Build vs. Buy: Adding AI Calls to ServiceTitan (The Real Math)

You run an HVAC company. You use ServiceTitan for dispatch and scheduling. You want your phone to answer after hours, handle appointment requests, and push jobs into ServiceTitan without waking anyone up at 11pm.

You've looked at the options and landed on the classic software question: build it or buy it?

Here's the actual math.

The Build Option

Building a voice AI integration for ServiceTitan means assembling several independent systems and making them work together reliably.

The AI engineer. Someone has to build this. A mid-level AI/ML engineer with voice and telephony experience costs $140–165k in salary plus benefits and overhead — call it $175k all-in for year one. You'll need at least 6 months of their time before anything is in production. That's $87k in labor before your first call.

Twilio. Voice infrastructure costs ~$0.017 per minute for calls plus ~$1/month per phone number. At 500 calls/month averaging 4 minutes each, that's $34/month in Twilio costs — practically free. The cost isn't the usage; it's the engineering time to build the Twilio integration correctly (webhook handling, call state management, recording, transcription).

The ServiceTitan connector. ServiceTitan has a developer API. Using it for booking requires understanding their data model: job types, booking sources, customer records, tech availability. Building a connector that correctly handles the edge cases — existing customer vs. new customer, service area validation, job type routing — takes 4–6 weeks for a developer who knows the API. Someone learning it from scratch: 8–12 weeks.

The voice AI itself. You need a system prompt that understands HVAC scheduling. You need intent classification that knows the difference between "my AC isn't cooling" (service call) and "I want a quote on a new unit" (sales lead). You need a feedback mechanism so the system gets better when it gets things wrong. Building this from scratch, without the vertical intelligence layer that WFW provides, means your engineer is also writing and maintaining an AI system.

TCPA compliance. Outbound calling in the US is governed by the Telephone Consumer Protection Act. Your AI calling customers back needs TCPA-compliant consent handling, opt-out management, and calling hour restrictions by time zone. Getting this wrong is expensive: $500–$1,500 per violation, class action exposure. Budget for a lawyer to review your implementation: $5–10k.

Year 1 total for build option:

• Engineering (6 months to ship): $87k
• Ongoing engineering (maintenance, improvements): $87k
• TCPA legal review: $7k
• Infrastructure (Twilio, hosting): $2k
• Total: ~$183k year 1, 6 months to first call

And that's if it goes well. Most custom AI voice integrations take longer and cost more than estimated.

The Buy Option

WFW's base plan is $299/month. The ServiceTitan integration is pre-built — it's a tool endpoint in the HVAC VIL (Vertical Intelligence Layer) that maps WFW's job booking tool calls to ServiceTitan's API.

Time to live: 2 days. Day 1: service account setup, first agent provisioned for your test office. Day 2: verify the ServiceTitan integration is booking correctly, configure your business hours and after-hours settings, go live.

What's included at $299/month:

• Voice infrastructure (Twilio + ElevenLabs, managed)
• Automatic provisioning (KB built from your website automatically)
• HVAC VIL (vertical intelligence pre-loaded for HVAC/plumbing terminology, job types, ServiceTitan)
• TCPA compliance (calling hour enforcement, consent tracking built in)
• Call analytics and transcript storage
• Feedback-driven optimization (30-day rolling pattern analysis, prompt suggestions)

At 500 calls/month averaging 4 minutes: call usage is metered separately at $0.04/minute → $80/month in usage. Total: $379/month.

Year 1 total for buy option:

• Monthly subscription + usage: $379 × 12 = ~$4,500
• Engineering time to integrate: 2 days = ~$1,600 (at $100/hour)
• Total: ~$6,100 year 1, live in 2 days

The difference is $177k and 6 months. That's not a close call for most HVAC operators.

What You Can't Buy Cheaply

The math above is clear. But it's worth being specific about what's actually hard to replicate.

The HVAC VIL. The vertical intelligence layer includes HVAC-specific terminology, job type mapping, common service question handling, and integration-specific tool definitions for ServiceTitan. Building this is months of domain research plus ongoing maintenance as job types and pricing change. WFW maintains it across all HVAC operators on the platform; you'd maintain it alone.

The provisioning loop. Automatically building a knowledge base from your website — and keeping it current as your services and pricing change — requires the crawl-diff-update pipeline described elsewhere in this series. This is not a weekend project.

The feedback-driven optimization. An AI that doesn't improve from its calls will make the same mistakes indefinitely. WFW's closed-loop optimization pipeline (call → transcript → extraction → pattern analysis → prompt suggestion → human review) is built infrastructure. Replicating it requires significant data engineering work.

These three things are what makes a voice AI actually good over time, not just functional at launch. You could build something functional in 6 months. You'd spend another 12 months building the infrastructure that makes it improve.

The Hybrid Path

Some HVAC software companies use WFW's API as infrastructure and build their own UI on top. Their customers interact with their branded dashboard, their pricing, their support team. WFW is invisible in the product.

This is the best-of-both model: the SaaS vendor builds what they're good at (their vertical software product, their customer relationships, their differentiated UX), and WFW handles the voice AI infrastructure underneath.

The economics work differently for this path — see White-Labeling a Voice AI Platform in 3 Days for the numbers.

When to Build

If voice AI is your core product differentiator — if the AI itself, not the scheduling workflow it enables, is what customers pay you for — build. Relying on a third party's infrastructure for your core value proposition is a strategic risk worth the cost of ownership.

If voice AI is a feature that enables your core product (an HVAC company whose core product is HVAC service, not software), buy. The engineering cost of building and maintaining voice AI infrastructure is not your competitive advantage. The time you'd spend building it is time you're not spending on what actually differentiates your business.

Most HVAC operators are in the second category.

---

That wraps the SaaS Integration series. Start with Adding a Voice Agent to Your SaaS in 3 API Calls if you're just getting started, or jump to the Developer Deep Dives series for the architecture underneath.