Answering & missed calls
How to test an answering service vs. an AI receptionist on your own calls
By ApexScale AI · Published
The short answer
Don't take anyone's word for reliability, ours included. Place the same 6 to 10 scripted test calls to each option from a number they don't know, at the same times of day, and score how fast each answered, whether it asked your questions, whether emergencies reached you, how complete and fast the lead arrived, and how the calls were billed. The scorecard below works for any live service and any AI receptionist.
Every answering service says it is reliable, and so does every AI receptionist company, including us. We have no independent data showing that Margot, our AI receptionist, is more reliable than a live answering service, and we have not found a published study that compares the two on home-service calls. So this guide doesn't ask you to believe anyone. It shows you how to find out on your own calls.
The method is a mystery-call test: a short set of scripted calls you place to each option, timed and written down, with a scorecard you fill in yourself. It takes a few evenings and a second phone. If you'd rather have someone else place the calls, the last section covers our free audit.
What does "reliable" mean for a contractor's phone?
For a contractor, reliable means five things you can check on a test call: the call is answered within the rings you'd accept, the caller is asked your questions, an emergency reaches you, the lead arrives complete and fast, and the bill stays predictable in a busy month. If you can't time it or check it, leave it out of the test.
- Answered fast. Count rings, or seconds, from the first ring the caller hears to the first word from a person or the AI. Decide your limit before you start, such as the most rings you'd wait through as a customer.
- The right questions. Write down the questions you'd ask on each call type: the address, what's wrong, whether water is still running, whether anyone in the home is vulnerable. Tick each one that was asked.
- Emergencies reach you. On an urgent call, time from the caller's first word to your phone ringing, and note what happened on the call you deliberately let ring out.
- The lead arrives complete and fast. Time from hang-up to the text or email landing on your phone. Check it has the caller's name, callback number, address or town, the job, how urgent it is and what the caller was promised.
- The bill stays predictable. Note how each test call was charged, and work out what the same bill looks like in your busiest month.
Which test calls should I place?
Place eight scripted calls that cover the hard cases: an after-hours emergency, a price shopper, an out-of-area caller, an existing customer, a caller who asks for a person, a hazard call, two calls close together, and a weekend estimate request. Use fictional details, and write down what a good answer sounds like before you dial.
Swap in your own trade's calls: for an electrician the hazard call is sparking or a burning smell, and for a roofer the emergency is an active leak in a storm. Our after-hours emergency calls guide helps you write the emergency rule both options will be tested against, and the AI receptionist evaluation guide adds edge cases such as an incomplete address or a caller who corrects a detail.
- After-hours emergency (after closing). Say: "My water heater is leaking all over the basement floor and I can't find the shutoff." A good answer asks your urgent-call questions, reaches you live or rings your on-call number under your rule, and tells the caller what happens next. Time the rings to answer and the seconds until your phone rings.
- Price shopper. Say: "How much do you charge to replace a toilet?" A good answer quotes only a price you approved, such as a trip charge, or none at all, takes the details and promises a callback time you can keep. Time hang-up to lead.
- Out-of-area caller. Give a town just outside your service area. A good answer catches it, tells the caller politely, and doesn't promise a visit you won't make. Note how far into the call the address was asked for.
- Existing customer. Say: "You were out last week and the drain is backing up again." A good answer marks the caller as an existing customer and follows your rule for them, such as a transfer when a visit is scheduled today. Time hang-up to lead.
- Caller who asks for a person. Say: "Can I just talk to a real person?" and, on the AI, "Am I talking to a robot?" A good answer is honest, then transfers or gives a clear callback time under your rules. Time how long until the caller is offered a way forward.
- Hazard call. Say: "I smell gas near the furnace." A good answer gives a safety message first, to get out and call 911 or the gas utility from outside, before it takes any details, and then alerts you. Time the seconds until the safety message.
- Two calls close together. From two phones, call within a minute of each other about two different routine jobs. A good answer handles both, with no busy signal, long hold or voicemail for the second caller. Time the rings and any hold for each.
- Weekend estimate request (Saturday or Sunday). Say: "I'd like a quote for a new water heater, no rush." A good answer takes the job details and tells the caller honestly when you'll be in touch. Time hang-up to lead.
What should the scorecard look like?
Use one row per test call and one column per option, filled in with what you heard and timed, not what a sales page says. Leave the result cells empty until you've placed the calls, then mark each pass, partial or fail against the good answer you wrote down, with the timings beside it.
Print it, or copy it into a spreadsheet with one tab per round of calls. If you're comparing more than two options, add a column for each.
| Test call | What to measure | Live answering service | AI receptionist |
|---|---|---|---|
| After-hours emergency | Rings to answer; your questions asked; seconds until your phone rang; what happened when you didn't pick up | Your result | Your result |
| Price shopper | No invented price; details taken; callback time promised; hang-up to lead | Your result | Your result |
| Out-of-area caller | Service area caught; what the caller was told; no visit promised | Your result | Your result |
| Existing customer | Marked as existing; your rule followed; hang-up to lead | Your result | Your result |
| Asks for a person | Honest answer; transfer or a clear callback time; seconds to get there | Your result | Your result |
| Hazard call | Safety message first; seconds until it was given; you alerted | Your result | Your result |
| Two calls close together | Both answered; rings and hold time for each; no voicemail | Your result | Your result |
| Weekend estimate request | Details taken; honest callback expectation; hang-up to lead | Your result | Your result |
| Every lead text or email | Name, number, address or town, job, urgency and what was promised; time it arrived | Your result | Your result |
| The bill for test week | How each test call was charged; what your busiest month would cost | Your result | Your result |
How do I run the test fairly?
