# Change Management and Lifecycle

*Audience: AI governance reviewers and use case owners.*

An approval is only as good as its last check. Change management keeps an approved AI use case honest: Relyance records what you approved, watches for anything that differs from it, scores each change, and decides which ones need a person. This page explains approval, changes, change requests, and the lifecycle stages a use case moves through.

:::info
Change management is in Beta and is available where enabled for your tenant, on top of AI Use Cases. Contact your account team for access. Where it isn't enabled, the use case drawer has no **Approval** card, **Lifecycle** field, or **Changes** tab.
:::

## How it works

1. **You approve the use case.** Approving saves a *baseline* - a snapshot of the use case as it is now: its components, their tools and data, where they're hosted, and their registration.
2. **Relyance compares every scan with the baseline.** Whenever something a use case is built from changes, Relyance looks for differences and bundles related ones into a single change - for example, a new tool on an agent together with the data categories that tool brought.
3. **Each change gets a risk score.** The score runs from 0 to 100 and maps to **Low**, **Medium**, **High**, or **Critical**.
4. **Low-risk changes are accepted by the engine.** Changes within your threshold are accepted automatically and folded into the baseline, with a short note from the **Relyance governance engine** explaining why.
5. **Higher-risk changes wait for a person.** Changes above your threshold stay open as **Needs review**, and the use case moves to **Re-review**. Accepting one updates the baseline and opens a re-assessment of the intake questions it affects.

You can also propose a change before it ships - see [Change requests](#change-requests).

## Approving a use case

The **Approval** card sits at the top of the use case's **Overview** tab. Before the first approval, it shows the steps to get there - **Register**, **Assess**, **Monitor** - and an **Approve** button.

1. Open the use case and, on the **Approval** card, click **Approve**.
2. Optionally add a **Note (optional)** to record why it was approved.
3. Click **Approve**.

Approving approves the use case as it is now. Later changes are compared against this baseline, and the next review is scheduled from the use case's risk level. If there are open changes, approving accepts all of them.

After approval, the card shows who approved it and when, and the next review date. It shows a banner only when something needs a person - open changes, or an overdue review - with **Review changes** to jump to the **Changes** tab.

The card's **⋯** menu has:

- **Check now** - once the use case is approved, look for changes immediately rather than waiting for the next scan.
- **Re-approve** - once the use case is approved, approve it again as it is now, making that the new baseline.
- **Quarantine**, **Retire**, and **Release** - see [Lifecycle stages](#lifecycle-stages).

## Lifecycle stages

The **Lifecycle** field in the drawer header shows where a use case stands. Relyance sets the first four stages for you; **Quarantined** and **Retired** are set by a person.

| Stage | When a use case is in it |
|---|---|
| **Register** | It hasn't been approved, and no assessment about it is in progress |
| **Assess** | It hasn't been approved, and an assessment about it is in progress |
| **Monitor** | It's approved, nothing is waiting for review, and its review isn't overdue. Relyance keeps checking it for changes |
| **Re-review** | It's approved, but at least one change needs review or its scheduled review date has passed |
| **Quarantined** | Someone flagged it to stop and be investigated. It stays here until someone releases it |
| **Retired** | It's no longer in use. Relyance stops checking it for changes |

To quarantine or retire a use case, choose **Quarantine** or **Retire** from the **Approval** card's **⋯** menu and confirm. To return a quarantined or retired use case to normal, choose **Release**; it goes back to whichever stage its approval and open changes give it.

:::warning
Quarantine is a governance flag in Relyance. It records that the AI system should be stopped and investigated, and shows that on the use case, but it doesn't switch anything off in your environment. Stopping the system is up to its owners.
:::

## Review dates

Approving a use case records it as reviewed and schedules the next review from its risk level. By default:

| Risk level | Next review |
|---|---|
| **High** | 90 days after approval |
| **Medium** | 180 days after approval |
| **Low** | 365 days after approval |

When the next review date passes, the use case moves to **Re-review** and the **Approval** card shows **Review overdue since** the due date. Re-approve the use case to record the review and schedule the next one. You can also edit **Last Reviewed** and **Next Review** directly on the **Governance Status** card.

## The Changes tab

The **Changes** tab lists everything that differs from the approved use case, detected or requested, newest first. Its label shows how many changes need review - for example, **Changes (2)**. Switch between **Needs review** and **All changes** at the top.

