Inbound Call Triage Brief

Triage inbound calls to an AI agent line — classify caller intent, screening questions, escalate-vs-handle rules, message-taking script, and spam signals before the agent picks up.

Purpose

Produce an inbound triage brief when a call arrives (or is expected) on an AI agent phone line — PollyReach, Agent Phone Call, OpenPhone, Twilio, etc. Covers intent classification, first questions, escalate-to-human rules, message-taking, and spam detection.

This skill does not answer calls, provision numbers, or access telephony APIs. It gives the agent (or human supervisor) a playbook for this caller/context.

Pairs with phone-call-prep-brief (outbound) and agent-phone-call-debrief (after-call).

When to use

Use when:

• Setting up how an AI line handles incoming calls
• A call is ringing and the agent needs a quick routing script
• Defining receptionist / front-desk behavior for an agent number
• User asks to screen, triage, or route inbound callers

Safety and boundaries

Do not collect passwords, OTP codes, full card numbers, or SSNs — even if callers offer them.

Emergency: If caller describes immediate danger (medical, fire, violence), instruct transfer to emergency services (e.g. 911) — do not attempt AI-only handling.

Harassment/spam: Document disengage script; do not engage debate.

Regulated industries (health, legal, financial): default to human handoff for advice-giving.

Recording: Include disclosure line if calls are recorded (high level only).

Required inputs

1. Line purpose — support, sales, booking, personal screening, mixed.
2. Business/context — who the number represents (company, individual, service).
3. Caller hint (optional) — known number, CRM name, or “unknown.”
4. Hours — 24/7 AI vs business hours with after-hours behavior.
5. Human backup — who can take escalations and how (warm transfer, callback).
6. Must-capture fields — name, callback number, account ID, etc.

Workflow

1. Classify expected caller intents for this line (≤6 categories).
2. For each intent: handle / message / escalate.
3. Draft opening script (≤20 seconds spoken).
4. Define escalation triggers (keyword, emotion, VIP, legal threat, etc.).
5. Output triage brief (format below).

Intent categories (starter set)

Adapt to line purpose; pick 3–6:

Intent	Typical handle
Booking / scheduling	Agent collects slots → confirm
Order / account status	Verify identity lightly → lookup or escalate
Support issue	Troubleshoot tier-1 → escalate if unresolved
Sales inquiry	Qualify → book or escalate
Wrong number / spam	Polite disengage
VIP / urgent	Fast-track human

Output format

Inbound triage brief — {line or business}

Field	Value
Line purpose	…
After-hours	…
Human backup	…
Recording disclosure	yes/no + one-liner

Opening script

What the agent says when answering (friendly, states who they represent, sets expectation).

Intent routing table

Intent	First questions	Agent can resolve?	Escalate when
…	…	yes / partial / no	…

Screening questions (unknown callers)

Numbered list, minimal PII.

Message-taking template

If taking a message: fields to capture + read-back script.

Spam / robocall signals

Behaviors that trigger short disengage (silent line, obvious pitch, repeated hang-up).

Escalation pack

• Trigger list
• Warm handoff line: “I’m connecting you with …”
• If human unavailable: callback promise script

Do not do on this line

Bullets (e.g. process payments, give medical advice, promise refunds without policy).

Quality bar

• Inbound-first — opening script assumes caller initiated contact.
• Escalation is explicit — not “use judgment.”
• Short scripts — speakable in one breath where possible.
• PII minimal — collect only must-capture fields.

Examples

Good opening: “Thanks for calling Acme Support — I’m the automated assistant. I can help with order status or scheduling a callback. What are you calling about today?”

Bad opening: “Hello, how can I help?” (no context, no boundaries)

Good escalate trigger: “Caller requests chargeback over $500 or mentions attorney.”

Bad escalate trigger: “If the caller is angry.” (too vague)