AI voice agents by role

AI customer support phone agent: answering service calls 24/7, and handing over well

The short answer

An AI customer support phone agent answers your service line at any hour with no queue. It checks who's calling, sorts out routine requests such as order status, balances and bookings, and passes complaints and complicated cases to a person with the details captured. Judge it on problems solved, not calls kept away from people.

By , founder and CEO of ShiftMateUpdated 15 min read
A customer on the phone, getting help from an AI customer support phone agent
Mike Steenkamp, founder and CEO of ShiftMate, a South African voice agent specialist
“We're ShiftMate, a voice agent specialist based in South Africa. We build inbound voice agents for each business, and train the people who take over when the AI shouldn't. This is our view of AI customer support by phone: which calls to hand over, which to keep, and how to tell if it's working.”
Mike Steenkamp, founder and CEO of ShiftMate, Umhlanga, South Africa
Key takeaways
  • An AI customer support phone agent handles frequent, simple, rule-based calls end to end and hands the rest over with a summary. It's one use of AI voice agents.
  • Measure resolution, repeat calls within 7 days and satisfaction for AI-served vs person-served callers. A high "containment" rate can hide customers who simply gave up.
  • Gartner's 2024 survey of 187 service leaders found 44% exploring a customer-facing AI voicebot, 11% piloting and 5% with one live. Most are still early.
  • Test every call type separately, from where your callers are, before launch. A rule that's right for one kind of call can be wrong for another.
  • ShiftMate builds each inbound support agent around your own call reasons and systems, and trains the people who take its handovers with AI practice calls.

What is an AI customer support phone agent?

An AI customer support phone agent is an AI voice agent that answers your customer service line in normal spoken conversation. The customer says what they need in their own words; the agent checks who they are, looks up their order, account or booking, sorts out the routine request, and hands anything else to a person with a summary. It's one of the main uses of AI voice agents.

You'll also see it called AI phone support, an AI customer service voice agent, an AI contact centre agent or a conversational IVR. The promise behind all of them is 24/7 AI customer support: no queue at 22:00, no queue on month-end, no queue when a storm knocks out the network and everyone phones at once.

Answer

Picks up at once

No menu and no hold music. Many customers can be helped at the same time, so peaks don't become queues.

Identify

Checks who's calling

With the same identity checks your agents use, before it shares anything personal.

Resolve

Sorts out routine requests

Order and claim status, balances, payment dates, bookings, simple changes, documents needed.

Hand over

Passes on what it shouldn't handle

Complaints, distress, exceptions and anything it isn't sure about go to a person, with the details already captured.

Which customer service calls should AI handle, and which should go to people?

AI should handle customer service calls that are frequent, simple and rule-based, where the answer is in your systems. People should handle calls that are emotional, unusual or need someone with authority. Before you scope an AI customer support phone agent, pull a month of call reasons from your system and sort them into the two columns below.

Let the AI handle itHand it to a personWhy
Where's my order, claim or delivery?A complaint or a disputeStatus is a lookup; a dispute needs judgement and authority
Balance, statement and payment date questionsA customer in financial or personal distressVulnerable customers need care, not a script
Bookings, changes and cancellationsAn exception to your rulesExceptions need someone allowed to make them
Password and account help, with proper ID checksAnything that looks like fraudFraud needs a trained person and a record
Opening hours, branches, documents neededAnything the AI isn't sure aboutGuessing is worse than handing over
Reporting a fault and getting a reference numberA customer who asks for a personArguing with that request loses trust fast

Be careful with the promise of automating almost everything. McKinsey describes a healthcare organisation weighing a plan to automate 80 to 90% of interactions within 24 months, designed only for simple, happy-path calls; real callers brought emotional calls about several things at once (McKinsey, June 2026). Choose the call types your AI customer support phone agent can finish properly, and start there.

How does a 24/7 AI customer support call work?

A call to an AI customer support phone agent works in five steps: answer, identify, look up, resolve or hand over, and log. Each step has a rule that keeps it safe.

  1. 1

    It answers at once

    It says it's an AI assistant, says the call is recorded, and asks how it can help. No menu: the customer explains in their own words.

