Triage Graphs

Triage Graphs: Troubleshooting Your Agent Can Follow

A knowledge base tells a caller what your policy says. It cannot work out why their connection keeps dropping. A Triage Graph maps the problem itself — the symptoms people report, the causes behind them, and the resolutions that genuinely fix them — so an agent can reason through a fault instead of reciting a document.

Ask it something vague and it asks a better question back. “No lights or fans at all?” narrows a complaint to one specific symptom before anything is proposed, the way a competent engineer would.

Graphs are yours to build, test and export, and the categories are free-form — the same engine that triages a dead router will walk an HR or finance question just as well.

A network engineer tracing a fault at a patch panel, the kind of troubleshooting path triage graphs describe
How it is built

How triage graphs are built

Three kinds of node and the links between them. That is the whole model.

Symptoms in the caller’s words

Name the complaint the way someone would actually say it, and give it the keywords they use. Up to twenty per symptom, so “no internet” and “wifi dead” land in the same place.

Causes worth ruling out

Behind each symptom sit the things that cause it, each carrying a diagnostic question the agent can ask to confirm or dismiss it.

Resolutions with real steps

An ordered list of steps and an estimated time to fix, so the person on the phone knows what they are in for before you begin.

Questions that narrow

Link one symptom to another with a branching condition. A vague report becomes a specific one before a single fix is suggested.

Ranked, not listed

Every link carries a confidence and a priority, so candidate causes come back most-likely-first rather than in the order you happened to type them.

Tickets and escalation

A resolution can require a ticket in WHMCS, ServiceNow or Zendesk, or hand the conversation straight to a named agent.

In the agent

Wired into a live conversation

A graph is not a document. It runs while the caller is still on the line.

Point a flow node at it

In the flow builder, a function node connects to a named graph. Mid-conversation the agent queries it and works from what comes back.

Test before you ship

Run any query against a graph in-app and inspect exactly what it matched, what it ranked, and what it would have asked next.

Import, export, seed

Move a graph between environments as a file, or start from a seeded default and edit it down to your estate.

Questions

Common questions about Triage Graphs

What teams ask before putting a graph behind a live number.

A knowledge base answers a question that has an answer written down. A Triage Graph handles a problem that has to be narrowed before it can be answered — it asks clarifying questions, weighs competing causes, and proposes a fix with steps.

No. Categories are free text, so the same symptom→cause→resolution shape works for HR, finance, legal or facilities questions. IT is simply the example most people recognise.

Each link between a symptom and a cause carries a confidence score and a priority. Matching a caller’s description to a symptom produces a match score, and candidate causes come back ranked and de-duplicated.

A resolution can be marked as requiring a ticket, and the graph names which system — WHMCS, ServiceNow or Zendesk. It can also escalate directly to a named agent instead.

Yes. Graphs are named and kept separate, so you can run one per product line, per department or per customer, and point different agents at different graphs.

Put a graph behind your next support call

Guided troubleshooting beats a well-written document the moment a problem has more than one cause.

Start with one recurring fault — the one your team answers every week — and map its symptoms, causes and fixes. Test it in-app until the ranking looks right, then point an agent at it and listen to the first fifty calls. Widen from there.