Building an AI Receptionist: Zoom Virtual Agent Concierge

Important Note: This guide is tailored for the ZVA 3.0 experience. If you’re using 1.0 or 2.0, please let me know in the comments and I’ll happily do a write-up on how to build an AI Receptionist for your ZVA version!

Why Build an AI Auto-Receptionist?

If you manage Zoom Phone, you already know the Auto Receptionist. It’s the front door of your phone system — callers dial in, listen to menu options, press a number, and get routed. It works. It’s reliable. It tells you almost nothing about why people are actually calling.

Zoom Virtual Agent (ZVA) changes that. Instead of routing calls based on key presses alone, a virtual agent listens to what callers say, understands their intent, and routes them accordingly — while capturing data on every interaction. It even assists with deflection, by being empowered to handle FAQs that bog down your frontline teams. The result is the same routing your team already manages, plus analytics that reveal what your callers actually need, enabling you to take your customer experience further.

This guide walks you through the entire process of building a ZVA-powered auto-receptionist, from selecting your first Auto Receptionist, to reviewing caller intent data in CX Analytics. This article is designed to be your guide to getting that first AI auto-receptionist up and running.

What You’ll Need Before You Start

• An active Zoom admin account

• Access to Zoom Phone + Zoom Virtual Agents

• Your company’s FAQ page URL or a document containing your most common caller questions

• A list of the extensions your Auto Receptionist currently routes to (known as an IVR Menu)

• An existing auto-receptionist in Zoom Phone (optional) 

Step 1: Choose Your Starting Auto Receptionist

Don’t try to convert every Auto Receptionist at once. Pick one. Ideally, choose one that gets a moderate volume of calls and has a few distinct IVR menu options — enough variety that the analytics will help you surface some new insights. More on this later, but for now, let’s keep it simple.

→ Zoom Admin Portal → Phone System Management → Auto Receptionists

Open the Auto Receptionist you’ve selected and take a look at the existing IVR menu (if you’re starting from scratch or you use a third-party for your auto-receptionist, still worth following along here. You’ll be getting started on Step 2 when we go to build!). Note the options and where each one routes. You’ll replicate these same destinations in your virtual agent, so nothing changes for your callers or your team — the routing stays the same.

What to note from your existing IVR

• Each menu option and where it routes (call queue, extension, voicemail, etc.)

• Your current business hours schedule

• The greeting message callers hear today

• Any overflow or after-hours routing you have configured

Step 2: Create the Virtual Agent

This is the step that catches most people the first time. In your Auto Receptionist settings, you’ll see a button labeled Add Virtual Agent. It looks like it’s going to let you build a new agent right there, right? It doesn’t. That button attaches an existing agent to the receptionist. To create a new one, you need to go to the Virtual Agent section of the admin portal.

→ Zoom Admin Portal → Virtual Agent → Voice Agent → AI Receptionist 

If you’ve ever built a Call Queue and then assigned it to a phone number, this is the same pattern. You build the agent first, then attach it to the Auto Receptionist. The workflow is: create the agent, configure it, test it, and only then connect it to a live number.

Step 3: Select a Template and Customize Your Greeting

Choosing a Template

When you create a new virtual agent, you’ll be prompted to select a template. Templates come pre-loaded with guidance — essentially training notes that tell the AI how to behave. Think of it as the cheat sheet you’d give a new receptionist on their first day.

Choose the template closest to your business type. If you’re running a retail location, pick the Retail template. A professional services firm might start with a general template and customize from there. The template isn’t permanent — it’s a starting point you’ll refine. Don’t overthink it!

Writing an Effective Greeting

The greeting is the single most important configuration choice you’ll make, and here’s why: it determines the quality of your analytics data.

The default greeting is something like “How can I help you today?” That sounds fine, but in practice, callers respond with one-word answers: “Bakery.” “Sales.” “Billing.” That gives you no more insight than a key press would.

Selecting a Voice

You’ll choose from a library of available voices. Custom voice uploads aren’t available yet, so pick one that fits your brand and region. If you’re a local business, choose a voice that sounds appropriate for your area — a mismatch between your brand identity and your virtual receptionist’s voice is one of the first things callers notice.

