How AI-powered lead routing transformed our emergency plumbing client in 2.5 days
We shipped a full-stack lead-gen platform with an AI voice agent, real-time dispatch routing, and call-to-job tracking for an emergency plumbing service. The build took 2.5 days. The first qualified job came in on day three.
A plumbing owner called us with a problem most trades operators recognize. Inbound emergency calls were landing in voicemail, web forms were piling up in an inbox nobody read on weekends, and the nearest available technician was getting picked the same way it had been picked in 1995: someone grabbing a clipboard.
We had 2.5 days. Here's what we shipped and why it worked.
What we built
- A conversion-first marketing site at emergencyplumberma.com, built mobile-first because emergencies happen on phones.
- A 24/7 AI voice agent that answers in under three rings, captures the address, the emergency type, and the urgency, and asks the qualifying questions a dispatcher would.
- A real-time dispatch router that pings the nearest available licensed plumber, with fallback rules if the first technician doesn't accept inside 60 seconds.
- End-to-end lead tracking from the first click or call to the completed job, including source attribution and time-to-dispatch metrics.
Why we could do it in 2.5 days
Not because we cut corners. Because we did the architecture work upfront and shipped vertical slices instead of building horizontally.
- Day 0.5 was scoping. We talked to the owner, mapped the call flow on a whiteboard, and agreed what counted as a qualified job. No design phase, no Figma, no procurement theater.
- Day 1 was the site and the voice agent in parallel. Two engineers, one stack. The voice agent uses a single LLM with a tight system prompt and a tool-calling schema for capturing structured fields.
- Day 2 was dispatch routing and tracking. The router is a lightweight queue, not a Kafka cluster. The owner can read the code if he wants to.
- Day 2.5 was integration and the first live test on a real call.
What we learned
Most trades software is over-engineered for the operator and under-engineered for the customer. The plumber doesn't need a Salesforce instance. He needs a phone that rings, an address he can drive to, and a job he can close.
We left HousecallPro alone for the parts it does well (invoicing, scheduling, payments) and built the parts it doesn't (custom voice agent, real-time dispatch, attribution) on top. See the full case study for the architecture and the post-launch numbers.
If you're a trades operator reading this
The takeaway isn't 'use AI for your phones.' It's that the 2025 stack for a service business has at least three pieces (site, lead capture, dispatch) and you don't have to glue them together by hand anymore. If you'd like to map yours, book a free growth audit and we'll spend 30 minutes on it.
Frequently asked questions
- How fast can you build a custom AI voice agent for an SMB?
- For a tightly scoped use case (inbound emergency triage, intake qualification, after-hours booking), 2.5 to 5 days from kickoff to live. Larger scopes (multi-language, deep CRM integration, full call analytics) take 2 to 3 weeks.
- Does this replace HousecallPro or ServiceTitan?
- No. We integrate with them. The voice agent and dispatch router sit in front of the scheduling and invoicing tools you already use, so technicians keep working the way they already work.
- How does the pricing model work?
- The build is a one-time engagement scoped to what your business actually needs. Runtime cost is the LLM tokens (cents per qualified call) plus the SMS or phone provider you choose. We give you a fixed quote on a strategy call once we understand the scope; happy to walk through it on a free growth audit.
