Skip to main content
Incidents represent elevated, persistent issues in your AI support operation. When a policy violation is severe enough, or when similar violations recur across multiple conversations, CraftCX promotes them into an incident—a tracked object your team can investigate, assign, and work through to resolution.
Incidents are part of the Support Observability feature set, which is currently in active development. Reach out to support@craftcx.com for early access.

​
What makes something an incident?

Not every policy finding becomes an incident. CraftCX applies a high threshold for promotion to keep your incident queue focused on what genuinely matters:
  • Recurring violations — the same type of policy deviation appears across multiple conversations
  • High severity findings — a single finding is severe enough to warrant immediate attention
  • Quality regressions — a measurable drop in AXIS scores that persists over time
This means incidents are a signal that something systemic needs fixing—a knowledge gap, a misconfigured AI behavior, an outdated policy, or a product issue surfacing through support.

​
Working with incidents

​
The incident queue

Incidents are organized into four views:
TabWhat’s shown
UnassignedActive incidents not yet assigned to a reviewer
AssignedActive incidents assigned to someone on your team
Assigned to meActive incidents assigned to you
ResolvedClosed incidents, including false positives

​
Incident details

Each incident shows:
  • Title and summary — what was detected and why it matters
  • Severity — high, medium, or low, based on the nature of the issue
  • Status — current triage state
  • Related conversations — the specific tickets that triggered the incident
  • Linked rule — the policy rule that was violated (if applicable)
  • Detected at — when the incident was first identified

​
Triaging an incident

Incidents move through a triage workflow:
StatusMeaning
NewFreshly detected, not yet reviewed
AcknowledgedSomeone has reviewed and is aware of it
Needs fixRequires action—AI config, policy update, or process change
ResolvedThe underlying issue has been addressed
False positiveIncorrectly flagged; no action needed
Once an incident reaches a terminal state (resolved or false positive), further transitions are blocked.

​
Assigning and setting due dates

You can assign an incident to any team member and set a due date. This keeps accountability clear and helps your team prioritize. Assignments and due dates are shown on the incident alongside the current SLA status.
Use the Assigned to me tab as your daily starting point. It shows only the incidents you’re responsible for, keeping your queue focused.

​
Audit trail

Every status change is recorded with a timestamp and the name of the person who made the change. This gives you a full history of how an incident was handled—useful for retrospectives and compliance reviews.

​
SLA and due dates

When an incident is created, you can set a due date to indicate when it should be resolved. CraftCX tracks the SLA status based on that date:
  • On track — due date hasn’t passed
  • At risk — due date is approaching
  • Overdue — due date has passed without resolution
Each incident links to the conversations that triggered it. Click through to view the full transcript and the specific policy findings that contributed to the incident.

​
Frequently Asked Questions

Any team member with access to the CraftCX platform can triage incidents. Incidents are created automatically by CraftCX based on policy findings and quality signals—you cannot create incidents manually.
A policy finding is a single observation about one conversation—the AI deviated from a specific policy requirement. An incident is a higher-level, persistent issue that may span multiple conversations or may represent a single very high-severity finding. Think of findings as evidence, and incidents as the cases built from that evidence.
The incident moves to the Resolved tab. It remains visible for your records and audit trail, but no further triage actions are available. If the same issue resurfaces later, CraftCX will create a new incident.
Yes. Set up a notification rule in Settings → Notifications to receive an alert via Slack, email, or webhook whenever an incident is detected. See Notifications for setup instructions.
Severity is derived from the underlying policy findings and quality signals that triggered the incident. High-severity findings—where the AI clearly violated a critical policy requirement with high confidence—produce high-severity incidents.

Policy Monitoring

Understand how policy violations are detected and surfaced as findings.

Notifications

Get alerted when a new incident is detected.