Build a Voice AI System for Your Plumbing Business
Syntora builds custom voice AI solutions for home service businesses like plumbing companies. We build systems that answer calls, qualify jobs, and schedule appointments without human intervention.
This is not an off-the-shelf chatbot. A production-grade system requires direct integration with your scheduling software and needs to understand the specific language your customers use. The build scope depends on your call volume and the complexity of your dispatching rules.
We recently built a system for a 12-person plumbing company that was losing 20% of calls to voicemail. Their four dispatchers could not keep up with the morning rush. In four weeks, we deployed an agent that cut their missed call rate to under 2% and reduced the average booking time from 7 minutes to 90 seconds.
What Problem Does This Solve?
Most plumbing companies first try a basic IVR phone tree. This frustrates customers who have to navigate menus like 'Press 1 for scheduling, Press 2 for billing' when they have a semi-urgent issue that does not fit a category. The result is that most callers just press 0, defeating the purpose and still tying up a dispatcher.
A step up are generic call center platforms. These services charge per minute and use general-purpose language models that struggle with industry specifics. They can book a 'plumbing appointment' but cannot distinguish the urgency between a 'dripping faucet' and a 'sewer backup'. This lack of domain knowledge means they cannot prioritize high-value emergency calls, routing everything into one queue for a human to sort out.
Consider a customer who calls at 8 AM saying their 'water heater is making a rumbling noise'. A generic bot, unable to assess the risk of a potential tank failure, asks a series of irrelevant questions and then defaults to 'Let me connect you to an agent.' That agent is already on another call, so the customer with a potential $2,000 replacement job hangs up and calls your competitor.
How Does It Work?
We start by integrating directly with your CRM and dispatching software, such as Housecall Pro or ServiceTitan. We use their APIs to pull 6 months of call logs and job notes. This data is used to fine-tune a language model from the Claude API, teaching it the specific terminology and call patterns of your business. All API connections are made with Python using the httpx library for asynchronous requests.
We build the core logic as a FastAPI service that understands conversational nuances. It can identify a high-priority job by keywords and context, then immediately check your dispatch board for the nearest available technician with the right skills. This custom logic, written in pure Python, can handle complex scheduling rules that are impossible in visual flow builders, like 'never book a boiler repair after 4 PM'.
The agent confirms appointment details, collects customer information, and creates a new job in your system automatically. The entire interaction, from first ring to a confirmed booking in your calendar, takes about 90 seconds. We build the system to handle up to 20 concurrent calls without a drop in its 500ms response time.
The final system is deployed on AWS Lambda, which keeps hosting costs under $50/month for a typical business handling 3,000 calls. We use a Supabase database for logging every transcript and agent decision, with structured logging via structlog. This provides a full audit trail and allows us to quickly diagnose and fix any call that was handled incorrectly.
What Are the Key Benefits?
Handle 10 Concurrent Calls Instantly
The system answers every call on the first ring, 24/7. No more missed leads or full voicemail boxes, even during the 8 AM Monday morning rush.
Pay for Usage, Not for Agents
A one-time build cost replaces dispatcher salaries or high per-call SaaS fees. Hosting on AWS Lambda costs pennies per call, not dollars per minute.
You Own the System and the Data
You receive the full Python source code in your company's GitHub repository. There is no vendor lock-in and no ongoing licensing fees.
Alerts Before Customers Complain
Built-in monitoring tracks the agent's accuracy. If the system misclassifies more than 5% of jobs in a day, we get an automated alert to investigate.
Connects Directly to Housecall Pro
The agent writes job details, customer info, and call transcripts into your existing dispatch software. No manual data entry for your team.
What Does the Process Look Like?
Week 1: Scoping and API Access
You provide read-only access to your CRM and phone system logs. We analyze your call patterns and deliver a detailed technical specification document.
Week 2: Core Agent Build
We build the voice agent using the Claude API and your call data. You receive a link to a staging phone number to test the agent's understanding.
Week 3: Integration and Deployment
We connect the agent to your live scheduling software and deploy it on AWS Lambda. The deliverable is a functional system tied to a live number.
Weeks 4-6: Monitoring and Handoff
We monitor live calls for the first two weeks, tuning the agent's logic based on real-world performance. You receive a runbook and full source code access.
Frequently Asked Questions
- How much does a custom voice system cost?
- The cost depends on the complexity of your scheduling rules and which CRM we integrate with. A simple booking agent for a single-location business is a 3-week build. A multi-location system with technician-specific skills routing takes longer. We provide a fixed-price quote after a 30-minute discovery call at cal.com/syntora/discover to review your exact needs.
- What happens if the AI can't understand a caller?
- If the agent fails to understand a request after two attempts, it automatically transfers the call to a human dispatcher's cell phone. This 'human-in-the-loop' fallback is a standard part of our build. The full transcript of the failed interaction is logged and flagged for review, so we can improve the agent's logic for the next time.
- How is this different from a service like CallRail?
- CallRail provides call tracking and analytics; it tells you where your calls came from. It does not answer the phone or interact with the caller. Our system is an agent that actively qualifies, schedules, and dispatches jobs. We often use CallRail data as an input to our system, but we replace the need for a human to answer the tracked number.
- Can it handle different accents or languages?
- Yes. The underlying AI models are trained on a massive dataset of global accents. For an initial build, we focus on English, but we can add Spanish language support. This involves providing Spanish call logs or transcripts for fine-tuning. We test the system's accuracy against common regional accents in your service area during the final week of the build.
- How does it handle true emergencies like a gas leak?
- We build an explicit emergency detection module. Keywords like 'gas leak,' 'carbon monoxide,' or 'major flood' immediately trigger a high-priority alert and an instant transfer to a designated on-call technician, bypassing the standard scheduling flow. This logic is hard-coded and tested rigorously before deployment to ensure reliability for critical safety events.
- Who owns the call data and transcripts?
- You do. All data is processed and stored within your own AWS account and Supabase database. Syntora only requires temporary access during the build and optional maintenance period. We do not use your data to train models for other clients. You have full control and ownership of all customer information.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call