How to Evaluate an Answering Service: 5 Checks

Last updated: September 12, 2026

What to check before choosing an answering service

Who answers at 2am, what the operators are scored on, how message accuracy gets measured, whether any report reaches your own account, and where a wrong message goes after somebody catches it. Those five checks separate two services that look identical on a website. Ask for a month of scored calls on a trial account before you sign.

Price and answer speed get compared to death, and both are easy to compare. The criterion that decides whether you keep a service for five years is message accuracy, and it never shows up in a sales call. It shows up three weeks in, at 6am, when the message waiting on your phone says a caller wanted to reschedule and the caller actually wanted to cancel.

Everything else demos fine. The portal, the app, the greeting in your own business name, a friendly voice on a test call you placed yourself. Accuracy stays invisible until a service has been answering your phone for a month, so an evaluation has to drag it into the open before anybody signs.

Answering service, call center, or voicemail

An answering service picks up in your business name, works from written instructions you gave them, and hands back a message or a dispatch. Those calls run short, usually under a minute, because the operator captures who called and what they need rather than solving it.

Voicemail deserves a fair mention. When your callers will happily wait until Monday, it costs nothing and has never written down a wrong callback number. A human earns the money on the caller who hangs up on a recording and dials somebody else.

How your account actually works

The account script decides your quality more than the service does. It holds the greeting an operator reads word for word, your definition of an emergency, the on-call schedule by day and hour, and the terms nobody is allowed to say, which is why somebody who has never heard of your business can answer correctly.

Ask who writes it, and ask how a change reaches the floor tonight. A service that emails you a blank form and types in whatever comes back has moved the quality problem to you, and the gap between emailing a schedule change and every operator seeing it is where after hours calls go wrong.

The five checks

Who answers at 2am

Coverage usually gets answered with a seat count. Push past it and find out whether the overnight shift is their own staff or an overflow partner, because services hand the small hours to each other under mutual arrangements, and that partner may be working from instructions that are a week old.

Escalation is the part nobody asks about. A written path names the second number, the wait between attempts, and the point where an operator stops dialing and sends the message anyway. Without one, somebody improvises at 2am with your customer on the line.

Answer time is where everybody opens, and this list almost left it off. It stays because the standard sits in your service agreement rather than in any industry rule, which makes it the one number you can enforce. A floor that picks up in nine seconds and takes the wrong callback number has still failed you.

What the operators are scored on

Find out whether a scorecard exists in writing, whether it is written per account, and which failures void a score outright. Most floors grade against one general rubric covering greeting, tone, and hold time, while the parts specific to you, your triage question and your on-call rules, sit outside it. Those are the parts your calls fail on, and the scorecard operators get graded on shows what an account level rubric covers.

Then ask how many of your calls a supervisor read last month. Manual review across the category runs at 1 to 3 percent of calls, per write ups from Verint and similar vendors, which is a fact about the work rather than an accusation. Say a floor takes 40,000 calls a month across 200 accounts and reads 2 percent. That is 800 scored calls, about four per account, enough to catch an operator who breaks your greeting every single time and thin for anything that goes wrong once a week.

Measuring message accuracy on your account

Two services can claim high accuracy while measuring completely different things. The common method counts complaints, so if nobody called to complain, the messages were accurate. That counts the errors you noticed, which is a biased sample of the errors that happened, since a wrong callback number mostly looks like a caller who never called back.

The other method compares the message against the recording that produced it. Somebody pulls the call, reads the message beside the transcript, and marks three failures: a required field left empty, a detail written down wrong, and a detail the caller never gave. That procedure sits in how message accuracy gets measured.

Nobody should be quoting you an industry accuracy number. No published benchmark held up when we went looking on 2026-09-08, so a service answering "we run at 99 percent" without saying how that number was produced is reading you its complaint count.

The report you get on your own account

Service agreements promise quality and reports show it. Ask to see a real monthly report from a comparable account with the identifiers removed.

A report worth reading names your account, the period, how many of your calls were scored, what rubric produced the score, and the critical fails with a reason attached to each. What makes it credible is two example calls, one strong and one weak, with timestamps you can open. A report carrying only good calls is marketing with a chart on it, and the field list is in what a scored client report contains.

What happens when a message goes wrong

Every service gets one wrong eventually, and the difference between them is what the next 24 hours look like. A critical fail that reaches a supervisor the same day is a different operation from one that surfaces inside a monthly average three weeks later.

Then ask where the fix lands. This one is uncomfortable for both sides, because the most common root cause of a bad message is a vague line in the account script rather than an operator having a rough night. When every operator misses the same criterion, coaching the floor changes nothing, and a service that has never once come back asking to rewrite a line of your own script is not doing root cause.

The questions to ask on the call

Read these out loud on the sales call and write the answers down. A service answering in adjectives is telling you something.

• Who is on the floor at 2am, and are they your employees or an overflow partner's?
• How does an operator find out my after hours rules changed yesterday?
• Is there a written scorecard for my account, and can I read the criteria?
• How many of my calls did a supervisor read last month?
• Message accuracy on my account: do you measure it, and against what?
• Can you show me two calls from last month, one good and one bad, with the recordings?
• When a message goes out wrong, who finds out, and does the fix land in my account script?
• Who inside your company can listen to a recording of my calls?

