Troubleshooting guide
AI Automation: Diagnostic Guide
AI Automation: Diagnostic Guide organizes the decisions that matter for teams processing documents, email, content, support requests, leads, files, and messy operational data: the current workflow, ownership, implementation choices, rollout risk, and acceptance evidence.
Working artifact
AI Automation acceptance checklist
Turn broad completion claims into checks that a project owner can repeat after handoff.
| Acceptance check | Evidence | Responsible owner |
|---|---|---|
| Prove document classification and extraction | Repeat the affected journey and test staff manually interpret repetitive documents | Owner of LLM and classifier integration |
| Prove email routing and support triage | Repeat the affected journey and test rules fail on natural-language variation | Owner of OCR and document pipelines |
| Prove content transformation and report generation | Repeat the affected journey and test AI output enters systems without review | Owner of rules engines and workflow orchestration |
Record the symptom before changing it
Staff manually interpret repetitive documents. Confirm who encounters it, where it occurs, and what changed before it appeared. Then distinguish the visible symptom from dependencies such as LLM and classifier integration.
- Save exact error text
- Record the last known working date
- List recent code, content, vendor, DNS, or account changes
Separate reachability, data, and behavior
For AI Workflow Automation, confirm account ownership, current exports or backups, recovery options, and recent changes before touching production. Preserve exact errors and timestamps that may disappear after a restart or update.
- Reachability check for rules engines and workflow orchestration
- Data or content check related to document classification and extraction
- Behavior check for email routing and support triage
Use stop conditions
Frame the first scope around document classification and extraction and one observable acceptance journey. Treat email routing and support triage as a later phase unless the evidence shows it is a true dependency.
- No confirmed backup
- Unknown production ownership
- Security or payment data may be involved
Verify the repair in the real journey
Repair fits when the core remains sound. Extension fits when the boundary around LLM and classifier integration is understood. Replacement fits when ownership, architecture, or operating risk prevents a responsible change.
- Document classification and extraction
- Content transformation and report generation
- Confidence thresholds and review queues