Skip to main content

Change Risk Advisory

An AI agent that reads every Change Request as it's being written, scores it across about ten signals, and gives the requester (and later the approver) a clear read of the risk plus specific mitigations to address. When the same change pattern runs cleanly for long enough, the agent proposes it as a Standard Change Candidate so future identical changes can skip CAB and ship faster.

It runs on the same AI-driven case-cluster engine that powers Problem and Major Incident Detection. The engine is general-purpose; this feature applies it to Change Requests instead of Incidents.

Where it shows up

SurfaceWho sees itWhat appears
Banner on the Change Request formRequesterRisk band (Low / Medium / High) plus a list of specific mitigations to address before submitting
Banner on the approver screenApproverSame risk band, with mitigations grouped, and any mitigation the requester hasn't addressed called out as outstanding
CAB review viewApprover / CAB chairAggregated risks for the change under review and an inline Suggest review action that drafts a note back to the requester
Problem / Major Incident / Change Management DashboardChange ManagerStandard Change Candidates surfaced from successful change patterns, with frequency, success rate, and rollback history

The risk read happens as the requester writes, not just at submit time. Adding a rollback plan or attaching a test report immediately lowers the score and updates the banner.

Personas served

  • Requester sees the risk band on the change form and the per-signal mitigations. They can fix the issues in line before submitting.
  • Approver sees the same risks with their resolution state. Mitigations the requester addressed are checked off; outstanding ones are highlighted. The Suggest review action drafts a note citing the outstanding items so the approver doesn't have to write it themselves.
  • Change Manager uses the dashboard to monitor risk trends and to convert proven-safe change patterns into Standard Change Candidates.

What you can do with this feature

CapabilityPage
The ~10 signals, how they become a risk band, outcome scoringRisk scoring
AI-generated mitigations on the form, on the approver screen, at CABMitigations
Detecting safe patterns, the dashboard surface, runtime policy rulesStandard Change Candidates
Per-tenant signal toggles, weight tuning, thresholds, policy rulesConfiguration

How it relates to Problem and Major Incident Detection

The same engine groups incidents into clusters and groups changes into pattern candidates. The difference is what it does with the cluster:

FamilyCluster signalAction
Problem / MI DetectionCluster of similar incidents in a short windowSurface as MI or Problem candidate; agent declares
Change Risk AdvisoryCluster of similar past changes with outcome dataSurface as Standard Change Candidate; manager accepts

The same clustering also powers a defensive signal on individual Change Requests: when a new change resembles a past cluster that produced incidents or rollbacks, the risk score goes up before the requester even submits.

Privacy

The clustering embeddings the engine uses to compare changes are stored isolated per tenant. No cross-tenant pattern matching happens. Your change history is only ever compared against your own change history.