How to Build an AI Voice Agent That Qualifies Leads and Books Meetings Automatically

A step-by-step breakdown of how we built Athena, an AI voice agent for a travel agency, using JustCall, Zapier, ActiveCampaign, and Calendly.

Why We Built This

You are 21 times more likely to qualify a lead if you respond within five minutes.

Our client, Greeking.me, is a travel design agency in Athens that gets 350–400 leads per month — but with a 24–48 hour average response time, many of those leads were gone before anyone followed up.

To solve this, we built Athena — an AI voice agent that calls leads within minutes of form submission, qualifies them with a few questions, and books a consultation call with a travel designer.

The result? Lead response time reduced to five minutes.

Here’s exactly how we did it.

The Tech Stack

Why Zapier over Make.com or n8n?

JustCall’s outbound AI voice agent integration was native on Zapier, which meant no custom API work needed. If that integration didn’t exist, we could have built it via JustCall’s API endpoint, but native is always faster and more reliable for client projects.

The Architecture: Two Automations

The entire system runs on two Zapier automations:

• Automation 1: Instant AI Call (the main flow)
• Automation 2: Book Meeting (triggered during Athena’s call)

Automation 1: Instant AI Call

This is the core automation. It listens for new leads and triggers Athena to call them.

Step 1: Trigger — New Contact Added to ActiveCampaign List

When someone fills out the trip form on greeking.me, they’re automatically added to a specific ActiveCampaign list called “Greeking.” This is our instant trigger.

Step 2: Wait 3 Minutes + Check Booking

Some leads book a Calendly call right after submitting the form. We wait 3 minutes, then check if they’ve already booked. If yes → terminate. No need to call them.

Step 3: Phone Number Normalization

The form already validates phone numbers, but we double-check by normalizing the format. International leads enter numbers in all kinds of formats — this step ensures JustCall can dial them.

Step 4: Calling Hours Check (8am–8pm Local Time)

This is critical for international leads. We determine the lead’s timezone from their country code, then check if it’s between 8am and 8pm in their local time. If it’s outside those hours (say, 2am in their timezone), we delay the call until 8am. You can’t call people at 3am — it’s illegal in most countries and a terrible experience regardless.

Step 5: Second Booking Check

If we delayed the call for hours (waiting for 8am in their timezone), the lead might have booked in the meantime. We check again before proceeding.

Step 6: Fetch Calendly Availability

We hit a custom Calendly API endpoint to pull all available time slots for the next 7 days across the travel designers’ Round Robin calendar.

Step 7: Parse Available Slots with JavaScript

The Calendly API returns a massive JSON response. We wrote a JavaScript function in Zapier to parse and compact this into a clean, usable format that we can pass to the AI agent as dynamic variables.

Step 8: Handle No Availability (Edge Case)

If somehow there are zero available slots in 7 days (rare, but possible), we email the internal team for manual follow-up. Athena still calls the lead to collect qualification info — no lead is wasted.

Step 9: Trigger JustCall AI Voice Agent

This is where Athena comes in. We trigger the JustCall “Create AI Voice Agent Call” action with:

• The AI agent ID (Athena)
• The lead’s phone number
• Consent flag (true, from form submission)
• Dynamic variables: lead name, trip details, travel dates, destination, party size, available Calendly slots, and detected timezone

Athena now has all the context she needs before the call even connects.

Setting Up the AI Voice Agent in JustCall

JustCall’s AI Voice Agent has 5 configuration tabs:

1. Behavior

Define the agent’s role, personality, and conversation style. Athena is friendly, professional, and concise. She identifies as an AI travel assistant upfront.

2. Knowledge

Business context and knowledge base. We added Q&As about Greeking.me’s services, common traveler questions, and objection handling (e.g., “Are you a real person?” → “I’m Athena, Greeking.me’s AI travel assistant. I’m here to collect a few details so we can match you with the perfect human travel designer.”)

3. Call Flow

The conversation script. We used the “script-based personalized message” greeting mode. The script follows a clear flow:

1. Greet by first name + reference trip details from the form
2. Disclose AI status naturally after 1–2 exchanges
3. Ask qualification questions (date flexibility, prior visits, tickets booked, special requests)
4. Present 2–3 available time slots in the lead’s local timezone
5. Book the call or handle objections
6. Confirm booking with details

The script uses dynamic variables throughout (CallerName, TravelDestination, TravelDate, AvailableSlots, etc.) so every call feels personalized.

