Omilia raises $67M: voice AI containment is the metric

Every voice AI demo sounds good. The vendor picks the script, the network, and the accent. The number that decides whether the thing pays for itself is duller: what share of calls end without a human, and can you audit why. Omilia just raised $67 million on that number, and it took the company 24 years to get here.

What actually happened

Per BusinessWire and TechCrunch:

• $67M Series B led by Expedition Growth Capital, announced August 6, 2026. The prior round was a $20M Series A from Grafton Capital in 2020.
• Omilia was founded in 2002 and is based in Athens. Live ARR has grown more than 10x since 2020, to over $60 million — with no equity raise in between.
• Named customers include Capital One, Discover, RBC, DWP, PSEG, and Taco Bell, which TechCrunch reports is deployed across 1,000+ locations.
• Headcount is about 500, expected to reach 600 by year-end. The money funds a first U.S. office, a go-to-market build-out, and CRO, CMO, and VP RevOps hires.
• Expedition's Oliver Thomas cited call containment, glass-box auditability, and cost predictability as the differentiators — not model quality.

Why call containment matters for your business

Containment is the only line that touches your P&L. If 100 calls come in and 62 resolve without a person, you saved 62 handle-times. If those 62 include 15 that resolved wrong and generate a callback, you saved 47 and bought yourself a complaint queue. Ask any vendor for containment rate and callback rate on the same cohort. One number without the other is marketing.

"Glass-box" is a procurement requirement, not a nice-to-have. When a customer disputes what your system told them, you need the transcript, the intent it matched, the policy it applied, and the version of that policy on that date. If the vendor can only show you a confidence score, you cannot defend the call. This is why regulated buyers pay a premium — and why you should copy their checklist even if nobody regulates you.

Twenty-four years is the story. Omilia is not an AI-native startup; it is a speech company that survived long enough for the models to catch up to its distribution. That is the pattern to watch in your own stack: the vendor with boring uptime, an auditable log, and predictable per-minute pricing usually beats the one with the better demo by month nine.

Sources: BusinessWire, TechCrunch.