How to evaluate an AI receptionist for a ServiceTitan shop: 12 questions to ask

Every demo sounds the same. Judge on the questions instead.

If you are shopping for the best AI receptionist for HVAC, plumbing, electrical, or garage-door work, you have noticed that every demo sounds alike. Warm voice, quick answer, a booking on a sample calendar, a slide about ServiceTitan. The differences between one ServiceTitan AI answering service and the next show up after you sign, in the write-back, the escalations, and the invoice.

So here is how to choose an AI receptionist when the demos will not separate them: skip the script and ask these twelve questions. They cover integration depth, call handling, measurement, and commercial terms. A vendor selling a message taker usually gets vague around question three.

Integration depth

The long version of this section is what a real ServiceTitan integration looks like. The short version is three questions.

1. Does it read live ServiceTitan data during the call?

A good answer: the AI recognizes an existing customer by phone number and can see service history, memberships, and open jobs while the caller is still talking. A bad answer: the AI collects a name and number and "your team can look them up." That is not reading ServiceTitan. That is asking your CSR to.

2. Does it create the job in ServiceTitan itself?

Ask to watch a job land in a live ServiceTitan account during the demo, with the right customer record, the right job type, and a capacity slot you could actually serve. All three have to be correct, because a duplicate customer or a wrong job type is cleanup work in the morning. If the vendor books into its own calendar and syncs "later," or hands you a form submission, that is a message taker with a nicer voice.

3. What gets written back on calls that do NOT book?

Most calls do not end in a booking, and this is where thin integrations go quiet. A good answer: every call gets a summary, notes about what the caller described, and a disposition on the ServiceTitan record, whether the caller was out of area, had a billing question, or hung up. A bad answer is "we send you a transcript." A transcript in an inbox is not a disposition your reporting can count.

Call handling

An AI answering service for contractors earns its keep on the hard calls, not the easy ones.

4. What happens on an emergency when the schedule is full?

Your dispatchers already have rules for this: which emergencies get squeezed into a full day and which ones wait until morning. A good answer: the vendor asks for those rules and encodes them, so a no-heat call in January is handled the way your best dispatcher would handle it. A bad answer: "we take a message and flag it urgent." An urgent flag does not fix a furnace.

5. How does it hand off to a human, and with how much context?

Some calls need a person, and the handoff is where callers get frustrated. A good answer: the AI escalates to your team with the full context of the conversation, so nobody asks the caller to start over. A bad answer: a cold transfer or a voicemail. Ask the vendor to show a live escalation, not describe one.

6. Does it know your after-hours and member pricing rules?

Member pricing, after-hours rates, and how the two combine are exactly what a caller asks about at 9pm. A good answer: the AI carries your rules for how member pricing changes after hours and answers from them, not from a generic script. A bad answer: "the AI does not discuss pricing." That means the caller either waits until morning or calls someone else.

7. Can it handle SMS and web chat on the same rails, or voice only?

Plenty of customers text or use the site chat before they ever call. A good answer: voice, SMS, and web chat follow the same rules and land on the same ServiceTitan record. A bad answer is a voice product with a partner's chat widget bolted on. That is two vendors and two sets of rules for one front desk.

Measurement

Every AI receptionist comparison for home services eventually comes down to one number, and vendors define that number differently. Pin it down.

8. How is booking rate defined?

There are two ways to count it. Raw booking rate is bookings divided by all calls, including wrong numbers, vendors, and billing questions. Adjusted booking rate is bookings divided by calls where the caller wanted to schedule and the AI ran the whole conversation. A good vendor tells you which one they quote, shows you both, and lets you audit the calls behind the denominator. A bad vendor quotes a single percentage with no definition, and two shops can hear the same figure while measuring different things.

9. Can it A/B compare call playbooks and roll back a change that hurts bookings?

The script will change over time: a new offer, a new job type, a wording tweak. A good answer: playbooks are versioned, a change is compared against the previous version on booking rate, and a change that hurts bookings rolls back automatically. A bad answer: "we update the prompt when you ask." That means nobody knows whether last month's edit helped or cost you jobs.

10. Do callers notice?

Ask for two numbers from production: the share of callers who detected they were talking to AI, and the share of calls that finished with zero issues. A good vendor has both and can show you the calls behind them. A bad vendor plays you their best recording. One great call proves nothing about the other eight thousand.

Commercial terms

Two questions on money. For the longer treatment, see what AI receptionists really cost.

11. What is the pricing model, and what is left off the pricing page?

Ask for the invoice, not the sticker. Specifically: is there an onboarding fee, a monthly platform fee on top of the quoted rate, a per-minute overage behind a minute cap, or a surcharge for the ServiceTitan integration? A good answer is a flat number per brand you can budget in January and in July. A bad answer is "it depends on volume," which means the price climbs in your busiest month.

12. Does ServiceTitan stay the system of record?

Some vendors want to become the place your customer data lives. That is lock-in dressed up as a feature. A good answer: every booking, note, and outcome lives in ServiceTitan, so if you ever switch vendors nothing has to be migrated. A bad answer: a separate customer database, a separate calendar, or "an export is available on request."

How FrontDeskPro answers the twelve

Here is our answer sheet, to hold against anyone else's.

• Answers 24/7 by voice, SMS, and web chat.
• Recognizes existing customers by phone number and books straight into ServiceTitan against live capacity, with the right customer, job type, and slot.
• Encodes your dispatch judgment: which emergencies to overbook into a full schedule, what saves a cancellation, and how member pricing changes after hours.
• Runs versioned call playbooks with A/B comparison on booking rate and automatic rollback when a change hurts bookings.
• Writes every outcome back to the ServiceTitan job and escalates to your team with full context.
• Production record: 8,712 calls handled and 1,201 jobs booked at a 67.2 percent adjusted booking rate over nine months across a PE-backed ServiceTitan home-services portfolio, including 478 after-hours bookings. 99.4 percent of callers never knew it was AI, and the zero-issue rate was 94.9 percent.
• Pricing starts at $1,000 per month per brand, with no onboarding fee, no platform fee, and no per-minute fee in the base model.
• ServiceTitan stays the system of record.

The detail is on the AI answering service page, and the production numbers are on the case study. To put the twelve questions to a live ServiceTitan account instead of a slide deck, book a walkthrough and bring your hardest call.