Voice AI market

Voice AI trends 2026: what's really changing, the evidence, and what it means for your business

The short answer

The voice AI trends that matter in 2026: agents moving from answering calls to resolving them, faster replies, testing every call before and after launch, local accents and languages, big funding and takeovers, stricter rules on AI calls, and people working beside AI on the harder calls. Adoption is still earlier than the headlines suggest.

By , founder and CEO of ShiftMateUpdated 15 min read
People and AI working together in a contact centre, one of the voice AI trends of 2026
Mike Steenkamp, founder and CEO of ShiftMate, a South African voice agent specialist
“ShiftMate builds voice agents in South Africa, so these trends aren't theory for us: they shape what we build. This guide covers the voice AI trends that matter in 2026, from faster replies and local voices to testing and tighter rules, and which ones a business deciding now should care about.”
Mike Steenkamp, founder and CEO of ShiftMate, Umhlanga, South Africa
Key takeaways
  • Adoption is earlier than the hype: in a Gartner survey of 187 service leaders in mid-2024, only 5% had a customer-facing generative AI voicebot deployed, 11% were piloting and 44% exploring. Start with our guide to AI voice agents.
  • Analytics, testing and QA is the fastest-growing part of the voice AI agents market, according to a Research and Markets report of 30 October 2026, which forecasts the market growing from USD 3.40 billion in 2025 to USD 43.83 billion by 2036.
  • Rules are tightening: the US FCC treats AI voices as artificial voices that need prior consent (February 2024), and South Africa's national opt-out registry is being phased in, with direct marketers registering from 15 September 2026.
  • People aren't going away: Gartner predicts no Fortune 500 company will have fully eliminated human customer service by 2028, and the calls left for people are the harder ones, which means more training, not less.

Why are voice agents moving from answering to resolving calls?

Voice agents are moving from answering to resolving because answering alone doesn't save money or please customers: an agent that answers a question and then transfers the caller has added a step. The value is in finishing the job (the booking made, the address changed) where that's safe.

The forecasts point that way. Gartner predicted in March 2025 that by 2029 agentic AI will resolve 80% of common customer service issues without human intervention, leading to a 30% cut in operating costs (Gartner). That's a forecast, not a result. McKinsey's June 2026 article on why voice agents struggle makes the same point from the failure side: leaders who ask how much they can deflect, instead of where an agent can truly resolve a call, end up with agents doing jobs they can't do (McKinsey).

Resolving calls also explains another of the voice AI trends in the round-ups: industry specialists. An agent that resolves insurance claims has to know claims, the rules and the words customers use. The Research and Markets report of October 2026 names banking, financial services and insurance as the largest users, healthcare as the fastest-growing, and patient engagement as the fastest-growing application (Research and Markets via Yahoo Finance). General agents answer; specialists finish.

What it means for a business: start with one call type that an agent can finish from start to end, connected to the system that does the work. Measure resolution and repeat calls, not containment (calls that never reached a person, which can hide callers who simply gave up).

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 fast does a voice agent need to be, and why is speech-to-speech growing?

A voice agent needs to reply about as fast as a person would. Research across ten languages found the usual gap between one speaker finishing and the next starting is a fraction of a second (turn-taking study, PNAS, 2009). Long pauses make callers talk over the agent, repeat themselves or hang up, however good the answer is.

That's why speech-to-speech systems, which work on the sound directly rather than turning speech into text and back, are among the clearest voice AI trends. The Research and Markets report says the older step-by-step (cascaded) approach still holds the bigger share, but speech-to-speech is growing faster, and that real-time models now handle lower delay, interruptions and more expressive voices.

What we learned: Speed (latency)
“One of our earlier AI interviewers was good, but slow to reply, and on its first test run it dropped a turn. That taught us that a clever answer delivered late loses to a simple answer delivered on time, so we rebuilt for speed.”
Mike Steenkamp, founder and CEO of ShiftMate

The trap in speed is measuring the wrong thing.

