Azaz Vepari07 Mins read
ServiceTitan's AI Booked 90% of Calls — What That Number Doesn't Tell You
ServiceTitan says one HVAC contractor's AI voice agent booked over 90% of 1,300 calls. Here is what the announcement actually reports, what it leaves out, and the test to run before you believe it about your own call mix.
In this article9 sections
One HVAC contractor’s AI voice agent handled more than 1,300 calls in just over three months. More than 90% of them ended in a booked job. 72% finished without a human touching the call.
Those numbers are real, they are large, and they are worth taking seriously. They are also one customer story, published by the vendor, using figures supplied by the vendor and the contractor. That is not an accusation — it is the category the evidence belongs to, and it changes what you can safely conclude from it.
So this is not a debunk. It is a reading of what the announcement says, what it does not say, and the test that turns someone else’s 90% into a number about your own phone.
What ServiceTitan actually published
| Figure | Published value |
|---|---|
| Calls handled | More than 1,300, in just over three months |
| Booked | More than 90% of those calls |
| Completed without human intervention | 72% |
| Incremental booked revenue | Estimated $74,000 |
| Reduced operating costs | Approximately $23,000 |
| Average call length | Around 3 minutes |
| Implementation time | Roughly 20 minutes |
The contractor is Bill Joplin’s Air Conditioning and Heating, in the Dallas–Fort Worth area. The agent was deployed in January, works across voice and SMS, and the team named it Karen. ServiceTitan’s Vincent Payen frames the product problem as “every missed call is money left on the table”. Bill Joplin’s service operations manager, Randi Thompson, describes the effect in the most concrete terms in the release: “My phone wasn’t ringing because our voice agent was on the phone with those customers.”
Announcement dated 9 June 2026. Every figure in the table above is from that page, and nothing has been rounded or restated.
A correction against our companion video. The video says the operating-cost saving was about $33,000. The published figure is $23,000. The article is the record, and $23,000 is the number to quote.
Why a native agent is different from an answering service
The interesting claim in this announcement is not the voice quality. It is where the agent sits.
An AI agent built into the system of record can read customer history, technician schedules, business capacity, job types and everything else already in the platform. A generic voice bot cannot.
Picture the difference during a heatwave. A homeowner calls and says the upstairs system is not cooling. A generic bot takes a name and promises a callback — it has produced a message, not a booking. An agent with access to the operating system can identify the customer, classify the job type, look at real availability and place the appointment on the correct technician’s calendar.
That connection is the actual story here. An AI CSR that sounds human but cannot see capacity, memberships, zones, job types or customer history is not saving the office work — it is generating cleanup. The bet ServiceTitan is making is that native context lets the agent complete the transaction rather than merely record the conversation.
It is a reasonable bet. It is also the thing to interrogate when a competitor shows you a similar demo: not how natural the voice sounds, but what the agent can see at the moment it commits to a booking.
What the 90% number does not tell you
Here is the part that deserves care.
The announcement also mentions that a maintenance campaign caused call volume to spike during the period. That single detail matters, because a maintenance campaign generates a very particular kind of call: existing customers, known addresses, a known job type, low urgency, low price sensitivity. Those are the easiest calls in the business to book.
What we do not get is the breakdown:
- new leads versus existing customers;
- maintenance appointments versus urgent repairs;
- calls transferred to a human mid-conversation;
- cancellations and no-shows after the booking;
- how many bookings became collected revenue.
None of that makes the result fake. It means you cannot lift one 90% figure into your own forecast.
If your call mix is weighted toward after-hours emergencies, warranty disputes, price shoppers, angry customers, commercial approvals or callers outside your service area, your number will look different — possibly very different. The right comparison is never the vendor’s percentage. It is your current answer rate, booking rate, cost per booked call, cancellation rate, and revenue actually collected after the booking.
Booked is not collected
Worth isolating, because it is the single easiest number to misread. $74,000 in incremental booked revenue is not $74,000 in the bank. Booked revenue becomes collected revenue only after the job is run, invoiced and paid — surviving cancellations, no-shows, scope changes and the occasional customer who books three companies and takes whichever arrives first.
The same applies to the $23,000 in reduced operating costs. Avoiding seasonal hiring is a genuine saving, and it is also not the same as removing a CSR line from the payroll.
These are described in the announcement as estimated business results. They are not audited figures, and the release does not claim they are.
Does this mean human CSRs are finished?
No, and the same release contains the reason.
If 72% of calls completed without human intervention, then roughly 28% still needed a person somewhere in the process. That is not a rounding error, and those exception calls are rarely the easy ones.
