Triage a support ticket with full context

Support's worst moment: ticket arrives, you don't know if it's the customer's first complaint or their tenth, and you don't know what engineering is doing about the underlying issue. So you ask the customer to remind you, or you escalate to engineering, and the customer feels unheard either way.

Recon fixes this by making the customer's full history queryable from your AI tool.

The flow

1. Ticket arrives in your help desk.
2. Paste the ticket (or just the customer name + issue) into your AI tool.
3. Run the triage prompt.
4. Get a draft reply with sources you can verify in 30 seconds.

The triage prompt

A ticket just arrived from {customer name}:

"{paste ticket body}"

Use Recon to:
1. Pull the customer's full account memory (get_account).
2. Find any prior tickets from this customer that look similar
   (search the customer's account observations).
3. Check if any engineering work is open on this specific issue
   (impact with entity_type "linear" or "sentry").
4. Tell me how many other customers are hitting the same issue
   right now (impact with the bug or feature name).

Then draft a reply that:
- Acknowledges what we already know (don't ask them to repeat history)
- States what we're doing (cite the linear issue if any)
- Sets a realistic next-step expectation

Use the customer's own previous tone. Cite source IDs at the bottom.

What good output looks like

Triage:

Acme Corp (Pro · 52 seats · risk: 0.78). This is their third export timeout report in 10 days. Prior tickets: ZD-19421 (May 17), ZD-19488 (May 21). Both linked to engineering issue REC-447 (open).

Cross-customer impact: 8 other customers have reported the same pattern in the last 14 days. impact entity=REC-447 returned: BoltCo, Vertex Labs, Quanta, Harbor, Tilt, Mercia, Northwind, Loom.

Engineering status: REC-447 has 4 commits in the last 48h and was last updated this morning. No estimate but movement.

Draft reply:

Hi Sarah,

Thanks for flagging this again. I see you've hit the export timeout three times this month (ZD-19421, ZD-19488, and today). I want to be straight with you: we know about the root cause and engineering is actively working on it. Internal tracking is REC-447, and they shipped four commits to it in the last 48 hours. I don't have a firm date yet but I'll commit to updating you Friday with whatever progress we have.

In the meantime, the temporary workaround is to export in batches of <500 rows. I can walk you through that on a call today if that helps.

[your name]

Sources: ZD-19421, ZD-19488, current ticket, REC-447, 8 impact results.

When to escalate

After running this prompt, escalation criteria are easier:

• Engineering has no movement on the linked issue → escalate, this is now business-impacting.
• More than 10 customers affected and most are Enterprise → escalate as a P0.
• The customer is above their risk threshold AND this is their third report → escalate to leadership, not engineering. This is a relationship problem.
• Everything else: the prompt's draft reply is usually fine to send.

When this saves the most time

The biggest wins are when the AI tool surfaces something you would have missed:

• "Three other Enterprise customers also reported this in the last week". You bundle into one comms message instead of N replies.
• "This customer's power user hasn't logged in since the bug started". You flag to CS that this is an adoption risk, not just a support ticket.
• "The customer asked for this exact feature six weeks ago and we promised by month-end". You reset the timeline honestly instead of restating the broken promise.