  2. 2

    It checks who's calling

    The same identity checks your agents use, before it shares anything personal. Under POPIA, personal information goes only to the right person.

  3. 3

    It looks things up

    In your order, billing, ticketing or booking systems, through connections set up for that job, with access only to what the task needs.

  4. 4

    It sorts it out or hands over

    It completes the routine request, or transfers the call to the right team with a short summary: who, what they want, what's been checked.

  5. 5

    It logs the call

    A transcript, the outcome and the reason, in your CRM or ticketing system, so a person can pick it up later and so you can review it.

A routine call and a handover
Made-up internet provider. A customer phones at 21:15 on a Sunday.
AI agent

Hi, you're through to the support line. I'm an AI assistant and this call is recorded. What can I help with?

Customer

Hi, I'm phoning about my fibre. It's been down since this afternoon.

On an inbound line, this is a normal opening, not a mistake.

AI agent

Sorry about that. Can I have your account number or ID number to find your line?

Customer

It's under Ravi Govender, 8203 ...

AI agent

Thanks, Ravi. There's a reported outage in your area since 14:00, and the network team expects it back tonight. I can SMS you when it's restored. Would that help?

Customer

Ja, but this is the third time this month. I want to cancel.

AI agent

I understand. Cancelling is something a colleague handles, so I'll put a request in for a call first thing tomorrow, with today's details, so you don't have to explain it again. Is this number best?

A cancellation is a decision for a person. The AI hands it over with context instead of trying to save the sale.

How is AI phone support different from IVR and conversational IVR?

AI phone support is different from IVR because the customer talks instead of pressing numbers, and the agent can often finish the request instead of only routing it. A conversational IVR sits in between: it understands speech, but mostly to send the call to the right queue.

IVR (press 1)Conversational IVRAI customer support phone agentPerson
How customers use itKeypad menusSay a reasonTalk normallyTalk normally
Sorts out the requestRarelyRarely: it routesOften, for routine requestsYes
After hoursPlays a messageRoutes to voicemailHandles the callOnly with night staff
When it can't helpBack to the menuInto a queueHands over with a summaryEscalates
Handles emotion and exceptionsNoNoRecognises them and hands overYes
Cost per extra call at peakLowLowPer minuteAnother agent

If your IVR mostly exists to send people to the right team, a conversational IVR may be enough. If most calls are the same few questions that your systems can answer, an AI customer support phone agent that resolves them will do more.

What do analysts expect from AI in customer service?

Analysts expect AI to handle a growing share of routine service calls, but not to remove people, and not always to be cheaper. These are predictions and surveys, not measured results:

Gartner survey, December 2024

Most teams were still exploring

Of 187 customer service leaders surveyed in mid-2024, 44% were exploring a customer-facing AI voicebot, 11% were piloting one and 5% had one deployed. Source.

Gartner prediction, March 2025

80% of common issues by 2029

Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention. Source.

Gartner prediction, September 2025

People stay

Gartner predicts that by 2028 no Fortune 500 company will have fully eliminated human customer service, and that by 2027, half the organisations that planned big service-staff cuts because of AI will drop those plans. Source.

Gartner prediction, January 2026

AI won't always be cheaper

Gartner predicts that by 2030, cost per resolution for generative AI will exceed USD 3, more than many offshore human agents, and that AI regulation will push assisted service volume up 30% by 2028. Source.

For an AI customer support phone agent, the shared message is this: AI takes the routine calls, people take the hard ones, and the handover between them is where service is won or lost.

What does an AI customer support phone agent cost?

An AI customer support phone agent is usually priced per minute of calls, as a monthly bundle, or as a custom contract for larger contact centres, plus set-up and the work of connecting your systems. Gartner, back in 2022, estimated integration at USD 1,000 to 1,500 per conversational AI agent, with some organisations citing up to USD 2,000 (Gartner, August 2022). Many platforms bill in US dollars, so in South Africa ask for a rand price, or at least know how the exchange rate moves your bill.

Published per-minute prices depend on what they include: one developer platform lists US$0.07 to US$0.31 a minute, built from itemised parts (voice engine, voice, AI model and a US$0.015 phone line), while hosting-only building blocks start at about US$0.01 a minute and bill the AI models and phone line separately (vendors' own pricing pages, checked 16 November 2026; see AI voice agent pricing).