4. Actions

• During call: Book a meeting (triggers Automation 2)
• After call: Send booking confirmation SMS, handle unqualified/interested leads, save qualification answers, extract caller information and call summary

5. Advanced Settings

Timezone, responsiveness, noise cancellation, voicemail detection, max call duration, end call on silence. We initially enabled voicemail handling but turned it off to save credits ($1/min adds up fast on voicemails).

Automation 2: Book Meeting

This automation fires when Athena books a call during the conversation.

Step 1: Webhook Trigger

JustCall sends a webhook when Athena’s “Book Meeting” action fires during a call. Payload includes: invitee email, name, phone, timezone, and requested start time.

Step 2: Timezone Conversion (JavaScript)

The lead says “Wednesday at 3pm” — but that’s 3pm in their timezone. Calendly needs UTC. We wrote a JavaScript function to convert:

const localTime = inputData.startTime;
const tz = inputData.inviteeTimezone;
const date = new Date(localTime);
const utcDate = new Date(date.toLocaleString('en-US', { timeZone: 'UTC' }));
const tzDate = new Date(date.toLocaleString('en-US', { timeZone: tz }));
const offsetMs = utcDate - tzDate;
const correctedUTC = new Date(date.getTime() + offsetMs);
output = { startTimeUTC: correctedUTC.toISOString().replace('.000Z', '+00:00') };

Step 3: Book on Calendly

Trigger “Book Meeting for Invitee” on Calendly with the converted UTC time, lead details, and the Round Robin event type.

Step 4: Success/Error Split

• Success: Add “Booked Via Athena” tag in ActiveCampaign for attribution tracking
• Error: Send email to internal team with all lead details for manual follow-up

Post-call automations:

• Booking confirmation email (via Calendly)
• SMS reminder 1 day before the call
• SMS reminder 1 hour before the call
• For unsuccessful bookings: SMS with direct Calendly booking link

Voice Selection: The ElevenLabs Detour

We initially wanted Athena to have a Greek accent — fitting for a Greek travel agency. JustCall supports custom voices via ElevenLabs, and we tested it on their trial plan. It worked beautifully.

But ElevenLabs custom voices require JustCall’s most expensive plan. When the trial ended, we switched to an American English accent. The client later requested we make it “more American”, which we adjusted.

Lesson: Budget for voice customization if accent matters to your brand.

Cost Reality: JustCall Pricing

Let’s talk about the elephant in the room. JustCall charges approximately $1 per minute for AI voice agent calls. Competitors like VAPI charge around $0.09 per minute — that’s more than a 10x difference.

For a client already on JustCall’s platform, the convenience of native integration made it the right choice. But for new projects, especially high-volume ones, the math doesn’t work. If your leads average 3-minute calls and you get 400 leads/month, that’s $1,200/month on JustCall vs. ~$108/month on VAPI.

Our recommendation: If you’re not already locked into JustCall, explore VAPI or similar platforms. If you are, the native Zapier integration saves enough development time to justify the premium for smaller volumes.

What I Learned Building My First AI Voice Agent

This was my first time building an AI voice agent for a client. No one showed me how — I learned every tool from scratch, did all the research myself, and built it end to end.

Here’s what I’d tell someone building their first one:

1. It’s not as technical as you think. JustCall’s interface handles the heavy lifting. You’re mostly writing prompts, configuring settings, and connecting automations. If you can build a Zapier workflow, you can build this.
2. The conversation script is everything. The AI is only as good as the instructions you give it. We iterated on the script many times before it felt natural. Test, listen to recordings, adjust, repeat.
3. Timezone handling is the hardest part. Not the AI, not the voice — correctly converting times across international leads and making sure you’re not calling someone at 3am. Build this logic carefully.
4. Always build fallbacks. What if there’s no calendar availability? What if the booking fails? What if the lead already booked? Handle every edge case. No lead should ever fall through.
5. Voicemails eat credits. If you’re paying per minute, disable voicemail handling or keep the message very short.
6. The client relationship matters. Greeking.me was an incredible client to work with. Their enthusiasm (“This is amazing — PUBLISH!”) made the iteration process smooth. Not every client will be this engaged.

Final Thoughts

AI voice agents aren’t science fiction anymore. With the right tools and some automation know-how, you can build one in weeks. The technology is accessible — the real skill is in the design: writing a natural script, handling edge cases, and making the experience seamless for the lead.

If your business has a speed-to-lead problem, an AI voice agent might be the highest-ROI automation you can build.

Built by Kubi Rich at AI Operator. Have a similar project? Get in touch.