> ## Documentation Index
> Fetch the complete documentation index at: https://docs.stackshift.cloud/llms.txt
> Use this file to discover all available pages before exploring further.

# Teach AI a repeatable support procedure

> Define the steps AI should follow for a specific kind of customer problem.

<Note>
  This guide includes features awaiting release. If an option is not available in your workspace, contact StackShift Support.
</Note>

## What you can do

Define the steps AI should follow for a specific kind of customer problem. Where to find it: AI → Procedures

## Before you start

* AI configuration access and a connected provider.
* The business tools and allowed actions needed for the procedure.

## Steps

<Steps>
  <Step>
    Create a procedure from a relevant template. Choose its eligible inbox and describe when it should apply, such as a failed order that can safely be retried.
  </Step>

  <Step>
    Write the instructions and choose the exact connected operations it may use. Do not ask customers to supply internal account identifiers as proof of ownership.
  </Step>

  <Step>
    Inspect the proposed steps before applying them to the draft. Check information collection, conditions, approvals, handoff and how success will be verified.
  </Step>

  <Step>
    Open Versions & tests and simulate ordinary cases, missing information, rejected approval and an uncertain external result. Correct the draft until the outcomes match your intentions.
  </Step>

  <Step>
    Review and publish the version. Check Procedure outcomes and active cases. Pause it to stop new cases or activate a previous published version when needed.
  </Step>
</Steps>

## Example

A failed-job procedure verifies ownership and failure state, requests a permitted retry, then checks the result before reporting success.

## Things to know

* A procedure is a problem-solving process; an ordinary workflow responds to a selected event.
* Publishing requires the relevant customer-AI and connector settings. A simulation does not authorize live actions.
* Pausing stops new cases; it does not undo an action or automatically change the version used by a case already running.
* Staff can answer a pending consultation, take over, resume or cancel a case where allowed. Customer confirmation and staff approval serve different purposes.

## What happens next

<Check>
  AI has a reviewed process with bounded actions and a defined handoff instead of open-ended permission to act.
</Check>

## If something goes wrong

<Warning>
  * Procedure not selected: check publication, applicability, eligible inboxes and customer-AI settings.
  * Action unknown: reconcile its original result before another attempt.
  * Cannot complete: verify success using the expected evidence instead of assuming a queued action succeeded.
</Warning>

## Related guides

<CardGroup cols={2}>
  <Card title="Review AI actions safely" href="/stackshift-support/ai-actions-and-approvals">
    Let AI request allowed business actions and have the right person approve sensitive work.
  </Card>

  <Card title="Evaluate and improve AI responses" href="/stackshift-support/ai-evaluations">
    Compare AI behavior with examples and use outcomes to choose improvements.
  </Card>

  <Card title="Support procedures and resolution" href="/stackshift-support/agentic-procedures">
    Create, test and publish support procedures with approved actions and human handover.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.