AI Support Triage

Webhook receives a support ticket, an LLM classifies urgency and category, the ticket routes to the right team channel, and escalation fires automatically on SLA breach. Ambiguous cases go to human review.

llm_callrouterhuman_reviewsla_deadlinetry_catch

What this workflow does

Receive ticket — Webhook endpoint accepts POST from your helpdesk (Zendesk, Intercom, custom). Maps subject, body, and customer info to context.

LLM classification — Calls your LLM provider (OpenAI, Anthropic, etc.) with a structured prompt. Returns urgency (critical/high/medium/low) and category (billing/technical/feature/account).

Confidence check — If the LLM confidence is below 0.7, routes to human review with a 30-minute deadline. The human can override the classification.

Route to team — Router sends a Slack message to the appropriate channel based on category: #support-billing, #support-technical, #support-product.

SLA timer — Per-step deadline based on urgency. Critical: 1 hour. High: 4 hours. Medium: 24 hours. On breach, escalation fires to #support-escalation.

Log & update CRM — Writes classification result to your audit log and updates the ticket status in your helpdesk via API.

Workflow definition

Copy this JSON and POST it to /sequences.

1{
2  "id": "ai_support_triage_v1",
3  "context_schema": {
4    "ticket_id": "string",
5    "subject": "string",
6    "body": "string",
7    "customer_email": "string",
8    "customer_plan": "string"
9  },
10  "blocks": [
11    {
12      "id": "classify",
13      "type": "step",
14      "handler": "llm_call",
15      "params": {
16        "provider": "openai",
17        "model": "gpt-4o-mini",
18        "api_key_env": "OPENAI_API_KEY",
19        "system": "You are a support ticket classifier. Return JSON with: urgency (critical|high|medium|low), category (billing|technical|feature|account), confidence (0-1), summary (1 sentence).",
20        "messages": [
21          {
22            "role": "user",
23            "content": "Subject: {{context.data.subject}}\nBody: {{context.data.body}}\nCustomer plan: {{context.data.customer_plan}}"
24          }
25        ]
26      }
27    },
28    {
29      "id": "confidence_check",
30      "type": "router",
31      "routes": [
32        {
33          "condition": "steps.classify.output.parsed.confidence < 0.7",
34          "blocks": [
35            {
36              "id": "human_review",
37              "type": "step",
38              "handler": "human_review",
39              "params": {
40                "review_data": "LLM classified ticket #{{context.data.ticket_id}} as {{steps.classify.output.parsed.category}}/{{steps.classify.output.parsed.urgency}} with {{steps.classify.output.parsed.confidence}} confidence."
41              },
42              "wait_for_input": {
43                "prompt": "Review classification for ticket #{{context.data.ticket_id}}. Subject: {{context.data.subject}}",
44                "choices": [
45                  { "label": "Accept classification", "value": "accept" },
46                  { "label": "Billing", "value": "billing" },
47                  { "label": "Technical", "value": "technical" },
48                  { "label": "Feature request", "value": "feature" },
49                  { "label": "Account", "value": "account" }
50                ],
51                "store_as": "human_classification",
52                "timeout": 1800000
53              }
54            }
55          ]
56        }
57      ]
58    },
59    {
60      "id": "route_to_team",
61      "type": "router",
62      "routes": [
63        {
64          "condition": "steps.classify.output.parsed.category == 'billing'",
65          "blocks": [
66            {
67              "id": "notify_billing",
68              "type": "step",
69              "handler": "ap://slack.send_channel_message",
70              "params": {
71                "auth": { "access_token": "credentials://slack-bot" },
72                "props": {
73                  "channel": "#support-billing",
74                  "text": "Ticket #{{context.data.ticket_id}} ({{steps.classify.output.parsed.urgency}}): {{steps.classify.output.parsed.summary}}\nCustomer: {{context.data.customer_email}} ({{context.data.customer_plan}})"
75                }
76              },
77              "deadline": 3600000,
78              "on_deadline_breach": {
79                "handler": "ap://slack.send_channel_message",
80                "params": {
81                  "auth": { "access_token": "credentials://slack-bot" },
82                  "props": {
83                    "channel": "#support-escalation",
84                    "text": "SLA BREACH: Ticket #{{context.data.ticket_id}} — {{steps.classify.output.parsed.urgency}} billing issue unresolved."
85                  }
86                }
87              }
88            }
89          ]
90        },
91        {
92          "condition": "steps.classify.output.parsed.category == 'technical'",
93          "blocks": [
94            {
95              "id": "notify_technical",
96              "type": "step",
97              "handler": "ap://slack.send_channel_message",
98              "params": {
99                "auth": { "access_token": "credentials://slack-bot" },
100                "props": {
101                  "channel": "#support-technical",
102                  "text": "Ticket #{{context.data.ticket_id}} ({{steps.classify.output.parsed.urgency}}): {{steps.classify.output.parsed.summary}}\nCustomer: {{context.data.customer_email}}"
103                }
104              }
105            }
106          ]
107        }
108      ],
109      "default": [
110        {
111          "id": "notify_product",
112          "type": "step",
113          "handler": "ap://slack.send_channel_message",
114          "params": {
115            "auth": { "access_token": "credentials://slack-bot" },
116            "props": {
117              "channel": "#support-product",
118              "text": "Ticket #{{context.data.ticket_id}} ({{steps.classify.output.parsed.category}}): {{steps.classify.output.parsed.summary}}"
119            }
120          }
121        }
122      ]
123    },
124    {
125      "id": "log_result",
126      "type": "step",
127      "handler": "log",
128      "params": {
129        "level": "info",
130        "message": "Ticket {{context.data.ticket_id}} classified",
131        "data": {
132          "urgency": "{{steps.classify.output.parsed.urgency}}",
133          "category": "{{steps.classify.output.parsed.category}}",
134          "confidence": "{{steps.classify.output.parsed.confidence}}"
135        }
136      }
137    }
138  ]
139}

Credentials to configure

openaiapi_keyOpenAI API key. Swap to anthropic, google, or any supported provider.

slack-botoauth2Slack bot token with chat:write scope for all support channels

How to trigger

1// POST /triggers
2{
3  "type": "webhook",
4  "sequence_id": "ai_support_triage_v1",
5  "config": {
6    "path": "/hooks/support-ticket",
7    "hmac_secret": "credentials://zendesk-hmac"
8  },
9  "context_mapping": {
10    "ticket_id": "body.ticket.id",
11    "subject": "body.ticket.subject",
12    "body": "body.ticket.description",
13    "customer_email": "body.ticket.requester.email",
14    "customer_plan": "body.ticket.requester.plan"
15  }
16}

Works with Zendesk webhooks, Intercom webhooks, or any system that POSTs ticket data.

Customize it

Swap LLM provider — change provider to anthropic, google, or groq for faster/cheaper classification.

Add auto-response — insert an email step after classification to send an acknowledgment to the customer with estimated response time based on urgency.

Multi-language — add a language detection step before classification. Route non-English tickets to a translation step first.

Sentiment analysis — extend the LLM prompt to also return sentiment. Route angry customers to a dedicated retention channel.

More templates