AI 办公提效Advanced & Implementation
Chapter 20
20 / 23

AI Handover & On-Call Response

⏱️ 20 min

Use AI to write handover packages, on-call guides, and incident report templates

CHAPTER PRACTICE GOAL
01Work problem

Use AI to write handover packages, on-call guides, and incident report templates

02Reusable output

A reusable work template with realistic input, a defined output format and a human review point.

03Definition of done

Run one real task, verify the critical facts and record the before-and-after time.

The biggest problems with handovers and on-call are "missing context" and "slow incident response." Use AI to generate handover packages, runbooks, incident report templates, and risk reminders.

1) Handover Package Template

Prompt:

You are a handover document assistant. Generate a handover package based on the following points.
Output structure:
- Project / business overview
- Current status (in progress / to-do / completed)
- Key contacts (role / channel / available hours)
- System / documentation links
- Risks and pending decisions (with deadlines)
- Common questions and answers
Mark sections that "require human confirmation."
  • Ask AI for both a "high-level summary" and a "detailed checklist" version for different readers.

2) On-call Runbook

  • Have AI generate "alert classification -> handling steps -> escalation path -> backup contacts."
  • Include an "info collection template" (questions to ask, logs/screenshots/links needed).
  • Require AI to remind: "don't paste sensitive logs or credentials in public channels."

3) Incident Reports & Post-Mortems

Generate an incident report in this format:
1) Time of occurrence / scope of impact / affected users
2) How it was discovered / current status
3) Temporary remediation steps
4) Next steps (with owner / timeline)
5) External communication messaging (optional)
  • Post-mortem template: timeline, root cause, fix, prevention, monitoring gaps to fill.

4) Handover Meeting Checklist

  • Must-ask items: dependencies and credentials, environment differences, release process, known gotchas, forbidden operations list.
  • Have AI generate a "question list + recording form" that can be archived right after the meeting.
  • For cross-timezone handovers, ask AI for a "short summary + detailed version" to reduce reading time.

5) Risk Controls

  • Permissions and credentials: never include these in AI conversations. Point to internal password vaults / access request processes.
  • Responsibility boundaries: clarify what needs escalation to manager / security team.
  • Audit trail: record handover time, participants, and confirmed items for traceability.

6) Practice

Pick a project or system you own. Write down the core info, have AI generate a handover package + runbook + incident report template, then ask a colleague to review and fill in the gaps.


7) Worked Example: Hand Over an Unresolved Incident Across Timezones

The Sydney team is ending its shift while service errors remain above normal and the impact is not fully known. The goal is not to copy the entire chat. The next team must be able to act within five minutes.

Turn the timeline below into a cross-timezone handover pack.
Use only recorded facts. Return:
1. Current impact: confirmed / unconfirmed
2. Actions taken and observed result
3. Active hypotheses with supporting and opposing evidence
4. The next three actions with owner and expected time
5. Conditions that require escalation
6. Placeholders for dashboard, ticket, and runbook links
Never include passwords, tokens, or customer personal data.

What a Receivable Handover Contains

SectionQuestion it must answerInadequate wording
Current impactWho is affected, since when, and what proves it“The system seems unstable”
Actions takenWhat changed, what happened, and can it be rolled back“Already handled”
Active hypothesisWhy it is suspected and what evidence is missingPresenting a guess as root cause
Next actionWho will do what by when“Keep monitoring”
EscalationWhich condition requires waking which ownerA contact list without triggers

8) Receiver Read-Back

A handover is not complete when the document is sent. The receiver restates:

  1. The most important current impact
  2. Who owns the next action
  3. Which action is prohibited or requires approval
  4. When the next update will occur

If the sender and receiver give different answers, the handover is not finished.

9) Incident Information Safety Gate

  • Logs exclude tokens, cookies, phone numbers, emails, and customer message bodies
  • AI-generated commands are tested safely and approved by a human before execution
  • Unknown root cause is labelled as an active hypothesis
  • External updates use only approved impact and timing
  • Every temporary action has a record, owner, and rollback path

10) Failure Modes and Fixes

FailureCauseFix
Long document still leaves the receiver unsureIt is organised by history, not actionPut impact, status, and top three actions first
Chat speculation becomes root causeFacts and hypotheses are mixedSeparate confirmed, hypothesis, and unknown
Many contacts still produce slow escalationNo trigger is attachedState exactly when each person is contacted
Incident knowledge disappears afterwardsNo final archiveWrite timeline, decisions, and retro link back to the knowledge base

11) Chapter Deliverable

Leave a reusable Handover Pack: five-minute brief, detailed timeline, next actions, escalation conditions, and read-back record. After the incident, continue to AI Knowledge Management to turn stable learning into a searchable runbook.