How to pick an AI front desk for a trades business

Search "best AI answering service for plumbers" and the top results are rankings, mostly written by the companies being ranked. We sell one of these things, so you should discount our opinion too. That's why this post is a checklist instead of a ranking: seven questions you can put to any vendor, with a scorecard at the end showing where ServiceFront passes and where it fails. Two of the rows are fails. They're in there.

The one test that disqualifies most of them

Before you read a single pricing page, run this test on the vendor's demo: ask the agent for a price you never gave it.

"What do you charge to replace a sump pump?" If it produces a number, walk away. The number might even land close. The problem is what the agent just showed you: it will say something plausible rather than admit it doesn't know. Next week it'll quote a customer $180 for a job you charge $450 for, and you'll spend a phone call arguing with your own front desk.

A made-up answer is a promise your business has to keep or clean up. Air Canada found out the hard way: its website chatbot told a passenger they could book a full-fare ticket and claim the bereavement discount back afterward, which the airline's actual policy didn't allow. In February 2024 a British Columbia tribunal ordered Air Canada to pay CA$650.88 in damages, holding the airline responsible for what its bot said (Moffatt v. Air Canada, 2024 BCCRT 149). Air Canada had argued it couldn't be held liable for information provided by its own chatbot, a position the tribunal read as suggesting the bot was "a separate legal entity that is responsible for its own actions." The tribunal called that a remarkable submission and rejected it.

That case is the rule now, for a national airline and for a two-truck plumbing shop alike: what your bot says, you said.

The checklist

1. It refuses to invent facts. The test above, plus a follow-up for the vendor: where do the agent's answers come from? "A profile of facts you verified" is the right answer. "We crawl your site" is weaker (your site is probably out of date). "A prompt we write during onboarding" is weakest. The source of truth decides whether the bot is a front desk or a liability with a chat window.

2. It's honest about being AI. Ask the demo agent directly: "am I talking to a person?" The only acceptable answer is an honest one. A bot coached to dodge that question will get caught eventually, and the customer who catches it stops believing everything else it said, including your real prices and your real hours.

3. There's a human handoff, and nothing dead-ends. What happens when the agent can't answer? The acceptable outcomes are: it says it doesn't know, takes a message, and gets the question to you. The unacceptable outcome is a guess. Ask the vendor to show you a conversation where the bot didn't know something.

4. The lead actually reaches you. An agent that answers beautifully and then files the lead in a dashboard you'll check on Sunday is a slower contact form. Ask: where does the lead land (text, email, push), how fast, and what happens when delivery fails. Then ask what an emergency lead does differently from a "can you quote a repipe" lead.

5. It knows the trade, not just your business. Ask the demo: "there's a sewage smell in my basement, can it wait until Monday?" A generic chatbot will chat. A trades-native agent will treat it as urgent and push toward a call. Your customers' first questions are triage questions, and a bot that can't triage is answering the easy half of the job.

6. The pricing is flat enough to predict. Per-minute and per-call meters mean your busiest month is your most expensive month, which is backwards. Ask what a slammed August costs, in a number. If the answer has the word "depends" in it, ask again.

7. You control what it's allowed to say. You should be able to set a quote range per service, switch quotes off entirely for jobs that need eyes on site, and approve the facts before they go live. An agent with discount authority nobody gave it is how a Chevy dealership's bot once agreed in writing to sell a 2024 Tahoe for a dollar.

Where ServiceFront stands, honestly

Run the same checklist against us and here's the scorecard:

Criterion	ServiceFront today
Answers your phone	No. It works on the web: your site, your hosted page, the links you share.
Books into your calendar	Not yet. It captures the lead and sets up the callback; calendar booking is on the roadmap.
Texts you the lead	On the roadmap. Today the lead lands in your email.
Grounded in facts you verified	Yes. If you didn't confirm a fact, the agent says so and takes a message instead of guessing.
Honest about being AI when asked	Yes.
Human handoff, nothing dead-ends	Yes. Every conversation ends in an answer, a captured lead, or your number.
Trade triage	Yes, for the trades we've built deep, plumbing first.
Flat pricing	Yes. Free to run read-only, $59 a month flat for lead capture. No meters.

If your biggest leak is the phone ringing while you're under a sink, the first row means we're the wrong fix today, and this comparison of who picks up what will serve you better than anything else we could tell you. If your leak is the customer who reads your site at 11pm and never calls, the rest of the table is the part that matters.

Run the test on us first

The grounding test works on us too, and it's free to run. Paste your site's URL at servicefront.ai and ask the agent for a price you never published. The right answer is that it won't give you one.