Support leadership needs a decision, not a replay of a ticket thread. This pack turns high-risk customer tickets, account notes and internal updates into a short escalation summary with evidence, a named decision and a clear next action.
Use it when a customer is blocked, a commitment may be missed, several teams are involved, or account risk is rising. It is for support managers briefing leadership, engineering, customer success or an incident lead.
Key point
Start with evidence, not prose
Extract facts and conflicts before drafting. A concise escalation that contains one invented certainty is worse than a longer one that names the gap.
1. Gather one source set
Create a working document called Escalation source pack. Put the material in date order where possible:
- Ticket IDs, customer messages and agent replies.
- Account notes, including renewal, usage or stakeholder context where it is relevant to the issue.
- Internal investigation updates and hand-off notes.
- Any commitments already made, including a promised update time.
- Names of teams already involved.
Remove material that is unrelated to the current decision. Keep the original wording for customer claims. A customer saying that their operation is stopped is evidence of a reported impact. It is not proof of a technical cause.
If you use an integration or API to collect the material, behaviour and available features can vary. Check the current xAI documentation overview before building the collection step into a support process.
Watch out
Do not blend sources without labels
A support note, an engineering finding and a customer assertion have different weight. Keep their source identifiers and dates in the source pack.
2. Extract facts before choosing a severity
Paste the source pack into Extract escalation facts. Read the Conflicts and ambiguity section before doing anything else. It tells you whether the issue is ready for leadership or whether you first need one answer from the account owner, support agent or investigating team.
Do not ask the model to assign an internal severity unless your organisation has supplied the criteria and you have checked the result. The useful outcome here is a record of what is known, what was promised and what remains unverified.
Pay particular attention to three fields:
- Business impact stated by the customer: retain the customer’s language, then identify it as reported.
- Promises and deadlines: a missed update commitment can itself require escalation, even if the technical issue is unchanged.
- Unresolved blocker: make this specific.
Waiting for engineeringis weak.Engineering has not confirmed whether the workaround preserves data submitted after 14:00is usable.
3. Make the sequence visible
Use Build an evidence timeline when several people have touched the case or the customer has contacted you more than once. Leadership can then see the difference between a new incident and a slow-moving response.
Check the timestamps against the ticket system before you reuse the timeline. This matters when exported notes omit time zones, merge several updates into one entry or use relative phrases such as “this morning”. Do not silently convert those phrases into a precise time.
Check
A useful timeline exposes the next gap
You should be able to point to the last customer contact, the last verified internal update and the next commitment. If any is missing, the escalation needs an open question.
4. Draft only the decision-ready summary
Run Draft the leadership escalation with the fact record and timeline. Send the output to the people who can make the requested decision, not to every team who has seen the ticket.
The Decision needed section is the test. Each item should answer four questions: who decides, what they decide, why now, and what happens if nobody decides. If the real request is only for an update from another team, say that plainly. Do not disguise an information request as a leadership decision.
Use Sharpen the decision request if the draft offers several possible actions. It is especially useful where a leader must choose between assigning a specialist, approving a customer communication path or setting a response owner. The appendix separates supported options from assumptions, so the meeting does not become a debate about basic facts.
Keep customer-sensitive detail to what the recipient needs. The summary should identify the account and issue, but it does not need to reproduce a full ticket transcript.
5. Check what could make the summary wrong
Run Check the escalation before sending against the original source pack. This is not a spelling check. Look for claims that have become more definite in the summary than they were in the evidence.
The most common errors are:
| If you see this | Treat it as | What to do |
|---|---|---|
| “The defect caused the loss” | An unverified causal claim | Replace it with the observed effect, or cite a confirmed finding. |
| “Engineering will fix this today” | A commitment | Keep it only if an authorised source gave that commitment. |
| A named owner with no source | An assumption | Change it to Owner to confirm, or obtain confirmation. |
| No customer update time | A material omission | Add the next update commitment or make it an open question. |
Check
The final document has six visible answers
A reader should find customer impact, evidence, actions taken, decision needed, owner and deadline without reading the source tickets.
When it does not work
If the output is vague, do not ask for a “better summary”. Add the missing input. Paste the last customer message, the latest internal finding, or the exact decision that leadership can make. Then rerun the relevant prompt.
If the evidence conflicts, keep the conflict in Open questions and assign someone to resolve it. If no one can own the next action or deadline, send a short request for that assignment rather than presenting a false sense of control. Escalate the uncertainty itself when the customer impact is already high.