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 it means for a business: ask suppliers for the delay from the end of the caller's sentence to the start of the real answer, measured on a call from a South African number.

Why is testing the fastest-growing part of voice AI?

Testing is the fastest-growing of the voice AI trends because businesses have learned that a voice agent which passes a demo can still fail on real people, and at AI volumes nobody can listen to enough calls to find out by hand. The Research and Markets report names analytics, testing and QA as the fastest-growing component of the voice AI agents market, because buyers now put compliance, performance monitoring and safety first (Research and Markets via Yahoo Finance).

Traditional quality checks review a small sample of calls, often quoted as 1% to 2% (a figure from a voice platform, repeated in the round-ups). That breaks down when an agent takes thousands of calls a day. Among conversational AI trends, automated review of every call is the obvious answer after launch. Most round-ups miss the step before it: testing the agent on difficult pretend callers before a single customer hears it.

What we learned: Testing before launch
“Before a version of one of our AI sales agents may phone anyone, it has to pass every practice situation three times: the busy prospect, the one who only wants an email, the one who asks if it's a robot, the one who says take me off your list. Passing once proves nothing.”
Mike Steenkamp, founder and CEO of ShiftMate

What it means for a business: write your test callers down (the angry one, the confused one, the one who asks if it's a robot, the one who wants to be taken off the list), set a pass mark for each, and run the full set again after every change. Our page on why AI voice agents fail explains the failures to test for.

Will voice AI understand local accents and languages?

Voice AI is getting better at local accents and languages, but it isn't there yet for most of Africa, and that gap is one of the voice AI trends that matters most for South Africa. The AfriSpeech-200 study found that off-the-shelf speech recognition makes several times more errors on African-accented English, including South African accents, than on standard US benchmarks (AfriSpeech-200, 2023). The Research and Markets report lists accuracy across accents and dialects as a challenge, and multilingual agents for emerging markets as an opportunity.

Local builders are responding. Cape Town's Untapped AI told iAfrica in June 2026 that its data shows drop-off is "significantly higher" when customers hear robotic or foreign accents (the company's own claim), and that it recorded 20 South African voice agents in a studio (iAfrica).

What we learned: South African voices
“The market responds better to natural, authentic South African voices. An American or British voice on a South African line sounds like a scam call. So our sales agents always sound South African, from the first word to the last, even on UK or US campaigns.”
Mike Steenkamp, founder and CEO of ShiftMate
What we learned: South African speech
“"Yeah, no" is a South African way of starting an answer. It isn't a no. An agent built on American speech hears a refusal; ours has to listen to what comes after it, and if nothing clear follows, ask.”
Mike Steenkamp, founder and CEO of ShiftMate

What it means for a business: test any voice agent on recordings of your own customers, including people who switch between English and Afrikaans or isiZulu mid-sentence, before you sign. For the full South African picture, see AI voice agents in South Africa.

Where is the money going in voice AI?

Money is one of the loudest voice AI trends: it's going into voice AI fast, and the larger companies are buying the smaller ones. CB Insights counted voice AI funding of USD 315 million in 2022 and USD 2.1 billion in 2024 (CB Insights). Forecasts of the market disagree because each firm defines it differently: Research and Markets puts it at USD 3.40 billion in 2025, growing to USD 43.83 billion by 2036; Grand View Research at USD 2.5 billion in 2025, growing to USD 35.2 billion by 2033 (Grand View Research). Both are forecasts.