Compare on cost per resolved call, not cost per minute. A cheap minute that ends in a repeat call costs you twice: once for the AI, once for the person who fixes it the next day. Our AI voice agent pricing guide breaks down every part of the price.

How do you know your AI customer support phone agent is working?

You know an AI customer support phone agent is working when problems get solved, not when calls get kept away from people. McKinsey warns that judging AI on containment, deflection or cost per contact makes it hold on to calls it should transfer, so repeat calls go up and satisfaction goes down (McKinsey, June 2026). Track these instead:

  • Resolution by call reason: the share of calls completed end to end, with the customer's problem actually fixed.
  • Repeat calls within 7 days on the same issue (the clearest sign the AI didn't really solve it).
  • Handover rate, and whether each handover summary was right, so the customer didn't repeat themselves.
  • Satisfaction for AI-served callers vs person-served callers on the same call reasons.
  • Waiting time and handling time for the calls that reach people.
  • Complaints that mention the AI, and requests for a person that weren't honoured.
  • Speed: the pause before each real answer, measured from where your callers are.
What we learned: Trust
“A spoken promise isn't a completed action. Our agents only tell a caller something is done once the system confirms it is done, not because the conversation went well.”
Mike Steenkamp, founder and CEO of ShiftMate

How do you test an AI phone agent before customers hear it?

You test an AI phone agent by putting it through practice calls with difficult pretend customers, one call type at a time, and listening to every one before a real customer hears it. Live monitoring after launch matters too, but it finds problems after customers have met them.

Practice callerWhat it testsPass if
The angry repeat callerStaying calm, owning the problem, handing overNo arguing, quick handover with the history
Several problems at onceKeeping track of each oneEvery issue is resolved or handed over, none dropped
The interrupterStopping and listeningIt stops mid-sentence and answers the new question
Strong accent on a noisy cell lineNames, numbers and account detailsIt gets them right, or reads them back to check
Switching languages mid-sentenceUnderstanding English mixed with isiZulu or AfrikaansIt follows, or offers a person who speaks the language
A vulnerable customerRecognising distressIt hands over at once, with care
"Just put me through to a person"Respecting the requestIt transfers or books a call back, without persuading
Fails identity checksProtecting personal informationIt shares nothing and offers another way to verify

Then measure speed. In normal conversation the gap between speakers is often around a fifth of a second (Stivers and others, PNAS 2009), and on a support line long silences make people repeat themselves or think the line has dropped.

What we learned: Testing each call type
“A rule we wrote for outbound calls broke inbound ones: it flagged a perfectly normal caller saying "Hi, I'm phoning about my fibre" as a mistake. The same sentence is right on one kind of call and wrong on another, so we test every call type on its own.”
Mike Steenkamp, founder and CEO of ShiftMate
What we learned: Speed (latency)
“We learned to measure the real answer, not the filler. A quick "Sure" makes a speed number look good while the caller is still waiting for what they asked. And we measure the round trip from a South African caller, not from a server next door to the AI.”
Mike Steenkamp, founder and CEO of ShiftMate

What an AI support agent won't fix

An AI customer support phone agent won't fix a broken process, and it changes the work your people do. Plan for both.

Broken processes

If customers phone because bills are wrong or deliveries are late, the AI will answer those calls faster, but the calls keep coming. Fix the reason people phone.

Systems it can't reach

Without read and write access to your order, billing or booking systems, it can only take messages. Connect first, then automate.

Harder calls for your people

Once AI takes the simple calls, the ones left for people are the angry, complex and emotional ones. One large operator told The National it has stretched agent training from three weeks to eight because of AI (The National, November 2026). Your team needs practice for those calls.

Power and network failures

Load-shedding and network outages cause call spikes. Decide what happens when your systems are down: an honest message, an SMS when it's fixed, or a call back.

Trust

Customers who've fought with bad IVRs start out suspicious. An agent that says what it can and can't do, and hands over when asked, earns trust; one that pretends loses it.