Asking for a month of scored calls on a trial account

One request sorts the field faster than the rest of the evaluation. Ask the service to take a slice of your calls for two weeks, then hand back the scored calls with the recordings attached and the criteria visible.

Most cannot do it, and some will offer a trial that proves only that their operators can answer a phone, which you already assumed. A service that comes back with fifteen scored calls, a rubric written against your account, and two calls they are not proud of has answered everything else on this page at once.

The limits are real. A trial runs while the service knows somebody is watching, and a quiet October tells you little about the first week of January. Run it anyway, because the discipline you see in a trial is the ceiling rather than the floor of what comes after you sign.

What a good service will admit without being asked

Some of the most useful answers in an evaluation are concessions, and a service that volunteers these is easier to trust on the rest.

• Answer time standards live in your service agreement, so a universal industry standard quoted at you is a marketing number.
• No published benchmark governs message accuracy, which means a target has to be built from a baseline on your own account.
• Automated scoring holds up on rule based criteria, whether the greeting matched or the callback number was read back digit by digit, and gets shakier on tone, as Aircall and StackAI both note. Whether an operator sounded like they cared is still a human read.
• A scorecard cannot fix a script, and a floor working from bad instructions fails consistently no matter who coaches it.

Pricing questions that actually matter

Pricing comes in three shapes, and the shape does more damage than the number. Billing by the minute rounds up to the next increment, so ten 40 second calls can bill as ten full minutes. Per call billing is simpler and quietly rewards a service for getting off the phone. A monthly plan with an included allowance is the common arrangement, and the allowance is where the surprises live.

Ask what counts as billable. Wrong numbers, spam calls, hold time, a call that transfers straight through to you, and the seconds an operator spends looking up your on-call schedule get handled differently by different services, and none of it appears on a pricing page.

Ask for the overage rate next to the effective rate inside your plan, because overage prices higher and a month at double your volume costs more than double. Then ask how you leave, since a twelve month term you can exit on 30 days notice is a different purchase from a twelve month term you owe in full, and both get described on a call as a twelve month agreement.

If the calls involve a medical practice

No federal agency certifies an answering service as HIPAA compliant, so a badge on a website is a claim the service wrote about itself. A service taking calls for a medical practice generally acts as a business associate under a signed agreement, and what protects you is that agreement plus what the service does with the information.

Ask for the agreement. Find out who inside the service can hear a recording, how long audio and messages are kept, and whether any audio gets sent to an outside company for transcription. HIPAA questions for a medical practice walks the full list, and your own counsel decides what your practice needs.

What to ask for in writing before you sign

Five requests, all of them in writing. A service running a real quality program can send every one inside a day, and a service without one will send you a case study instead.

1. The scorecard your account would be graded against, with the criteria and the auto-fail list written out.
2. A sample of their own scored calls, two from last month on an account like yours, with the recordings attached and the scores visible.
3. Their quality process in plain terms: how many calls get read, by whom, and what happens to one that fails.
4. Escalation and on-call rules exactly as they would be entered for your account, including who gets called when the first number does not answer.
5. Pricing with the overage rate, the billing increment, and what counts as a billable call.

Nothing there requires a project on their end. Services running CallSniper can produce that scored call report for one account in a few minutes, and a service scoring by hand can do the same with a supervisor and an afternoon.

The comparison only works if both services get the same list on the same day, so send it to the two on your shortlist and give them until Friday. QA for telephone answering services covers the operator side of this conversation, if you run a service rather than buy from one.

If neither can produce a scored call on your own account, keep looking, or stay where you are and put that request into your next quarterly review.

Frequently asked questions

What is the difference between an answering service and a call center?

An answering service answers in your business name, works from the instructions you wrote, and hands back a message or a dispatch, usually in under a minute. A call center runs longer calls on one program with agents in your queue all shift, and the deliverable is the real split: a message on one side, a resolved call on the other.

How much does an answering service cost per month?

Pricing arrives as per minute, per call, or a monthly plan with an included allowance, and the shape decides what a busy month does to your bill. Ask for the overage rate beside the effective rate inside the plan, since overage prices higher and a spike month costs more than the arithmetic suggests.

What questions should I ask an answering service?

Start with overnight coverage and escalation, then what operators are scored on, how message accuracy gets measured, what reporting reaches your account, and what happens after a bad message. Read the question bank above out loud on the call and listen for answers with a number or a document behind them.

Is an answering service better than voicemail?

For callers who will wait until Monday morning, voicemail is fine, cheap, and it has never taken down the wrong number. A service earns the money on callers who hang up on a recording and dial somebody else, and on anything that has to reach a person tonight.

How do I know if my answering service is doing a good job?

Ask for scored calls on your own account, meaning how many were reviewed last month, against what criteria, and which ones failed. Listening to two of them yourself, one they chose and one you picked at random, settles most of it in ten minutes.

Sources

• Trinsic call notes, 2026-08-27 and 2026-09-03, on sub 60 second calls and answer time standards set per account. Verified internally.
• Research pass on 2026-09-08 found no published message accuracy benchmark for answering services. Verified as an absence that day.
• Manual QA covering 1 to 3 percent of calls, from Verint, Gistly, and ViciStack. Third party, unverified, setup only.
• Aircall and StackAI on rule based criteria scoring more reliably than tone. Third party, unverified.
• Business associate framing for a service handling calls for a covered entity. Pending counsel review.

Related industriesAll industries