The upset customer. The gas smell. The commercial account with a purchase order. The membership dispute. The homeowner who wants a technician today when there is no capacity left to give them. Those are the calls where judgement, liability and the customer relationship all land at once.
What changes is the shape of the office job, not its existence. Instead of spending the day repeating availability and collecting addresses, the team handles exceptions, quality control, escalations, outbound recovery and the relationships that keep customers. Plenty of businesses will genuinely avoid adding seasonal CSR headcount on the back of this. The shops that remove every human safety net because a vendor published a 90% booking story are learning the wrong lesson from it.
The pricing the announcement does not include
The release does not publish what this contractor paid.
It does not state the AI agent’s cost, which ServiceTitan package is required, what usage limits apply, or whether voice and SMS volume create additional charges. That is normal for a customer-story press release, and it is also the gap that makes ROI arithmetic impossible from the outside.
Before you calculate a return, get the full number in writing: subscription, implementation, phone, per-usage and support. Then measure it against completed and collected jobs — not answered calls.
For where this sits in the wider picture, our guide to AI workflows for HVAC businesses covers the seven workflows that are genuinely ready now and the controls each one needs, and our HVAC CRM comparison covers what ServiceTitan costs as a platform, where it publishes no price at all.
The 30-day test that settles it for your business
Run the agent against your actual call mix and measure these seven things. Do not use the vendor dashboard as the scoreboard.
- Answer rate — what share of inbound calls got picked up at all.
- Booking rate — of answered calls, how many became a job on the calendar.
- Transfer rate — how often the agent handed off to a person, and at what point.
- Bad bookings — wrong job type, wrong slot, outside your service area, duplicate.
- Cancellation and no-show rate on AI-booked jobs, compared against human-booked ones.
- Collected revenue from those jobs, not booked revenue.
- Hours your team spent fixing its mistakes — the number nobody tracks and the one that decides it.
If those numbers improve once the full cost is included, the agent is working. If the only thing that looks good is the vendor dashboard, it is not.
Run it during a normal month, not during a maintenance campaign. A campaign month flatters any booking system, which is precisely why it is worth noticing that this one happened during a spike.
Frequently asked questions
Did ServiceTitan’s AI really book 90% of calls?
That is what ServiceTitan published on 9 June 2026, for one HVAC contractor over just over three months and more than 1,300 calls. It is a vendor-published customer story using figures from the vendor and the contractor, not an independent benchmark across many businesses.
Is an AI voice agent worth it for a small HVAC company?
It depends on your call mix and your missed-call rate, neither of which this announcement can tell you. A shop losing bookings to unanswered phones during peak season has a clear problem to solve. A shop whose calls are mostly complex, commercial or emergency work has a much harder case, because those are the calls the agent hands back.
What does ServiceTitan’s AI agent cost?
The announcement does not say, and does not state which package is required or whether usage creates additional charges. ServiceTitan does not publish platform pricing either — it is quote-only, priced per technician. Ask for subscription, implementation, phone, usage and support as one written number.
Will an AI agent replace my CSRs?
Not on this evidence. 72% completed without a human means around 28% did not, and exception calls are usually the hardest ones. The realistic outcome is the office spending less time on repetitive booking and more on escalations, quality control and customer relationships.
What is the difference between an AI voice agent and an answering service?
Access to your operating system. An answering service takes a message; a native agent can identify the customer, read real capacity and place a booking on the right calendar. If the tool cannot see your schedule, memberships and job types, it is producing work for the office rather than removing it.
How should I measure an AI CSR pilot?
Answer rate, booking rate, transfer rate, bad bookings, cancellations and no-shows, collected revenue, and the hours spent correcting its errors. Compare against your own current numbers, and run it in a normal month rather than during a campaign.
The verdict
ServiceTitan has shown that a native AI voice agent can do real booking work, at meaningful volume, for at least one established HVAC contractor. That is genuinely important, and a year ago it would have been a demo rather than a case study.
But the winner here is not AI replacing every CSR. The winner is the contractor who gives AI the repetitive calls and keeps humans responsible for exceptions, safety and customer trust.
Treat the 90% as evidence that the category works, not as a forecast for your phone. Then go and produce your own number.
Sources
Sponsored
Ready to upgrade your dispatch?
FieldCamp uses AI to optimize routes, auto-assign jobs, and eliminate scheduling chaos — so you can run more jobs with fewer headaches.
Tested and written by
Azaz Vepari
Editor, Modern Field Service
Listen to this review
The same comparison as a podcast episode, for the drive between jobs.
Tool profiles
Pricing, best fit and limitations for each tool in this review, one page each.
