Summary
A coaching firm asked me to take their weekly check-in grind off their hands. They needed a system that would call each client, gather progress answers, and surface insights to coaches—without anyone lifting a finger. I delivered an AI-driven accountability bot that handles the calls, turns transcripts into clear takeaways, and emails coaches a polished recap. Behind the scenes, a Retool dashboard tracks call stats and spins up on-demand reports.
The Client’s Brief
My client’s goal was straightforward yet time-consuming: ensure every client received a friendly check-in, capture their responses, and deliver those insights back to coaches efficiently. They wanted:
- Automated, scheduled calls with each client.
- Actionable summaries coaches could actually use.
- A single dashboard and weekly email that removed manual follow-up.
Tech Stack & Approach
To keep costs low and iteration fast, I leaned heavily on low-code tools:
- Vapi + Twilio for AI voice calls and reliable phone lines.
- n8n as the orchestration heart—queuing calls, normalizing data, triggering AI prompts, and firing off emails.
- OpenAI to summarize client transcripts into coach-ready insights.
- Supabase for call logs, open-rate metrics, and sentiment flags.
- Retool to build a responsive owner dashboard without a full frontend rewrite.
Building the Solution
The first prototype stumbled when I queued too many calls—n8n nodes timed out, and Vapi rejected retries. Slipping in an n8n queue node throttled calls just enough to keep things smooth.
Then came the Twilio trial hurdle: only verified numbers could be dialed. A quick upgrade to a paid Twilio line unlocked full dialing capacity.
Finally, each service spoke its own JSON dialect. A small JavaScript transformer inside n8n now normalizes Vapi’s output before handing it to OpenAI, ensuring consistent transcripts and summary quality.
Impact for the Client
- 92 % call completion across the first 30 clients, up from 40 % manual outreach.
- 80 % reduction in coach admin time—dropping from roughly 5 hours to under 1 hour per week.
- 58 % average email open rate, a solid jump from 34 % on their old manual recaps.
What I Learned
Orchestrating multiple low-code tools can be trickier than writing custom code—but it lets you pivot faster when requirements change. Handling telecom quirks (like verified-number restrictions) was less about AI magic and more about reliable plumbing. And storing clean data in Supabase early made dashboarding a breeze.
Next Steps
The client’s already talking about swapping Vapi for OpenAI’s Text-to-Speech once latency improves, adding real-time sentiment alerts, and even letting coaches reply to recaps—pushing their notes straight back into the database.
If you could hand off any part of your client workflow to an AI right now, what would it be? Let me know in the comments—I’m always scouting the next bottleneck to crush.