Each change card shows:

- **What changed** - the differences it bundles, such as **Tool added** or **New data category**.
- **Its source** - **Detected** by a scan, or **Requested** by your team.
- **Its risk** - the level and score, with **How the risk was scored** to expand the factors and their points.
- **The engine's note** - from the **Relyance governance engine**: why it accepted the change, or why it needs a person.
- **Its status** - see below.

| Status | Meaning |
|---|---|
| **Needs review** | Its risk is above your threshold, so it waits for a person to accept it |
| **Accepted** | A person accepted it, and it's part of the approved baseline |
| **Accepted by engine** | Its risk was within your threshold, so the engine accepted it on its own |
| **Reverted** | A detected change that no scan shows any more, closed on its own |

### Accepting a change

On a change that needs review, click **Accept**. Accepting updates the approved use case. If the change affects intake questions, Relyance also opens a re-assessment of just those questions, carrying the rest of the last intake forward, and the change offers **Open re-assessment**.

### What Relyance detects

| Change | What it means |
|---|---|
| **Component added** / **Component removed** | A model, agent, MCP server, or vendor joined or left the use case |
| **Model changed** | A component now uses a different model |
| **Tool added** / **Tool removed** | An agent or MCP server gained or lost a tool |
| **New data category** / **Data no longer processed** | A component started or stopped processing a data category |
| **Data seen in prompts** | A data category appeared in the prompts sent to a model |
| **Hosting region changed** | A component moved to a different region |
| **Endpoint moved** | A component's endpoint changed |
| **Component no longer registered** | A component lost its registered status |
| **Policy violation** | A policy violation opened on a component |
| **Became over-privileged** | A component gained more access than it needs |
| **System prompt edited** | A system prompt changed |

By default, **Tool removed** and **System prompt edited** don't raise a change on their own.

### How a change is scored

Each difference a change bundles adds points, and the total is scaled by the use case's own risk level:

- **Base points by kind.** Riskier kinds score higher - becoming over-privileged, a model change, or a component losing its registration score more than a removed component or a data category that's no longer processed.
- **Extra points for risky variants.** Sensitive data, a destructive tool (one that sends, writes, changes, or deletes things), hosting outside your allowed regions, and the severity of a policy violation each add points.
- **Use case risk.** The total is raised for a **High**-risk use case and lowered for a **Low**-risk one.

The score is capped at 100 and banded:

| Score | Level |
|---|---|
| 0-24 | **Low** |
| 25-49 | **Medium** |
| 50-74 | **High** |
| 75-100 | **Critical** |

By default, the engine accepts changes up to **Medium** on its own, and **High** and **Critical** changes wait for a person. Contact your account team if your organization needs a different threshold, review cadence, set of ignored changes, or allowed regions.

## Change requests

A change request lets your team propose a change before it ships and see what it would raise. Open the use case's **Changes** tab and click **New change request**.

### Describe the change with AI

1. In **What do you want to change in \<name\>?**, describe the change in your own words. For example: *Swap GPT-4o for a model we already approved, or remove Zendesk now that tickets live in Intercom*.
2. Click **Describe with AI**. Relyance reads your request, finds matching components in your inventory, and checks each option against what was approved. This usually takes under 10 seconds.
3. Review the draft, marked **Drafted by AI**, and adjust it if needed.

Or click **Fill in manually** to build the request yourself.

### Build or adjust the request

1. Enter a title, and under **Why**, what the change is for.
2. Under **The change**, choose an operation and what it acts on:
    - **Add a component**, **Remove a component**, or **Replace a component**.
    - **Add a tool** or **Remove a tool** - pick an agent or MCP server and type the tool name. Tick **It sends, writes, changes or deletes things** for a destructive tool.
    - **Start processing data** or **Stop processing data** - pick a data category.
    - **Change a setting** - describe a change no scan can see, such as a guardrail or a prompt change.
3. Click **Add another change** to bundle more operations into the same request.
4. Check **What this would raise** - the score, the factors behind it, and whether it's **Within your threshold** (the engine would accept it on its own) or **Above your threshold** (it will wait for a person).
5. Click **Submit request**.

### What happens after you submit

- **Within your threshold** - the governance engine accepts the request and applies it.
- **Above your threshold** - the request waits on the **Changes** tab as **Needs review** until someone accepts it.

Accepting a request approves it in advance:

- **Component changes** are applied as soon as the request is accepted.
- **Tool and data changes** are what scans see, so they're counted as approved once a scan detects them. Until then, the change shows a **Monitoring** badge listing what it's waiting for. When a scan sees it ship, the change shows the date it shipped, and no further review is needed.
- **Setting changes** can't be seen by a scan, so a person always accepts them.