Deal (checked 16 November 2026)What happenedSource
Sierra (customer service agents)Raised USD 950 million in May 2026, valued at over USD 15 billionTechCrunch
Parloa (Berlin, customer service agents)Raised USD 350 million in January 2026 at a USD 3 billion valuationReuters
PolyAI (London, enterprise voice agents)Raised USD 86 million in December 2025 at a USD 750 million valuationForbes
Salesforce and CrestaAgreed to buy Cresta in November 2026; price not disclosedSalesforce

What it means for a business: consolidation is good for stability and bad for choice. Before you sign, check that you can export your recordings, transcripts and agent setup, and what happens to your contract if the supplier is bought. Watch the cost per resolved call, not just the per-minute price: Gartner predicts that by 2030 the cost per resolution for generative AI in customer service will exceed USD 3, higher than many offshore human agents (Gartner, January 2026). Prices for voice AI in 2026 vary widely, and we found no reliable source showing how fast per-minute prices are falling, so we don't quote a trend; our AI voice agent pricing guide shows what each price includes, and voice AI companies covers who's who.

What rules on AI calls are coming in?

Of all the voice AI trends, the rules on AI calls are the one most businesses underestimate. They're tightening on three fronts: disclosure, consent and fraud. In the US, the FCC ruled on 8 February 2024 that AI voices count as "artificial" under the TCPA, so AI calls need prior express consent, in writing for telemarketing. In Europe, the EU AI Act's transparency duties are due to apply from August 2026, according to the Research and Markets release. Fraud is the background: the US FTC reported USD 12.5 billion in consumer fraud losses in 2024, about USD 2.95 billion of it from impostor scams (both cited in the Research and Markets release). Cloned voices can make callers more suspicious of every voice they don't know.

South Africa moved too: the Consumer Protection Act Amendment Regulations took effect on 15 April 2026. South Africa's national opt-out registry is being phased in: direct marketers register with the National Consumer Commission from 15 September 2026, cleanse their lists monthly (free cleansing December 2026 to April 2027), and consumers can block marketers from May 2027 (NCC). POPIA section 69 already required consent for direct marketing by automatic calling machines, with a narrow exception for existing customers. Gartner also predicts that by 2028 regulatory changes related to AI will increase assisted service volume by 30%, as people use their right to reach a person (Gartner, January 2026). This is general information, not legal advice.

What we learned: Honesty
“Our agents say they're an AI and that the call is recorded. People forgive a machine that's honest with them; they don't forgive one that pretends.”
Mike Steenkamp, founder and CEO of ShiftMate

Voice-cloning fraud and voiceprint log-ins

The technology that makes agents sound human also makes voices easy to copy. With impostor scams costing US consumers about USD 2.95 billion in 2024, a voice on its own is weaker proof of identity than it used to be. If customers log in by voiceprint, add a second check, and never let an agent accept a voice alone as proof for a payment or an account change.

What it means for a business: build disclosure, consent records, calling times, opt-outs and a route to a person into the agent from the first version. Retrofitting them later costs more than building them in.

What won't voice AI solve in 2027, and where do people fit?

Voice AI won't fix a broken process, won't make a bad offer good, won't replace judgement on sensitive calls, won't understand every caller on every line, and won't remove the need for people. None of the voice AI trends of 2026 changes that; the checklist at the end is what we'd do in 2027.

The future of voice AI agents isn't a world without people. Gartner predicts that by 2028 none of the Fortune 500 will have fully eliminated human customer service, and that by 2027 half of the organisations that planned big service workforce cuts because of AI will drop those plans (Gartner, September 2025). Adoption is also earlier than the headlines: in Gartner's mid-2024 survey of 187 service leaders, 44% were exploring a customer-facing generative AI voicebot, 11% were piloting one and 5% had one deployed (Gartner).

What changes is the work people do. When AI takes the simple calls, people get the angry, complicated and emotional ones. In South Africa's outsourcing sector, which created 26,346 new jobs serving international clients in 2025 according to industry body BPESA (BPESA), one large contact centre company told The National in November 2026 that AI adoption had pushed its training from an average of three weeks to eight (The National). Investors have noticed the training side too: a16z wrote in 2025 that realistic voice agents can act as "simulators" for coaching and training (a16z).

