A plumber runs 36 AI-built Zaps around one system of record. The Zaps that catch mistakes matter most

A trades owner who calls himself "not a computer person" now runs 36 Next-Gen Zaps, and the design rules behind them are worth copying.

What happened

Zapier published a customer story on 3R Plumbing & Mechanical, a residential HVAC and plumbing company in Massachusetts. Ross Hunt started it in 2022 from his garage with one pickup truck. Early revenue was about $300,000 a year; this year it is on track for nearly $3.8 million, with 15 employees.

Hunt joined the Next-Gen Zaps early access program in August 2026. He describes what he needs to Claude, which is connected to Zapier through Zapier MCP, and Claude writes and publishes the workflow. Hunt reviews the logic, tests each one and checks that the data lands where it should. Today 36 Next-Gen Zaps are live, running about 1,400 times a week on top of his classic Zaps.

His core rule: ServiceTitan is the database, and everything else reads from it or writes back to it, with Zapier in between. The Zaps he values most catch mistakes. A booking left pending for 20 minutes triggers an email. A daily check flags Facebook ad leads still pending after 24 hours. Another compares booked jobs against the tracking number the customer actually called to catch wrong campaign tags. A data watchdog checks every feed each morning and emails him if anything is stale. A cleanup Zap closed thousands of leftover estimates, with a guard so it could never touch a sold one.

Zapier announced Next-Gen Zaps at ZapConnect 2026, adding branches that rejoin, nested loops, error handling that keeps a run going, and pauses for a human reply.

My take

The headline is the AI building the Zaps. The lesson is everything Hunt did around it.

• One system of record. Every workflow points back to ServiceTitan, so there is one truth to check.
• Alerts for silence, not just for errors. A lead that sits untouched for 20 minutes never throws an error, yet it is the most expensive failure in home services.
• Guards on destructive jobs. A cleanup that cannot touch sold estimates is a cleanup you can let run.
• A watchdog on the data feeds, because a dashboard built on a stale feed looks fine and is wrong.
• A human still reviews and tests every build.

If you run a service business, start with the missed lead alert and the stale data check. They are small, cheap and they protect revenue you already paid to get.