Skip to main content
This guide includes features awaiting release. If an option is not available in your workspace, contact StackShift Support.

​
What you can do

Enable customer-facing replies for a chosen scope, with rules for handing work to a person. Where to find it: AI → AI settings → Customer replies; AI → Conversations

​
Before you start

  • Accurate published knowledge for the questions AI should answer.
  • Workspace AI configuration permission and eligible customer-reply capability.

​
Steps

1
Review Availability and resolve the specific missing settings or requirements shown for customer replies.
2
Open AI settings and enable Reply to customers automatically for selected eligible inboxes. Only new messages in unassigned inbox conversations are eligible; staff replies pause AI.
3
Select who receives a handoff and review the customer-reply controls. Decide which questions must reach staff instead of an automated reply.
4
Use a test conversation with a routine question and another that needs a person. Check both the answer and the handoff.
5
Review AI Conversations and outcomes after activation. Disable or narrow the policy if answers are inaccurate, then improve the source knowledge before expanding it.

​
Example

AI answers opening-hours questions from a published article while a disputed charge is handled by a person.

​
Things to know

  • Staff copilot and customer replies are separate settings. Enabling copilot alone does not let AI send messages.
  • Customer replies do not automatically authorize refunds or other business actions.
  • Availability explains current access; an unavailable capability needs its stated requirement resolved rather than repeated clicking.

​
What happens next

Eligible customer messages can receive AI replies; conversations outside the policy or needing help are handed to staff.

​
If something goes wrong

  • AI does not reply: check the eligible inbox, unassigned state, automatic-reply switch, budget, channel status and handoff reason.
  • Wrong answer: pause or narrow the policy, correct knowledge and evaluate the change.

Set up AI, credits and your own API key

Choose how AI is paid for and which assistance your workspace can use.

Evaluate and improve AI responses

Compare AI behavior with examples and use outcomes to choose improvements.

Review AI actions safely

Let AI request allowed business actions and have the right person approve sensitive work.