Zapier MCP connection management: prevent broken workflows from hidden connections
Stop broken Zapier MCP workflows before they happen. Learn connection governance: naming conventions, registries, and cleanup rules for multi-org setups.
How an AI Voice Agent Replaced 90+ Weekly Confirmation Calls
Learn how a market research firm automated 90+ weekly confirmation calls with an AI voice agent built on Twilio, 11 Labs, and Zapier. See the exact architecture.
TABLE OF CONTENTS
A market research firm running 15–30 paid focus group sessions per day was burning director time on manual phone confirmation calls — one call per participant, every day. After a Connex build using Twilio, 11 Labs, Zapier, and Airtable, an AI voice agent now handles those calls automatically, confirming attendance or routing rescheduling without any human picking up the phone. Estimated impact: ~5–6 hours per week of director time recovered, equivalent to ~$11,000–$13,000 per year at a blended $45/hr rate.
Estimated impact, based on reported session volume and a blended $45/hr labor rate.
The Situation
This is a ~10-person market research company that runs paid focus group studies for corporate clients. They book participants into scheduled sessions — two daily time slots, 9:30 AM and 2:00 PM — and manage everything in Airtable: participant records, confirmation status, session assignments, and rescheduling links.
At the time of the engagement, each session had 15 seats. The business was actively expanding to 30 seats per slot. That growth made the confirmation problem urgent.
The Problem
24 hours before every session, a director had to manually call each registered participant to confirm attendance. Two slots, 15 participants each: 30 calls per day. No shows cost real money — failed fill rates mean unhappy clients and partial payments.
The calls themselves weren't complex — "You're confirmed for tomorrow at 9:30 AM, press 1 to confirm or 2 to reschedule" — but they were repetitive, time-specific, and had to happen on a fixed daily schedule regardless of what else was going on. The director was the bottleneck.
With an expansion to 30 seats per slot on the horizon, the manual approach would become untenable. 60 calls per day isn't a workflow — it's a full-time phone job.
What They'd Already Tried
The team had a Zapier account and used it for other workflows. They'd connected Airtable to various tools before. But automating outbound phone calls requires more than standard Zapier connectors — it requires an AI voice layer (to speak naturally) plus a telephony API (to actually make the call), connected in a way that handles branching responses: what does the system do if someone presses "reschedule"? What if the call isn't answered? Standard no-code tools get 80% of the way there and stall on the edge cases.
What We Built
A Connex consultant built an AI voice confirmation system that triggers automatically from Airtable, 24 hours before each session. Here's the architecture:
- Airtable holds the participant list with confirmation status, session time, phone number, and a pre-populated Calendly reschedule link.
- Zapier triggers daily at the scheduled time, filtered to participants whose confirmation status is empty or "not confirmed."
- For each participant, Zapier calls the 11 Labs API (using XI-API-KEY header authentication — not the built-in Zapier app, which caused calls to drop) to generate an audio script with the participant's session details.
- Twilio is set up directly (not via the Zapier Twilio app) to make the outbound call, play the AI-generated audio, and capture the keypress response: 1 = confirmed, 2 = reschedule.
- If the participant presses 2, the system reads their Calendly reschedule link from Airtable and delivers it via SMS through Twilio.
- After each call, Airtable is updated automatically: confirmation status set to "confirmed" or "reschedule requested," no manual entry required.
A key discovery during the build: the standard Zapier Twilio integration caused calls to disconnect after delivering the message. The fix was bypassing the Zapier-native Twilio app and connecting to Twilio's API directly with custom authentication. That edge case — the kind that stalls a DIY build — was diagnosed and resolved in session.
Session 2 completed the implementation and tested end-to-end with a live call, including the reschedule-link flow. Session 3 refined the audio script (fixing a cutoff issue by updating the prompt to read the full script without interruption) and confirmed the reschedule-link field was correctly populated from Calendly into Airtable.
The Results
The AI voice agent now handles confirmation calls for every session without director involvement. Before the build: 30 manual calls per day (two daily slots × 15 participants). After: zero manual calls. The system scales directly to 60 calls per day as capacity expands to 30 seats — no additional staffing, no process change.
Estimated time recovered: ~30 calls/day × ~2 min each ≈ 1 hour/day ≈ 5–6 hours/week of director time. At a blended $45/hr labor rate, that's approximately $11,000–$13,000/year in recovered capacity — not counting the value of eliminating missed calls during busy periods.
Confirmation status now flows back into Airtable automatically. The team can see real-time fill rates without chasing anyone. And when someone presses 2 to reschedule, they immediately receive the link — the session doesn't just dead-end.
What the Client Said
The project was booked specifically as an outbound AI calling agent, with a phase 2 discussed (adding inbound handling for missed calls). The engagement moved from sales call to first build session within days, and the system was fully functional after three sessions.
Frequently Asked Questions
Does this work if our participant list is in a different tool, not Airtable?
Yes. The trigger layer (Zapier) can pull from most scheduling or CRM tools — the key data needed is a phone number, a session time, and a confirmation status field. Airtable was used here because it was already the team's participant database.
What happens if a call isn't answered?
Twilio handles unanswered calls — the status comes back and Airtable can be updated to reflect "no answer" so staff know which participants to follow up with manually if needed. A retry or SMS fallback can be added as a second step in the workflow.
How long did this take to build?
Three one-hour ZoomFlow sessions: the first established access and tested call flow, the second completed the implementation and ran a live test, and the third refined the voice script and confirmed reschedule-link delivery. Total: 3 hours of billable build time.
Can the AI voice be customized to match our brand's tone?
Yes. The 11 Labs voice and call script are fully configurable. You choose the voice, write the script, and the prompt can reference participant-specific fields from your database (name, session time, location) to personalize each call.
If your team is spending hours on repetitive outbound calls that follow a predictable script, that's a workflow ready for automation. The edge cases — authentication, keypress handling, reschedule routing — are where DIY builds stall. Book a ZoomFlow session and we'll build it with you live, start to finish.
Related Posts
How to build a break-fix AI ops agent to troubleshoot Zapier + Airtable failures (Zeke playbook)
Learn how to build a break-fix AI ops agent that monitors Zapier + Airtable failures, diagnoses root causes, and proposes safe fixes with human approval gates.
How to automate 24-hour appointment confirmation phone calls with AI
Build AI appointment confirmation calls that reach leads 24 hours before their visit, capture outcomes, and sync Airtable and Calendly with Zapier.
