Omilia Built a Bot People Actually Say Thank You To
The Cyprus garage startup bootstrapped for 18 years, countersued the giant that tried to crush it, and just raised $67 million to finally kill the phone tree.
Here is a small, strange fact about Omilia's customer-service bots: people thank them. Not sarcastically. Callers finish an automated conversation about a lost card or a delayed flight and say "thank you" before they hang up, the way you would to a person. Dimitris Vassos, who founded the company in 2002, treats that reflex as his real benchmark. Accuracy scores and latency graphs are for the board deck. The thank-you is the proof.
On August 6, 2026, that quiet obsession got an expensive vote of confidence. Omilia raised a $67 million Series B led by Expedition Growth Capital, a transatlantic software investor, to fund an aggressive push into North America. The number that should make competitors uneasy is not the raise. It is the one underneath it: live annual recurring revenue past $60 million, grown more than tenfold since the last round, with no fresh equity in between. Omilia spent the interim doing the least fashionable thing an AI company can do right now. It made money.
OriginsA garage in Athens, a grudge, and an engine
Vassos did not arrive at conversational AI through a research lab. He spent his early career at IBM UK, working on voice products that shipped into more than 70 countries. When he moved back to Athens in 2002, he started Omilia - Greek for "speech" - out of a garage, with a goal that sounded almost petty at the time: destroy the IVR, the "press 1 for billing" phone menu that everyone hates and nobody had bothered to fix.
For its first years, Omilia was a contact-center integrator stitching together other people's software. The turning point was a betrayal. The speech-recognition vendor Omilia depended on tried to poach its clients directly. Faced with losing the business it had built, the team made the harder choice and wrote its own speech engine from scratch - and, as Vassos tells it, a better one. The dependency that nearly killed the company became the thing the company sold.
"It works so well that it also compels people to say thank you at the end for the service they received."Dimitris Vassos, Founder & CEO
That is the pattern worth stealing from Omilia's early story. A constraint arrived - no money, then no supplier - and each time the response was to build the missing piece rather than route around it. Slow, unglamorous, and in hindsight the entire moat. By the time the current wave of AI hype crested, Omilia already owned its stack top to bottom.
The long middleEighteen years without a term sheet
Omilia took no outside funding until 2020, when Grafton Capital put in $20 million - the company's first-ever round, after eighteen years of bootstrapping. In the same stretch it found itself tangled with Nuance, the speech-tech heavyweight, and countersued over antitrust claims rather than settle quietly. A small Cyprus-registered company picking a legal fight with an industry giant is not the behavior of a startup hedging its bets.
The bootstrapped years look, in retrospect, like a deliberate filter. Companies that raise early are pushed to grow before they are ready and to demo before they can deliver. Omilia had the opposite pressure. Every feature had to be paid for by a customer who was already using the product in a live contact center, at scale, with real callers who would notice if it broke. There is no faster way to learn what actually matters than shipping to a bank's phone line on a Monday morning.
The productWhat "self-learning agentic" actually means
In February 2026 Omilia launched what it calls the first enterprise-grade self-learning agentic CX platform. Strip the marketing and the idea is concrete. Instead of a single chatbot answering questions, you get a workforce of AI agents that handle a customer end to end across voice, chat, IVR and messaging - checking a balance, rebooking a flight, verifying identity - and hand off to a human only when they should.
The word doing the heavy lifting is "self-learning." The agents improve from real outcomes: whether an issue was resolved, what a human agent corrected, what a customer said afterward. Every conversation, machine-handled or human-handled, becomes training signal. Omilia also insists the system is a "glass box" rather than a black box - the enterprise can see why an agent did what it did. In a market nervous about AI it cannot audit, transparency is quietly becoming the feature that closes deals.
The moat in agentic AI is not the model. It is the feedback loop - the thing that turns a billion messy conversations into a system that gets better while you sleep.
The scale claims are the kind that only make sense after two decades of infrastructure work. Omilia says the platform can hold more than 50,000 concurrent voice interactions for a single customer while keeping responses under a second. It also promises "zero-day go-live" - a bold thing to say about enterprise software, and only credible because the boring plumbing was built long before the demo.
| Step | What happens |
|---|---|
| 1. Converse | AI agents handle voice and digital contacts end to end, across channels. |
| 2. Observe | Every interaction is scored on resolution, escalation and customer feedback. |
| 3. Learn | Real outcomes and human-agent corrections feed back as training signal. |
| 4. Improve | Agents get more accurate, faster and more personalized over time - autonomously. |
The fieldMicrosoft, NICE, PolyAI - and a bootstrapper
Omilia is not alone in wanting to automate the contact center. Microsoft folds conversational AI into its Dynamics and Azure stack. NICE owns the CXone platform that many enterprises run their support on - and, tellingly, Omilia integrates with it rather than fighting it. PolyAI, the well-funded London startup, chases the same enterprises with lifelike voice agents. It is a crowded, serious market with deep-pocketed players.
Omilia's edge is not spend. It is time in the trenches. Nearly a billion interactions a year, deep integrations into systems like NICE CXone and Microsoft Dynamics 365, and a decade-plus of running production voice at scale add up to a kind of institutional muscle memory that money cannot buy quickly. The Gartner analysts noticed: Leader in the 2022 Magic Quadrant, Customers' Choice in 2024.
| Player | Angle | Where Omilia differs |
|---|---|---|
| Microsoft | AI inside Dynamics / Azure | Omilia integrates with it, stays CX-native |
| NICE | CXone contact-center platform | Omilia runs on top of CXone, not against it |
| PolyAI | VC-funded lifelike voice agents | Omilia: 1B+ live interactions, 24 yrs in market |
What's nextAthens, New York, and a phone tree's funeral
The Series B money has a clear destination. Omilia, founded in Athens and registered in Cyprus, is opening its first U.S. office in the second half of 2026, with the company now describing itself as Athens- and New York-based. Vassos says demand for the self-learning platform is accelerating fastest in the United States. After twenty-four years of building somewhere the tech press rarely looks, Omilia is planting a flag in the market it has to win.
There is a lesson buried in the two-decade gap between the garage and the New York office, and it runs against almost everything the current AI cycle rewards. Raise slowly. Own your stack. Build the boring infrastructure. Measure yourself by whether a ninety-year-old can use the thing without help - and whether, at the end, they bother to say thank you. Omilia spent eighteen unglamorous years on that, and the $67 million is the market finally agreeing it was time well spent.