for their end users.

Delivered: a session pack with six slides in a neutral house style, the demo script with ten prompts, a personalisation text for Copilot and a scheduled prompt for daily sector news; a technical briefing on API and data warehouse connections in Copilot Studio, an initial diagnosis for the document extraction agent and a rationale for the use case canvas approach. Several Copilot Studio agents are now in use within the organisation, including a policy conditions agent, a sales bot and a contract extractor. By separating the included and the paid version explicitly, the organisation can now assess the extension of its licences per user on demonstrated use. The engagement did not stop at the session: technical questions about agent implementations followed, as did the follow-up update and a question about the discovery approach.