Questions to ask a supplier of AI phone support

  • Which identity checks does it run before sharing personal information, and are they the same as our agents' checks?
  • Can it write to our systems (update a booking, log a fault, open a ticket), or only read them?
  • What does the person who takes a handover see: a summary, the transcript, what was already checked?
  • Can it report repeat calls within 7 days by call reason, not just containment?
  • Where are recordings and transcripts stored and processed, and how is POPIA handled?
  • How long is the pause before each real answer, measured from a South African cell phone?
  • What does a resolved call cost, all in, in rand?
  • Can our team change answers and handover rules without waiting for the supplier?
  • Can we start with one call type as a pilot before signing a long contract?

Support judged on the repeat call

We think the honest test of AI support is the repeat call. If a customer rings back within a week about the same problem, the AI didn't fix it, whatever the containment rate says. The support lines that get this right will automate fewer call types than they could, finish each one properly, and hand the rest over with the whole story attached. We'd like to hear from support teams in call centres, and with businesses that have never had a support line at all, that are willing to measure it that way.

We build each inbound support agent around your call reasons, systems and handover rules, in a South African voice, and before it takes a real customer it has to get through practice calls with the angry caller, the confused one and the one with two problems at once, while AI practice calls get your own agents ready for the angry and complicated calls it passes to them.

Tell us which calls fill your queue and the repeat-call rate you'd like to beat, and we'll say where an agent would help. For a small business front desk, read the AI receptionist guide; for a chat agent on your website, see AI Customer Service Chat.

What we learned: What AI does for call centres
“In a call centre, voice AI is support, not a replacement. It trains the people, generates the leads, and makes the low-value calls the operators don't want to make, so they spend their day on the conversations that need a person.”
Mike Steenkamp, founder and CEO of ShiftMate

Questions people ask

What is an AI customer support phone agent?

An AI customer support phone agent is an AI voice agent that answers your customer service line at any hour. It checks who's calling, sorts out routine requests like order status, bookings and balances, and hands complicated or emotional calls to a person with a summary of the call.

How much does an AI customer support phone agent cost?

It's priced per minute, as a monthly bundle or as a custom contract, plus set-up and integration. One developer platform lists US$0.07 to US$0.31 a minute, built from itemised parts (voice engine, voice, AI model and a US$0.015 phone line), and Gartner estimated integration in 2022 at USD 1,000 to 1,500 per conversational AI agent. Compare suppliers on cost per resolved call in rand.

Can an AI phone agent replace a call centre?

No. It can take a large share of routine calls, but complaints, exceptions and vulnerable customers need people. Gartner predicts that by 2028 no Fortune 500 company will have fully eliminated human customer service. The best results come from AI and people working together, with clean handovers.

How is an AI phone agent different from IVR?

IVR makes callers press numbers to move through a menu and mostly routes calls. A conversational IVR understands speech but still mainly routes. An AI phone agent lets callers say what they need and can often complete the request itself, handing over when it can't.

Is AI phone support safe and legal?

It can be, with the same identity checks your agents use, access limited to what each task needs, and a record of every call. Callers should hear that they're talking to an AI and that the call is recorded. In South Africa, POPIA governs personal information, recordings and transfers abroad. This is general information: take legal advice.

What are the common mistakes with 24/7 AI customer support?

Judging it on containment instead of problems solved, automating calls your systems can't finish, weak handovers that make customers repeat themselves, refusing callers who ask for a person, and skipping tests on real accents and noisy lines. Watch repeat calls within 7 days: they show what the AI didn't really fix.

How do you test an AI customer service voice agent?

Run practice calls with difficult pretend customers before launch: angry, confused, interrupting, accented, switching languages, vulnerable, or asking for a person. Test each call type on its own, measure the delay from where your callers are, and listen to every call.

Does ShiftMate offer an AI support phone agent?

Yes, and it starts from your queue. ShiftMate builds each support phone agent on your top call reasons and the systems that hold the answers: it resolves what those systems allow and hands the rest to a person with a summary, so no customer repeats the story. ShiftMate also sells AI practice calls that train human support agents.

Shorter queues, without losing the human touch

A support agent built around your call reasons, finishing the routine ones and handing over the rest with context. Send us your top ten call reasons and we'll mark which ones AI can resolve and which need a person.