The AI Agents Racing to Answer Your Support Calls

Call a bank, an airline, or a cable company today and listen to the first three seconds. The pause before a voice arrives is shrinking, the voice is smoother than it used to be, and more and more often it is not attached to a human being. That gap - the moment between your call connecting and someone saying hello - has become one of the most fought-over pieces of real estate in enterprise software. Eight companies in particular are spending hundreds of millions of dollars to own it.

They do not look like a group. One bootstrapped itself from a Croatian village. One spun out of a Cambridge University lab. One is run from Bozeman, Montana. Two of them have already been swallowed by bigger companies. What they share is a bet: that the call, the chat, and the support ticket are about to be handled by AI agents that resolve problems rather than deflect them, and that whoever builds the best version gets paid for every conversation a company no longer has to staff.

The money says this is serious

You can argue about whether the technology is ready. The capital already voted. In December 2025, PolyAI raised $86 million at a $750 million valuation, up 50 percent from the $500 million it commanded a year and a half earlier. That round brought its total to roughly $200 million, from backers including Khosla Ventures, Hedosophia, Georgian, and Nvidia's venture arm. Cresta, which coaches contact-center agents in real time, has raised more than $270 million and touched a $1.6 billion valuation. Even the open-source option in the field, Botpress, closed a Series B at a $120 million valuation - more than double its previous mark.

The reason the numbers look like this is boring and unavoidable: answering the phone is expensive. A large enterprise pays for seats, training, attrition, night shifts, and holiday coverage. An AI agent does not sleep, does not quit, and gets cheaper per conversation as it scales. If a vendor can prove it resolves even a chunk of routine calls without a human, the return on investment writes itself. That is the pitch every one of these eight companies makes, in slightly different accents.

Two exits, and not much middle

The clearest signal about where this market is heading is not a funding round. It is the acquisitions. In August 2024, SoundHound bought Amelia for about $80 million. Amelia had raised more than $189 million over a long life that began in 1998 as IPsoft, one of the original "digital employee" companies. The buyer paid less than half of what investors had put in. A year later, Zendesk agreed to acquire Forethought, an agentic customer-experience platform that had raised north of $100 million and once counted Ashton Kutcher and Diddy among its early backers.

Read those two deals together and a pattern appears. In AI customer service, the sustainable positions are at the extremes. You either get big enough to be a platform in your own right, or you get absorbed into one that customers already use. The uncomfortable middle - a strong product with real revenue but no distribution of its own - is where companies get bought at a discount. Forethought's own pitch, that it "sits on top of Zendesk and Salesforce," told you how the story would end. If your entire value is a layer on someone else's stack, the someone else can decide to own the layer.

Voice is the expensive frontier

Text chat is close to a commodity now. Any competent team can wire a language model to a help center and answer typed questions. Voice is harder, and that is exactly why the money is chasing it. A natural phone call has to handle interruptions, accents, background noise, and the specific human impatience of someone who dialed because typing did not work. Do it badly and you get the robot everyone hangs up on. Do it well and callers do not always notice they were never talking to a person.

PolyAI is the loudest bet here. It spun out of Cambridge University's Machine Intelligence Lab in 2017, and its chief executive, Nikola Mrksic, was the first engineer at VocalIQ, a voice startup Apple acquired. The company sells enterprise voice agents that answer and resolve calls, and its valuation reflects investor conviction that natural-sounding voice is the defensible part of this business. Quiq, built in Bozeman by Mike Myer and William O'Neill, comes at the same problem from messaging, arguing that enterprises want one connected agent across voice, text, and chat rather than a separate bot per channel.

The coach in your ear

Not every company in this race is trying to remove the human. Cresta's most interesting product is not a bot at all - it is the software whispering to the human agent in real time, suggesting the next sentence, flagging a compliance risk, surfacing the answer before the caller finishes the question. NICE, the publicly traded contact-center giant out of Ra'anana, Israel, plays the same double game with its Enlighten AI platform, which both automates simple interactions and augments the people handling the hard ones. NICE's cloud roots trace back to its $940 million purchase of inContact in 2016, a reminder that this "new" market has an incumbent layer with real scale.

The augment-versus-replace split is the quiet philosophical fault line of the whole sector. Replace-first vendors sell a cost-cutting story to the CFO. Augment-first vendors sell a quality-and-retention story to the head of customer experience. Most buyers, sensibly, want both, which is why nearly every company here now claims to keep "humans in the loop" while also promising to shrink the number of humans you need in that loop.

Build it yourself

For teams that do not want to hand the keys to a black box, Botpress offers the contrarian route. Founded in Quebec in 2017 by Sylvain Perron and Justin Watson, it is MIT-licensed and open source, which means a developer can read the code, self-host the whole thing, and build a support agent without paying a per-conversation toll to anyone. In a market where most vendors sell an opaque service, "you can inspect and own it" is a genuine differentiator, and it explains why Botpress raised on a developer-tools story rather than a pure customer-service one.

Infobip sits at the infrastructure end too, though it arrived from a completely different direction. It began in 2006 in the Croatian town of Vodnjan, bootstrapped with a small family loan, and grew into a messaging giant connected to more than 800 mobile operators - all without outside funding until it became Croatia's first unicorn in 2020. Its conversational-AI product, packaged under a layer it calls AgentOS, is less a standalone bot than a way to add AI to the messaging pipes it already runs for banks, retailers, and airlines. When you already move billions of messages, adding an agent that answers them is a natural next move.

What it means when you dial

Strip away the funding rounds and the acquisition drama, and the shift underneath is simple to state. The old chatbot was measured by deflection - how many people it kept away from a human. The new AI agent is measured by resolution - how many problems it actually solves without one. That is a higher bar, and it is the reason the good ones now feel useful instead of infuriating. It is also why the bad ones will be exposed faster: a scripted tree could hide behind "let me connect you," but an agent that promises to resolve your issue has nowhere to hide when it cannot.

For customers, the practical effect over the next year is a support experience that gets suspiciously better in places and stubbornly worse in others, depending entirely on which of these companies your bank or airline chose. For the industry, the direction is consolidation. NICE and Zendesk-scale platforms will keep absorbing the specialists, the open-source and infrastructure players will arm everyone else, and the pure voice bet - PolyAI's - will either become the standard for phone support or the cautionary tale about how hard the phone really is. The next time you call and a calm voice picks up on the first ring, it is worth remembering how much money is riding on whether you can tell.