Introduce the assistant's role
Microsoft's human–AI interaction guidelines organize recommendations for first contact, everyday use, errors and how a system evolves. Applying them to customer service starts with a short introduction that accurately reflects the scope.
Fictional opening example: ‘I am this team's virtual assistant. I can explain the services and organize your request. If needed, I can pass the conversation to a person.’ This wording should only be used where those options actually exist.
Distinguish answers, requests and confirmations
Saying a request was received is different from confirming that it was resolved. A progress message should match a verifiable state. If an action depends on approval, the customer needs to understand what is still pending.
- Before the action
- Explain what will happen when there is a meaningful choice. Do not hide the effect behind a generic confirmation.
- While waiting
- Report the known status and a possible next step. Avoid time estimates or availability the operation cannot support.
- When an answer is missing
- Acknowledge the limit and offer a handoff. Repeating the same question should not be the person's only option.
The handoff is part of the service
In a fictional scenario, the assistant collects a description of the problem and identifies that the team needs to review a document. The handoff should bring together the topic, confirmed information and unresolved question, respecting the access rules defined for the process.
The person taking over needs to distinguish what came from the customer from what the AI suggested. The aim is to avoid treating an assistant's assumption as fact or asking the customer to tell the whole story again.
Test the experience from beginning to end
We recommend simulating ambiguous requests, corrections and requests to speak with a person. Observe whether there is an understandable path, whether notices match what happened and whether the team can continue the conversation.
This is a service design criterion. A1's process assessment connects the interface to operational responsibilities and limits, so that clarity in the conversation reflects the work behind it.