Step 4: Upload Your Knowledge Base

The knowledge base is where you teach the agent what to say. If it helps, think of this as the right side of the brain — language, context, and content. When a caller asks a question, this is where the agent looks for the answer. For our first AI auto-receptionist, we typically want to start with our company’s FAQs page. It’s a great place to start because it provides so much deflection and value at such a low level of complexity. 

You have two options for populating the knowledge base:

1. Sync a website URL — the agent pulls content directly from your site

2. Upload content manually — paste or upload FAQ documents directly

Your FAQ already covers the most common reasons people call. Start there, review the analytics after a week or two, and add more content if you see gaps.

Keeping the knowledge base current

Treat the knowledge base like a living document. Whenever your website FAQ changes, update the agent’s knowledge base to match. If a customer would get a different answer from your website than from the virtual agent, something is out of date. Build a simple review cadence — monthly is a good starting point, or whenever you make significant changes to your FAQ page.

Step 5: Configure Skills — Your Routing Logic

If the knowledge base is the right brain (what to say), then skills are the left brain (what to do). Skills define the actions the agent can take, and for Zoom Phone administrators, this is where everything will start to feel familiar.

Each skill maps directly to an IVR action you’re already managing: transfer to a department, route to an extension, send to a call queue. The difference is that instead of a key press triggering the action, the AI determines the caller’s intent and triggers it automatically.

Use your existing routing tools

This is important: You’re going to include give the virtual agent guidance on which should route to which department(s). Start with basic phrases like “bakery” or “cake”. Even if someone were to say something you didn’t inform the AI on, like “pink icing”, the AI is capable enough to use the topics you have provided across to determine which call routing makes the most sense. 

Assuming we don’t need to set up routing by name or scheduling (which are fairly straightforward!), we are now all set to finish our agent build. You will be prompted to select if you want to replace a current auto-receptionist or set this up as a new one from here. Congrats, you’ve just published your first AI Auto-Receptionist!
 

Step 6: Test Before Going Live

The testing interface lets you simulate a call. Type or speak a phrase like “I need to reach the bakery” and verify that the agent routes you correctly. Then try a few more scenarios:

• A straightforward department request (“I need to talk to someone in the deli”)

• A question the knowledge base should answer (“What are your hours?”)

• An ambiguous request (“I have a question about my order”)

• Something outside the agent’s scope (“I want to file a complaint”)

What to look for

Pay attention to more than just routing accuracy. Listen for whether the greeting sounds natural, whether the agent correctly identifies your business name, and whether the handoff to a live queue feels smooth. Small issues are easy to fix now and painful to discover from a real customer.

Step 8: Review Analytics and Iterate

This is where the investment pays off. CX Analytics gives you something you’ve never had with a traditional IVR: visibility into caller intent.

→ Analytics & Reports → Virtual Agents → Query Insights

Query Insights shows you what callers actually said — not just which department they routed to, but the specific question or request that prompted the call. Where your old IVR might tell you that 200 people pressed 1 for Member Services last month, Query Insights can show you that 11 of those callers were specifically asking about gas prices, 8 were asking about membership renewal, and 15 were trying to reach a specific person.

That’s the difference between knowing where callers went and understanding why they called. And that understanding is what lets you make informed decisions about staffing, training, FAQ content, and service design.

What to do with the data

• Review weekly for the first month to spot patterns and catch any gaps in your knowledge base or skills.

• Update your greeting if callers are still giving one-word answers. The greeting is your biggest lever for analytics quality.

• Add knowledge base content if you see repeated questions the agent couldn’t answer. This is where deflection rates are impacted.

• Build on your skills to give your agent new capabilities. The possibilities here are nearly endless, which is why we started with the basics today!

• Export and share insights with department leads. The data is useful beyond IT — it tells operations and customer service teams what customers actually need.

What’s Next?

Once you’re comfortable with your first virtual agent, the natural next step is expanding it’s capability with more skills and knowledge or moving on to additional Auto Receptionists for your business. Each new agent you build will go faster because the patterns are the same — knowledge base, skills, guidance, test, deploy.

I truly hope this article helped you in your AI journey with Zoom. If it did, please drop a comment or a like, I’ll be active in the comments replying to any questions you have!