Skip to content
agentusehub
OperationsAdvancedAgencies

Turn client Slack requests into a triaged queue

An n8n agent reads requests in shared client channels, classifies and sizes them, and files them in Airtable with an owner and due date.

Time saved
4h / week
Setup time
4 hours
Running cost
$10/mo
Cost range
Under $25/mo

The problem it solves

Agency clients drop requests into shared Slack channels at all hours: copy tweaks, urgent bug reports, ideas for new campaigns. Account managers scroll back every morning to work out what was asked, and small requests slip through. Nobody can see how much unplanned work each client is generating.

How it flows
  1. New Slack message
  2. Is it a request?
  3. Classify + size
  4. Queue in Airtable

Step-by-step setup

  1. Self-host n8n

    Run n8n with Docker on a small VPS behind HTTPS, since Slack only delivers events to public HTTPS URLs. Store the Slack, Anthropic and Airtable credentials in n8n's credential manager.

  2. Listen to client channels

    Add a Slack Trigger node for new messages and keep only channels whose names start with client-. Drop bot messages and thread replies so the agent sees only new top-level messages.

  3. Filter out chatter first

    Add a Basic LLM Chain node with the request filter prompt below, which answers REQUEST or NOT_REQUEST. Only requests continue to the agent, which keeps model costs low.

  4. Configure the AI Agent node

    Connect an Anthropic Chat Model to an AI Agent node and give it two tools: an Airtable search tool to find duplicate open requests and an Airtable create tool to add records. Paste the triage prompt as the system message.

  5. Assign owners from a lookup

    Keep a Clients table in Airtable with the account manager and SLA hours for each channel. The agent sets Owner and Due Date from that table instead of guessing.

  6. Confirm in the thread

    Add a Slack node that replies in the original thread with the request ID, a one-line summary and the due date. Tick a Misfiled checkbox in Airtable for wrong calls and review them every Friday.

Example prompts

Triage system message
You triage client requests for {{agency_name}}. Each input is a Slack message from a client channel.

For each request:
1. Search the Requests table for an open request from the same client with a similar summary. If one exists, do not create a new record; return its ID.
2. Otherwise create a record with:
   - Summary: one sentence starting with a verb
   - Type: Bug, Content change, New work, Question or Access
   - Priority: Urgent only if the client says something is broken or live; otherwise Normal
   - Effort: S (under 1 hour), M (half a day) or L (more than a day)
   - Source link: the Slack message permalink
3. Mark New work with Effort L as "Needs estimate" so the account manager can quote it.

Never promise a due date sooner or later than the SLA in the Clients table.
Request filter
Read this Slack message from a client channel. Answer REQUEST if the client is asking for something to be done, changed, fixed or answered. Answer NOT_REQUEST for thanks, greetings, acknowledgements or general chat. If the message both thanks the team and asks for something new, answer REQUEST. Reply with one word only.

Message: {{message_text}}
Sales · Beginner
  1. Form submit
  2. Fetch company site
  3. Claude scores fit

Auto-qualify inbound leads before a rep sees them

Score every demo request against your ICP, update HubSpot and ping the right rep in Slack within a minute of the form submit.

n8n · Claude · HubSpot · Slack
Operations · Beginner
  1. Call ends
  2. Transcript to Claude
  3. Notes + field updates

Meeting notes that update the CRM for you

After each sales call, Zapier sends the transcript to Claude, which writes the notes and updates next steps and deal fields in Salesforce.

Zapier · Claude · Salesforce
Customer Support · Intermediate
  1. New ticket webhook
  2. Claude classifies
  3. Draft from help docs

Triage support tickets and draft first replies

Every new Zendesk ticket gets a category, priority and sentiment tag plus a draft reply in an internal note before an agent opens it.

Zendesk · n8n · Claude