1,361 Calls, 72% Booked With Zero Human Help: A Texas HVAC Shop's AI Voice Agent Numbers

Source and scope

This is an industry profile, not a customer testimonial. Bill Joplin's Air Conditioning and Heating is not an Otherworlds AI client. The numbers and quotations below come from a single public source that names them directly:

• ServiceTitan blog, April 30 2026 — "Karen, Lucy, Piper and Jenny Are ServiceTitan AI Virtual Agents. And They're Basically Employees Now" by Brendan Meyer.

ServiceTitan is the CRM/platform vendor. Their published numbers are vendor-reported, not independently audited. We flag this because it matters when you read the conclusions below.

Who Bill Joplin's is

Bill Joplin's Air Conditioning and Heating is a residential HVAC service and installation company based in McKinney, Texas — a fast-growing suburban community north of Dallas in Collin County. The company handles AC repair, heating service, maintenance plans, and full system replacements. The person ServiceTitan quotes is Randi Thompson, who manages office operations and made the decision to deploy an AI voice agent.

The problem: after-hours, overflow, and a small front office

The scenario matches what most HVAC operators between 5 trucks and 50 trucks experience:

• Phone volume is spiky. Summer heat waves and winter cold snaps bring 5–10x the baseline call load.
• Office team is small. They answer during business hours, handle dispatch, process payments, and chase callbacks. When calls pile up, jobs get lost.
• After-hours is worst. The calls that come at 8 PM on a Thursday — AC died, baby won't sleep — those are the highest-urgency, highest-close-rate calls. They hit voicemail.
• Traditional answers: live answering service ($2,000–$4,000/month, takes messages but can't book), overflow CSR (hard to hire, hard to retain), or "text us" (loses older customers who still call).

Thompson needed a way to capture the calls already arriving that nobody was handling.

What they deployed

In January 2026, Bill Joplin's went live with ServiceTitan's AI Virtual Agent — a voice agent integrated into ServiceTitan's telephony and CRM layer. They named her "Karen."

The naming decision itself reveals how the team treated the deployment. They originally planned "Keith" as an inside joke about multiple employees sharing that middle name. When ServiceTitan's voice defaulted to female, they needed a K-name. Someone suggested Karen. Thompson's published line: "You just blame it on Karen."

Karen handles:

• Inbound service and repair calls (AC not cooling, heat not working, unusual noises, thermostat issues).
• Maintenance plan bookings and reschedules.
• Emergency triage — routing true emergencies (burning smell, gas suspicion, no-heat in winter with infants) to the on-call tech.
• CRM lookups — identifying returning customers, pulling their system history.
• Direct booking onto the ServiceTitan dispatch board, confirmed via SMS.

The published numbers

ServiceTitan reported the following for January through late April 2026:

Metric	Value
Total calls handled by Karen	1,361
Calls ending in a booked job	Over 90%
Booked with zero human intervention	72%

Thompson's one-line summary that ServiceTitan published:

"She's the lowest paid employee I've got, and she never complains about it."

That's a joke. But the underlying point: the cost of the agent is a fraction of the fully-loaded cost of a second CSR, and unlike a human dispatcher, the agent doesn't need health coverage, vacation accrual, or a shift schedule during a heat wave.

What those numbers actually tell you

The 90% booking rate is strong in context. Industry average for human phone answering in home services — even your best CSRs — runs 70–85% booking conversion on qualified inbound calls. A 90%+ rate means the agent isn't just answering, it's closing conversations effectively.

The 72% zero-intervention number is the one that matters most. Karen is not transferring the majority of calls to a human for the actual booking. She is completing the conversation, identifying the job, checking the calendar, reserving the slot, confirming the customer, and logging the ticket — without Thompson's team touching the call. That's the workload-relief metric.

Context caveats to apply before drawing conclusions:

• ServiceTitan published this, not an independent auditor. Treat it as directional — "high enough for the vendor to feature as a flagship story" — rather than peer-reviewed.
• 1,361 calls over roughly 3.5 months is about 12–13 calls/day. That's plausible for after-hours overflow at a single-location residential HVAC company in North Texas during a warming spring.
• The 90% "booked" figure includes the 72% zero-intervention + a portion of transferred calls that humans also converted. So the model works: AI handles the clear cases, humans rescue the tricky ones. Humans aren't idle — they're just handling 28% instead of 100%.

Why this deployment proves the category has matured

Bill Joplin's is one of four ServiceTitan AI Virtual Agent deployments that ServiceTitan featured in a single April 2026 article. The four span Texas, New York, Georgia, and California — independently owned shops ranging from plumbing-only to HVAC-and-plumbing. That's the market signal: this is no longer a beta. Real operators are running production workloads.

And the period ServiceTitan reports (January through April) captures the spring booking surge and a late cold snap in North Texas. The agent held through real seasonal load, not a quiet pilot period.

What an HVAC operator should take from this

1. If the 72% figure is even directionally accurate, the ROI math is overwhelming. Compare it against the cost of an answering service that books zero jobs, or a second CSR at $3,500+/month fully loaded, or the revenue lost from the 40% of after-hours calls that currently hit voicemail.

2. "Zero intervention" is the metric that matters, not "answers the phone." Plenty of services answer. Very few complete the booking on your calendar without a human re-keying the details. The CRM integration layer is what converts "answering" into "revenue captured."

3. Start in overflow mode. Bill Joplin's did not replace their front-office team. They gave the agent the calls nobody was handling — after-hours, overflow, peak-season spillover. Keep your humans for complex diagnostics and relationship-heavy work. Stop losing simple "yes I can come Tuesday at 2" bookings to voicemail.

4. Track the transfer rate, not just the booking rate. If an agent books 90% but transfers 30% to humans for "clarification," real workload savings are less than the headline implies. If transfer is 28% (as the 72% zero-intervention suggests), you're getting genuine relief. Ask your vendor to report this.

5. The "lowest paid employee" framing is the right mental model. Karen doesn't replace your team. She fills a gap they physically cannot — 11 PM, 2 AM, Saturday morning, Thanksgiving. The humans come in Monday and find the bookings already on the board.

Related reading

• HVAC After-Hours Calls: The Revenue Math You Are Leaving on Voicemail — quantify what your own missed calls cost before evaluating options.
• How a Small HVAC Shop in Colton, CA Put an AI Agent on the Overnight Shift — sister profile on another ServiceTitan AI Virtual Agent, qualitative rather than quantitative.
• Agent+ HVAC Voice + Chat Agents — our proprietary-model alternative for operators who want the same outcome with different model-stack and integration choices.
• AI answering service — platform-level service page.
• AI for HVAC hub — all our HVAC analysis content.

Numbers in this article are vendor-reported by ServiceTitan, not independently audited. We have linked the primary source for verification. If you run a similar deployment with independently trackable numbers, we would like to profile yours.