The AI-voice-agent pitch is everywhere in 2026: answer every call, never miss a lead, cut the receptionist budget. The employment data backs the adoption story — call-centre hiring is shrinking in almost every economy that tracks it. But a new performance analysis has complicated the assumption underneath it, and for the people who work the phones, the gap between what AI gets deployed for and what it actually delivers is where the job security lives.
🔍 THE BOTTOM LINE
An October analysis by Bold Agent Kit, a company that builds AI receptionists, examined roughly 14,000 real inbound customer calls and found human receptionists achieved about four times stronger inbound intake conversion performance than AI agents — 6,343 AI-handled calls against roughly 7,657 human-handled ones. That lands in the same week as Andreessen Horowitz’s chart-heavy review of Revelio data showing global call-centre employment growth has turned sharply negative, including in lower-middle-income countries that had been insulated through 2024. Both findings are real, and together they describe a labour market that is automating faster than it is able to replace well: the jobs are being restructured before the technology has proved it can do the work.
What 14,000 calls showed
Bold Agent Kit’s study is the first notable dataset of its kind to come from inside the industry — the company sells AI voice agents, so the result is against its own commercial interest. The analysis measured three outcomes: successful intake conversions, calls abandoned without meaningful progress, and missed calls. Humans won the conversion battle by roughly a factor of four. The researchers’ explanation is not that AI cannot take a message; it is that the calls that turn into business are the messy ones — the caller with a non-standard problem, the half-sentence explanation that a person can finish, the moment where a caller decides whether to trust the business on the other end.
The company’s framing is careful: AI answered more calls and handled volume fine. It just converted fewer of them. For a small business, that distinction is the whole ballgame — a missed conversion costs more than a missed call.
The obvious caveat: the study is small, self-reported, drawn from one platform’s client base, and it is not peer-reviewed. It is also the kind of finding that should be read as a snapshot of AI voice capability in late 2026, not a permanent truth. The same caveat cut the other way in the industry’s favour for two years; the conversion gap is data, not doctrine.
The employment data running the other way
While the performance debate plays out, the hiring effect is already visible. Revelio employment data, cited in Andreessen Horowitz’s 9 October charts newsletter, shows global call-centre employment growth has flipped negative after more than a decade of consistent expansion — and, strikingly, lower-middle-income countries that stayed positive through most of 2024 have now followed, with growth negative everywhere as of the final quarter of 2025. Outside China, call-centre employment has lagged broader white-collar service work in every major non-US market.
This confirms and extends the Goldman Sachs finding from September: call-centre employment sat 39 percent below trend in the US in that report, 33 percent below in Canada, with software publishing and consulting close behind. a16z’s reading of the new data is blunt but fair-minded — growth began slowing when interest rates rose, slightly before generative AI reached the market, but the downtrend has steepened and accelerated since. Their conclusion: AI is the most plausible driver, and the mild good news is that US call-centre workers appear to have found work elsewhere, because overall unemployment has stayed relatively stable.
Why the gap matters for workers and employers
Here is the uncomfortable synthesis. Companies are cutting call-centre hiring on the assumption the agents will handle the load. The conversion data suggests that, for the revenue-generating slice of the work, they currently do not. What tends to follow in labour markets is not a reverse of the cut but a re-shaping of the remaining role: fewer people doing pure call handling, more doing what the AI fails at — the complicated conversion calls, the follow-ups, the escalations — with the AI layering the routine traffic underneath.
That is, on paper, a better job than pure switchboard duty. Whether it becomes one depends entirely on how employers staff the transition. The Workday workforce report found jobs being rewritten faster than removed, and this is a textbook case: a voice-agent supervisor who rescues the leads AI drops is doing measurably different work from the receptionist of 2023, with more of the skills — judgement, rapport, closing — that resisted automation in the first place.
There is also a fresh signal here for the many NZ businesses — trades, clinics, hospitality — that adopted AI receptionists during 2025-26 cost pressure. If the four-fold conversion gap holds even approximately, the practical advice is to route rather than replace: let AI take after-hours overflow and routine bookings, and put humans on the calls where a customer is deciding whether to spend money.
The New Zealand read
New Zealand has no separate call-centre employment series that moves like the global one, but the exposure sits mostly in Auckland-based corporate contact centres, and the hiring signal is visible in the SEEK data — high-automation-exposure job ads (the band that includes telemarketers and keyboard operators) fell 2.6 percent year on year in September while overall ads grew 7.8 percent. The country’s labour market remains tight enough, at 5.6 percent unemployment, that displaced contact-centre workers have somewhere to go — hospitality, tourism and the South Island regions are all hiring hard for summer, per SEEK’s September report. The conversion-gap finding is genuinely relevant to NZ’s small-business economy, where the owner often is the receptionist after hours, and where the “AI answered it but nobody booked anything” experience is quietly common.
FAQ
Did AI receptionists really lose to humans in a head-to-head study? On conversion, yes: an analysis of about 14,000 real inbound calls by Bold Agent Kit found humans achieved roughly four times stronger intake conversion performance, though AI handled more calls and more volume. The study is a single-platform snapshot, not peer-reviewed.
Are call-centre jobs declining? Global call-centre employment growth has turned negative across income groups, according to Revelio data cited by Andreessen Horowitz, including in lower-middle-income countries as of late 2025. Goldman Sachs separately put US call-centre employment 39 percent below trend in September.
If AI underperforms, why are companies still cutting call-centre jobs? Hiring decisions are being made on cost models and the assumption capability will catch up. The jobs are being restructured — fewer pure call handlers, more escalated-conversion specialists — before the technology has fully earned the cut.
What should small businesses do with AI phone agents? The data suggests routing, not replacement: let AI handle after-hours and routine bookings, and keep humans on the calls where customers are deciding whether to buy — that is where the four-fold gap shows up.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