Give both options the same instructions, then place the same scripts at the same times from a number neither one knows. Repeat the full set at two or three times of day, and screenshot every lead text or email so its arrival time is on record. Fix your own instructions between rounds, never during one.
Ask each provider how you can call it before your customers do: a test number, a trial, or the first days after go-live while your current setup stays in place. Don't announce the test to the people answering in advance; that keeps it a fair look at a normal day.
To pick the times that matter, look at when your calls actually arrive. A missed-call audit of your own call log shows the hours you miss most, and those are the hours to test.
- Set both up the same way first: the same questions, the same service area, the same emergency rule and the same person to reach. A miss caused by your instructions is a setup problem, not a service problem.
- Keep the forwarding the same, so the caller hears the same rings before either option picks up. If your phone rings first and forwards when you don't answer, use the same delay for both.
- Call from a number neither option has seen, such as a family member's phone or a spare line, so nobody recognizes you.
- Read the scripts word for word. Use fictional names and your own shop address, never a customer's details.
- Run the full set at two or three times of day: early morning, midday while you'd be on a job, and after hours. Place the weekend call on a Saturday or Sunday.
- Write down when you dialed, the rings, what was said and what happened. Screenshot each lead text or email so its arrival time is on it.
- Tell your on-call team that test calls are coming, and end each call by saying it was a test, so nobody is sent out to a fictional job.
How do I read the results and the bill?
Start with the calls that cost most when they go wrong: the hazard call and the emergency. A fail there outweighs a slow text on a price shopper. Then compare how complete and fast the leads were, and finally the bill: how each test call was charged, and what your busiest month would cost under each billing model.
Per-minute services charge more in the months you're busiest, and the billing increments matter. Ruby's published billing, for example, rounds each call up in 60-second increments and counts hold time, while PATLive's published billing charges the first minute in full, then 6-second increments. Find your test calls on the invoice and check how many minutes each was charged. Ask each provider in writing what happens to your calls when volume spikes.
A flat-priced option costs the same in a storm week as in a quiet one, so check what it includes instead: which calls it answers, what it won't do, and what happens to a call it can't handle. Our guide to what an answering service costs a contractor works a storm month through both models, and the receptionist cost calculator runs your own minutes and call counts.
If an option failed a call because of your instructions, fix them and re-run that call before you decide. If it fails the same call twice with good instructions, that is your answer.
Where does Margot fit, and when is a live answering service the better choice?
Margot fits owners who miss calls on the job and after hours and want every lead texted to them at one flat monthly price. A live service may suit you better if you get under about 150 minutes of calls a month, want a person on every call, have Spanish-speaking callers, or need calls entered in your job software.
Margot is ApexScale AI's receptionist for contractors: an AI, not a person. She answers your forwarded missed and after-hours calls as your business, asks your questions, checks the address is in your service area, texts you every lead and live-transfers urgent calls under rules you write. She answers in English only, says she is an AI if a caller asks, does not dispatch, and has no integration with job software. We expect her to do well on this test, but your scorecard is the evidence, not our expectation.
On price, our home page says it plainly: under about 150 minutes of calls a month, a pay-as-you-go per-minute service can cost less than our flat Front Desk plan. Plans with bundled minutes move that line, so run your own numbers before you choose.
There are two free ways to start. Hear Margot answer as your business on a demo call of about two minutes, before you sign anything. Or request a free missed-call audit: Leo places five short, timed calls to your current line over about a week, at the times you're usually busy, and sends you a one-page report. If an answering service covers your line today, that report gives you an outside look at its column of the scorecard.
To fill in Margot's column on your own line, run the test in your first two weeks. Full refund if you cancel within 14 days of Margot answering your first call.
Frequently asked questions
Is an AI receptionist more reliable than an answering service?
Not as a rule, and we don't claim it. We have not found an independent study comparing the two on home-service calls. Reliability depends on the setup, your instructions and the time of day, which is why this test runs on your own calls.
How many test calls do I need?
Eight scripted calls, repeated at two or three times of day, is usually enough to see a pattern without tying up your week. Re-run any call that failed once you've fixed your instructions.
Should I tell the answering service I'm testing it?
Not in advance, so you see a normal day. Do tell your own on-call team, and say it was a test at the end of each call so nobody is sent out.
Can I test Margot before I pay?
You can hear her on a free demo call that answers as your business. The full test on your own line happens after go-live, which is usually within 2 business days of signup. Full refund if you cancel within 14 days of Margot answering your first call.
What if both options fail the same call?
Then the problem is probably your instructions, your call forwarding or the script itself. Rewrite the rule for that call, check your carrier's forwarding settings, and test it again on both.
Related reading
AI receptionist evaluation
Do customers hang up on an AI receptionist? What callers react to and how to test it
What an Answering Service Costs a Contractor
Virtual receptionist vs. answering service vs. AI receptionist: how they differ on real calls
Answering Service for Plumbers: What to Require Before You Sign
Compare
AI receptionist
What does a water damage company lose when a 2 a.m. call goes to voicemail
Every lead lands on your phone
When Margot answers a call you can't, you get a text like this the moment it ends. Hear her answer as your business, or see your own numbers first with a free missed-call audit.
Sources
- Pricing, PATLive (billing increments, and the pay-as-you-go per-minute rate behind the roughly 150-minute comparison with ApexScale's flat Front Desk price). Accessed October 19, 2026.
- Plans and pricing, Ruby (billing increments and hold time). Accessed October 13, 2026.
Third-party figures describe their publishers' own products, samples, or surveys. They are context, not ApexScale results.
Full refund if you cancel within 14 days of Margot answering your first call.