Catch angry support tickets before the customer escalates

Build me an app my support leads open every morning to catch the tickets that are about to blow up, before the customer escalates. The main surface is a triage board of open Zendesk tickets ranked by escalation risk, which a lead works from top to bottom until their queue is empty.

Load the board from Zendesk with Search Tickets for open and pending tickets, scoped to the group or brand the signed in lead has selected, and use List Tickets as the fallback when no filter is set. For each ticket, pull the thread with List Ticket Comments and keep only the comments written by the end user. Never score agent replies, since a calm professional response would otherwise mask how upset the customer actually is. Send that customer-only text to JigsawStack Sentiment Analysis, which returns a document level sentiment score and emotion alongside sentence level results. Combine the document level negativity with how long the customer has been waiting since the last public agent reply into a single escalation risk score, and sort the board so the angriest and longest waiting tickets sit at the top.

Each row shows the requester, the subject, the dominant emotion, the sentiment score, hours waiting, current priority, the group or brand, and a mood trend arrow. Compute the arrow by scoring the customer comments posted after the last agent reply separately from the ones posted before it, then show whether the thread got worse, got better, or held steady since your team last responded. Treat a thread that turned more negative after an agent reply as the strongest escalation signal on the board.

Clicking a row opens the full conversation in a panel beside the board, rendered in order with end user messages and agent replies visually distinct, built from List Ticket Comments plus Show Ticket for the ticket properties. Above the transcript, show a JigsawStack Summarize Text recap of what the customer actually wants, generated from the customer's own messages rather than the whole thread. Use the sentence level output from Sentiment Analysis to find the single most negative sentence in the conversation, highlight it in place in the transcript, and repeat it in a callout at the top of the panel so the lead immediately sees the line that tipped the ticket negative.

From both the row and the detail panel, let the lead raise the ticket priority, add an escalation tag, post a public reply to the customer, and post an internal note, all through Update Ticket. When adding the tag, preserve the ticket's existing tags instead of overwriting the array. Also give them a button that sends a heads-up to the on-call channel using Slack Send a Message, prefilled with the ticket link, the requester, the risk score, the dominant emotion, and the one line recap, and editable before it goes out.

Add a Draft a recovery reply button that kicks off a background agent. The agent reads the entire comment history for that ticket with List Ticket Comments, works out what went wrong and what the customer is asking for, and drafts an apology that names concrete next steps rather than generic sympathy. It must post that draft onto the ticket as an internal note through Update Ticket, never as a public reply, so a human approves and sends it. Show the run status on the row while it works, and drop the finished draft into the detail panel when it lands so the lead can read it in context.

Keep a per-user reviewed today marker so a lead can clear a row from their own queue once they have worked it, without affecting what a teammate sees, and show a running count of what is left. Reset those markers each morning so the board refills with what needs attention. Remember each lead's group or brand filter between sessions. Cache the sentiment and summary results per ticket against the id of the most recent comment, so reopening the board does not rescore threads that have not changed since the last look.