The Berlin company teaching machines to answer the phone - and convincing the Global 2000 to let them.
PARLOA, Berlin / Munich / New York. The logo is two overlapping lenses, refined from hundreds of drafts to mean "mutual understanding." The product spends its days on hold music's grave.
Somewhere right now, a customer at Allianz, Booking.com, or Swiss Life calls a help line. A voice picks up. It is calm, fluent, and quietly competent. It routes the call correctly 97 times out of 100. It switches languages mid-sentence. It does not sigh, does not transfer you to a department that no longer exists, does not put you on hold while it "checks." It is not a person. It is Parloa.
For most of the last decade, "talking to a robot" was a punishment customers endured, not a service they enjoyed. Parloa's bet is that this was never a law of nature - just a limit of the technology. Founded in 2018 in Berlin and now valued at $3 billion, the company sells one thing: AI agents good enough that you stop wishing for a human. The market, judging by the $560M-plus investors have poured in, finds the pitch persuasive.
Here is the uncomfortable arithmetic of the contact center. Calls are expensive. Humans are expensive. So for twenty years, the dominant strategy was to make calling slightly miserable - menus, hold queues, scripts - in the hope you would give up and email instead. The "press 1 for billing" maze was not a bug. It was a budget decision.
The first generation of chatbots promised relief and mostly delivered frustration. They handled the easy questions and collapsed at the first surprise. They could read a script but could not hold a conversation. And voice - the channel people actually reach for when something has gone wrong - stayed almost untouched, because spoken language is messy, interruptible, and unforgiving of error.
Parloa's founders had spent years close enough to this problem to take it personally. Their earlier venture, Future of Voice, built voice experiences for Red Bull, Vodafone, and Deutsche Bahn. They watched enterprises want conversational service and lack any reliable way to build it. The gap between the demo that wowed a boardroom and the system that survived a million real calls was enormous - and nobody had crossed it.
Malte Kosub is the kind of founder who sold his first company - an e-commerce startup called Wandnotiz - in 2017 and immediately went looking for a harder problem. He found it in language. His co-founder, Stefan Ostwald, brought the engineering depth in voice and machine learning. In 2016 they started a consultancy. By 2018 they had concluded that consulting on the problem was not the same as solving it, and turned Parloa into a product company.
The bet was specific and contrarian: that the hardest channel - voice - was the one worth winning, and that enterprises would pay for reliability over novelty. Not a clever demo, but an agent that performs on call number 999,999 as well as on call number one. That meant building not just the AI, but the unglamorous machinery around it: testing, monitoring, optimization, compliance. The boring parts. The parts that make a Fortune 500 risk team say yes.
It took years before the world's appetite caught up to the bet. When large language models arrived, Parloa was not scrambling to start - it was already holding the infrastructure that turned a flashy capability into a dependable enterprise system. Timing favors the prepared, and occasionally even rewards them.
Note the gap and then the sprint: four quiet building years, then a valuation curve that needed a bigger chart. Patience, followed by a sudden lack of it.
Parloa's core product is the AI Agent Management Platform (AMP) - a name as understated as a fire extinguisher, and roughly as useful in a crisis. It lets a large enterprise design, test, deploy, monitor, and improve AI agents that handle real customer conversations across voice, chat, WhatsApp, and Microsoft Teams.
Purpose-built for the hardest channel - voice - where customers interrupt, mumble, and change their minds. Cited deployments hit 97% routing accuracy and 60% faster resolution.
A visual builder lets teams design and adjust conversational flows, with real-time translation across 140+ languages and integrations into Salesforce, Microsoft, and Genesys.
Real-time guidance surfaced to human agents mid-call - because the goal was never to remove people, just the parts of the job that made them want to quit.
Crucially, it runs on Microsoft Azure and ships with the compliance acronyms enterprises require - GDPR, SOC 2, HIPAA. Boring? Completely. But boring is what lets a bank hand its phone lines to a machine and still sleep at night.
A pitch is a story. A customer logo is a fact. Parloa has collected an uncomfortable number of facts for any competitor to ignore: Global 2000 brands across finance, insurance, travel, retail, and telecom now run customer conversations through its agents.
A guest list of companies that do not, as a rule, gamble their phone lines on a startup. They did the math first.
Bar height tracks round size. The valuation note underneath is where it gets genuinely strange: $1B to $3B in eight months.
Microsoft (Azure hosting + Teams), PwC and KPMG (enterprise rollout), and CCaaS/CRM integrations with Genesys and Salesforce. Translation: Parloa plugs into the systems enterprises already can't live without.
Strip away the funding headlines and the ambition is almost old-fashioned: make a company feel like it actually knows you. Parloa frames its mission as closing the "company-customer relationship gap" - turning each interaction into something personal, and a little preemptive, that gets better the longer you stay.
In 2025 the company published the "Parloa Promise," a public commitment to three things: agent reliability, relentless innovation, and human-centric responsible AI. It is the kind of statement that is easy to write and hard to keep - which is presumably why they wrote it down where customers can hold them to it. Kosub has taken the responsible-AI argument to venues like the World Economic Forum, positioning Parloa as both operator and policy voice.
Whether that is visionary or merely well-marketed depends on execution - and execution, in this market, is measured one phone call at a time.
Competitors have noticed. Sierra, Decagon, Cresta, PolyAI, and Cognigy are all chasing versions of the same future, alongside the cloud giants' own bots. The contest will not be won on demos. It will be won on the millionth call - the one that goes wrong in a way nobody scripted - and on which agent handles it without a human having to apologize afterward.
If Parloa is right, the contact center stops being a cost to minimize and becomes a relationship to deepen. If it is wrong, it will be a very well-funded footnote. The investors have placed their bet. The enterprises have placed theirs. The rest of us will vote, one call at a time, on whether the voice on the line has finally earned its keep.