A Practical Call Taxonomy: Categorize Calls Before You Automate

The most common mistake in healthcare phone automation isn't choosing the wrong technology. It's automating before you know what you're automating.

Practices that implement voice AI without a clear picture of their call mix end up with a system optimized for the calls they assumed they were getting, not the ones they're actually getting. The result is poor routing accuracy, frustrated patients, and a system that needs immediate reconfiguration.

The fix is simple: build a call taxonomy before you build (or buy) anything.

Why Call Mix Is Rarely What You Think

Practice managers consistently overestimate scheduling calls and underestimate billing, records, and general inquiry calls. This isn't because they're wrong about their practice — it's because scheduling calls are more memorable (they involve decisions) while routine calls (directions, hours, insurance questions) fade into background noise.

Here's a typical call mix audit result for a mid-sized specialty practice, compared to initial estimates:

The discrepancies matter because they drive agent configuration. If you build a system assuming 60% scheduling calls and 15% billing calls, but the reality is closer to 38% and 22%, your billing agent is under-resourced and your scheduling agent is over-optimized.

How to Build Your Call Taxonomy

Step 1: Two-week call audit

Pull your phone system's call records for two weeks (or have staff log calls by type in real time). The goal is a count of calls by type — not a qualitative evaluation.

Categories to track:

• Scheduling (new appointment, follow-up, reschedule, cancellation)
• Billing (balance inquiry, payment, insurance question)
• Medical records (request, transfer, portal help)
• New patient (first contact, insurance check, provider question)
• Clinical / provider message (not an emergency)
• Urgent / triage
• General information (hours, directions, services)
• Other / unclassified

Step 2: Identify sub-intents

Within each category, document the most common sub-intents. For scheduling, this might be:

• "Book new patient appointment"
• "Reschedule existing appointment"
• "Cancel appointment"
• "Check appointment time"
• "Add or change appointment type"

Sub-intents are the raw material for AI intent configuration. The more specific they are, the more accurately the system will route.

Step 3: Map escalation triggers

For each call type, document the conditions that require human escalation:

• Billing: insurance disputes, hardship requests
• Records: third-party requests requiring authorization
• Clinical: any symptom-related question
• Urgent: emergency keywords (chest pain, fall, breathing difficulty)

Escalation logic is often the hardest part of AI configuration — building it from a documented taxonomy is much faster than discovering it through failures in production.

Step 4: Document the data

Create a simple reference document: call types, volumes, sub-intents, escalation triggers. This is your AI configuration brief. It becomes the source of truth for onboarding any voice AI system.

What a Good Taxonomy Enables

A documented call taxonomy doesn't just improve AI accuracy — it clarifies your operational picture in ways that help beyond phone automation:

• Staff training: new hires know exactly what types of calls to expect and how to handle each
• Capacity planning: you know peak call types by hour and day of week
• Performance tracking: you can measure first-call resolution by type, not just aggregate
• Escalation discipline: staff know exactly when to escalate vs. handle — no guesswork