1. Write down the task and its limits
Define what the role should do, what it should not do and who reviews its output. For example: the AI Support Employee answers pricing and account questions and passes refund requests to the team.
2. Give it only the right knowledge
The AI Support Employee answers from the help articles your company publishes. Review them before you start, remove outdated ones, and keep public information apart from internal company data.
3. Limit permissions
Serene works within the permissions of the person using her, so she sees only what that person may see. Give each role only what it needs.
4. Define the approval step
Actions Serene proposes, such as sending a message or changing a record, wait for a person to approve them. Decide who approves, and review the records regularly.
5. Set a cost cap
Serene uses credits from the company wallet. The owner can set a spending cap and receives a warning when usage rises.
6. Plan for correction and shutdown
Know how to fix wrong information in the help articles, how to route conversations straight to your team, and how to stop the AI Employee when needed.
Checklist before you start
A written task with clear limits.
Up-to-date help articles.
The least permissions needed.
A person responsible for approvals.
A spending cap and warnings.
A shutdown route the team knows.
Read next: AI Support Employee and AI governance.