  • Start with one call type an agent can resolve end to end, connected to the system that does the work.
  • Write down your difficult test callers and pass marks before launch, and rerun them after every change.
  • Measure resolution, repeat calls and complaints, not just containment.
  • Test on your own callers' accents, languages and phone lines, and measure delay from where they are.
  • Design disclosure, consent, calling times and opt-outs in from day one, for every country you call.
  • Keep a fast route to a person, with the call summary handed over.
  • Train your people for the harder calls AI leaves them, before those calls arrive.

The trend we're betting on

Of all the voice AI trends on this page, the one we think matters most for a business deciding now is specialisation. Faster replies and more natural voices reach every supplier at much the same time. What will separate them is how well an agent knows one kind of call, one industry and one country's callers, and whether the supplier can prove it on difficult practice calls before launch and after every change.

That's where we put our effort: our AI outbound sales agent and AI practice calls run on one engine built for South African calls, described on our Gen 4 voice engine page.

Tell us which calls you'd automate first, and we'll tell you which of these trends changes them.

What we learned: Where the market is going
“Voice AI isn't replacing call centres. What it's doing is making voice available and affordable to businesses that could never afford a call centre, and supporting the call centres that exist: training their people, generating leads, and making the low-value calls their operators don't want to make. This space is changing and opening up.”
Mike Steenkamp, founder and CEO of ShiftMate

Questions people ask

What are the biggest voice AI trends in 2026?

The biggest voice AI trends in 2026 are agents that resolve whole calls, faster speech-to-speech replies, testing every call before and after launch, local accents and languages, industry specialists, heavy funding and takeovers, stricter rules on AI calls, and people working beside AI on the harder calls.

What is the future of voice AI agents?

The future of voice AI agents is more calls finished end to end by AI, with people taking the complex and emotional ones, under clearer rules on disclosure and consent. Gartner predicts no Fortune 500 company will have fully eliminated human customer service by 2028, so most businesses should plan for people and AI together.

How big is the voice AI agents market?

About USD 2.5 billion to USD 3.4 billion in 2025, depending on the research firm. Research and Markets forecasts USD 43.83 billion by 2036; Grand View Research forecasts USD 35.2 billion by 2033. The estimates differ because each firm defines the market differently, so always name the firm and year when quoting one.

How widely are AI voice agents used in customer service?

Still early. A Gartner survey of 187 customer service leaders in mid-2024 found 5% had a customer-facing generative AI voicebot deployed, 11% were piloting one and 44% were exploring one. We found no reputable survey of AI adoption in South African contact centres when we checked in November 2026.

How much does voice AI cost per resolved call?

Prices vary widely by what they include (phone line, speech, AI and set-up), so compare the all-in cost per resolved call, not the per-minute rate. Gartner predicts that by 2030 the cost per resolution for generative AI in customer service will exceed USD 3, higher than many offshore human agents, so automation isn't automatically cheaper.

What are the common mistakes businesses make with voice AI?

Three stand out: measuring containment (calls that never reached a person) instead of resolution, which hides callers who gave up; launching without testing on difficult pretend callers, so customers find the failures; and ignoring accents and languages, so the agent mishears the people it serves. Each is cheap to avoid before launch and expensive after.

Are AI phone calls legal in South Africa?

Yes, within the rules. POPIA section 69 requires consent for direct marketing by automatic calling machines, with a narrow existing-customer exception, and the National Consumer Commission's opt-out registry is being phased in: marketers register from 15 September 2026, and consumers can block them from May 2027. This is general information; take legal advice.

Will voice AI replace call centre agents?

Not fully. Gartner predicts that by 2027 half of organisations that planned major service workforce cuts because of AI will drop those plans. AI takes routine calls; people handle the complex, emotional and sensitive ones, which need more training. One South African contact centre company reported lengthening training from three weeks to eight.

See which trends change your calls

Voice agents built for South African callers and proven on difficult practice calls first, with training for the people who work beside them.