Skip to content

Sort and route requests

Requests arrive in every shape: a forwarded email, a pasted Slack thread, a half-finished form, a screenshot of a complaint. Before anyone can act, someone has to read it, decide what it is, judge how urgent it is, and figure out who owns it. That reading-and-deciding step is where things stall.

This walkthrough shows you how to build a Triage Assistant that does the first pass for you. You paste in a raw request, and it hands back a clean, consistent summary: what the request is about, how urgent it is, and who should pick it up. You review it, tidy it with one click if needed, and route it to the right person or team. Anything sensitive stays out of the version you forward.

By the end you’ll have a reusable assistant your whole intake desk can share, so every request gets sorted the same way no matter who’s on duty.

A single, reusable Triage Assistant with three things attached:

  1. A system prompt that tells it how to classify and what to output.
  2. Knowledge (optional) — your triage guidelines, so its categories and owners match how your team actually works.
  3. A Guardrail Pack (optional) — so sensitive details never end up in a summary you route onward.

Once it’s built, the day-to-day flow is just: paste the request, read the result, route it.


An Assistant is a saved configuration your conversations can run against — a system prompt, an optional knowledge source, and an optional Guardrail Pack bundled under one name. Building a Triage Assistant once means everyone gets the same classification every time.

Open the assistant picker from the conversation header, choose New Assistant, and give it a clear name like Triage Assistant.

In the system prompt, describe the job in plain language. Something like:

You are a request triage assistant. For every request I give you, return a structured summary with exactly these fields:

  • Summary — one or two sentences describing the request.
  • Category — one of: Support, Sales, Billing, Bug, Feedback, Other.
  • Priority — High, Medium, or Low, with a short reason.
  • Suggested owner — the team or role that should handle it.
  • Next step — the single most useful action to take next.

Keep it concise. If the request is unclear, say what’s missing instead of guessing.

Save the assistant. You now have a repeatable triage step you can use in any conversation.

For a deeper tour of system prompts, personas, and when to attach knowledge, see Choosing an assistant.

Step 2 — (Optional) Attach your triage guidelines

Section titled “Step 2 — (Optional) Attach your triage guidelines”

If your team has written rules — how you define “High” priority, which complaints go to which desk, what counts as a bug versus feedback — give them to the assistant as Knowledge.

Add a knowledge collection containing your triage guidelines, routing matrix, or SLA policy, then link it to the Triage Assistant. From then on, the assistant classifies by your rules and can cite the line it relied on, so its decisions are explainable rather than a black box.

See Upload a file to work with for how to bring a document in, and Use Chat as an internal knowledge base for building a shared, citable collection your whole team draws on.

Step 3 — (Optional) Add a Guardrail Pack for sensitive requests

Section titled “Step 3 — (Optional) Add a Guardrail Pack for sensitive requests”

Intake requests often carry things you don’t want forwarded — card numbers, health details, personal contact information, internal account IDs. A Guardrail Pack lets you set those boundaries once and trust them on every request.

Attach a Guardrail Pack to the Triage Assistant, and the summary it produces is checked before it reaches you. Sensitive details are kept out of the routed output, so the version you hand to another team is safe to forward without anyone having to scrub it manually.


Now the everyday part. Open a conversation, make sure the Triage Assistant is selected in the header, and bring in the request.

You have two easy options:

  • Paste it. Copy the raw email, ticket, or form response straight into the message box and send.
  • Attach it. Use the + menu (or drag and drop) to attach a file — a PDF, a Word doc, a spreadsheet, an email export. Chat reads the file, then triages it. See Upload a file to work with for supported file types.

You don’t need any special command. The assistant already knows its job from the system prompt — just give it the request.

Example — pasting a ticket:

A customer wrote in: “I’ve been charged twice for my May invoice and your billing line just rings out. This is the third time this has happened and I’m about to cancel. Order #44821, account under maria.lopez@example.com.”

Within a moment you’ll get a structured summary in the assistant’s voice — for the example above, something like:

Summary: Customer reports a duplicate charge on their May invoice and an unanswered billing phone line; threatening to cancel.

Category: Billing

Priority: High — duplicate charge, repeat occurrence, and churn risk.

Suggested owner: Billing / Accounts team.

Next step: Confirm and reverse the duplicate charge on order #44821, then reply to the customer directly.

Now you’ve turned a wall of text into a category, a priority, an owner, and a clear next move — in seconds.

Not quite right? You don’t have to retype your instructions. Use the Refine this button under the response to ask the assistant to improve its own answer, or just reply in plain language:

  • “Make the next step more specific — name the exact form to send.”
  • “Re-check the priority; this customer is on our Enterprise plan.”
  • “Shorten the summary to one line for a Slack handoff.”

The output sharpens until it’s ready to route. For the full set of one-click reply actions, see the surfaces of Chat.

With a clean summary in hand, hand it off:

  • Copy it. Use the copy button on the response and paste the summary into your ticketing tool, your team’s tracker, or an email. The next person sees a clear category and recommendation instead of raw text.
  • Route it inside Chat. If your intake desk lives in SUPERWISE Chat, share the conversation or post the summary into the owning team’s Channel or shared conversation. Everyone there sees how the request was classified and who’s picking it up — no separate status update needed.

See Find and organize your conversations for sharing a conversation and creating channels, and keep a folder per queue (for example Support intake or Sales leads) so related requests stay together.


  • Build the assistant once, share it. A shared Triage Assistant means it doesn’t matter who’s on the desk — every request is sorted the same way.
  • Run a busy desk in a channel. When the whole team triages in one channel, classifications and handoffs are visible to everyone, so nothing falls between the cracks.
  • Tune the prompt over time. When a category keeps getting misread, add a line to the system prompt or a rule to the guidelines. Triage gets sharper the more you use it.
  • Let guardrails do the scrubbing. Once a Guardrail Pack is attached, you don’t have to remember to remove sensitive details before forwarding — that boundary is kept for you.

Messy, inconsistent input becomes a clear type, priority, owner, and next step — every time, no matter who’s handling it. Handoffs get faster, fewer requests get stuck, and the sensitive details that shouldn’t travel stay put. Your intake desk runs on one shared standard instead of everyone’s best guess.