What is Omilia's self-learning CX platform and how is it different from other voice AI tools?

Omilia's Cloud Platform (OCP) is a vertically integrated voice AI stack that owns every layer from acoustic models to agentic orchestration. Unlike platforms assembled from third-party ASR, LLM, and TTS APIs, OCP runs proprietary models purpose-built for enterprise voice. Its self-learning pipeline ingests every live conversation to discover new resolution patterns and improve automation performance without modifying the live runtime. The result is compounding containment improvement over time, which most orchestration wrappers on frontier LLMs cannot replicate because they lack the closed-loop training infrastructure.

What does zero token cost passthrough mean and why does it matter for enterprise buyers?

Most enterprise voice AI platforms today are thin orchestration layers on top of frontier LLMs like GPT or Claude. Every customer call generates tokens billed at the underlying API rate, so call volume translates directly into unpredictable infrastructure cost. Omilia runs its own proprietary models, so there is no external token bill. The platform prices per resolved call, giving finance teams a fixed, auditable cost that scales with business outcomes rather than with model pricing decisions made by a third-party lab. For high-volume operations running hundreds of thousands of calls per day, this is the difference between a predictable budget line and an open-ended exposure.

Which regulated industries does Omilia serve and what compliance certifications does it hold?

Omilia serves banking, insurance, healthcare, and government agencies. Its platform simultaneously holds FedRAMP authorization, PCI DSS Level 1, SOC 2 Type II, ISO 27001, HIPAA, and GDPR compliance. Customers include Capital One, Discover, RBC, First Financial Bank, PSEG, Purolator, and the UK Department for Work and Pensions. The walled garden architecture, where no customer utterance data leaves Omilia's certified environment, is what makes simultaneous PCI Level 1 and FedRAMP authorization possible. Both mandates require data sovereignty that external API dependencies break.

What does Omilia's $67M funding round signal for the enterprise voice AI market?

It confirms that proprietary, domain-specialized voice AI is pulling away from general-purpose LLM orchestration for regulated enterprise use. The 10x ARR growth to $60 million without additional equity capital since the Series A is the clearest signal: Omilia sold into the most demanding enterprise buyers in the market, banking and insurance, and scaled without needing to raise more money to fund growth. Expedition Growth Capital's decision to lead a $67M round now, with ARR already at this level, is a bet that the contact center transformation cycle is still early and that domain-specialized, compliance-first platforms have structural advantages over general-purpose alternatives.

How does Omilia's self-learning pipeline work technically?

The self-learning pipeline runs entirely offline, separate from the live interaction path. After each call, OCP ingests the conversation, identifies resolution patterns that the live system did not handle optimally, and auto-generates updated Playbooks for the agentic Task Agents. Operators review and approve changes before they go live. The pipeline also uses the call corpus to train and evaluate Omilia's proprietary NLU and ASR models continuously, compressing the annotation cycles that normally take months into days. Omilia says the AI Bootstrapping engine can ingest existing call recordings, transcripts, standard operating procedures, and API specifications to generate production-ready Task Agents from scratch.

Omilia Raises $67M for Self-Learning Enterprise Voice AI

Omilia announced a $67 million Series B round today, August 6, 2026, led by Expedition Growth Capital. The company’s agentic voice AI platform now handles more than one million calls per day for a single Tier 1 US bank and more than 600,000 calls per day for a multinational insurance company. Annual recurring revenue has grown more than 10x to over $60 million since its Series A, without additional equity capital in between.

That combination, enterprise-scale deployments in the most demanding regulated industries, compounding ARR growth without further dilution, and a Forrester Wave Leader position in Q2 2026, is a precise description of product-market fit in a market that has been promised but not quite delivered for a decade.

What Omilia Actually Ships

The Omilia Cloud Platform (OCP) is not a wrapper on GPT or Claude. It is a vertically integrated stack that owns every layer: proprietary automatic speech recognition, natural language understanding, agentic orchestration, and Lexis, Omilia’s own generative text-to-speech model launched earlier this year. No customer utterance leaves Omilia’s certified infrastructure. No external API token bill accrues on the back of every call.

The platform runs two core execution modes that can be blended within the same deployment. Deterministic miniApps handle intents where guaranteed, scripted behavior is required, such as payment authorization flows or regulatory disclosures. Autonomous Task Agents handle complex, multi-step interactions end to end, processing intent shifts, tool failures, and edge cases in real time without human involvement. The orchestration engine dispatches tool calls in parallel, keeping total round-trip time within sub-300ms latency budgets even when querying multiple backend systems simultaneously.

That architecture matters for enterprises in regulated industries. PCI DSS Level 1 and FedRAMP authorization both require data sovereignty that external API dependencies break. Omilia is simultaneously certified under FedRAMP, PCI DSS Level 1, SOC 2 Type II, ISO 27001, HIPAA, and GDPR. The walled-garden architecture is the mechanism that makes it possible to hold all of those certifications at once.

The Self-Learning Differentiator

The feature that separates Omilia from the generation of agentic voice platforms built in 2025 and 2026 is its offline self-learning pipeline.

Most contact center AI platforms today are built on a fairly static architecture. The operator configures flows, deploys them, and monitors performance. When something breaks or degrades, a developer intervenes to update the configuration. The feedback loop is slow, manual, and capacity-constrained.

Omilia’s pipeline is different. After each live call, OCP ingests the conversation into an offline learning engine that runs separately from the production runtime. The engine discovers new resolution patterns, identifies flows that are underperforming, auto-generates updated Playbooks for the Task Agents, and surfaces them for operator review and approval before going live. The system also continuously trains and evaluates Omilia’s proprietary NLU and ASR models on the live call corpus, using what Omilia calls AI Bootstrapping to compress annotation cycles from months to days.

The result is compounding containment improvement. Most voice AI deployments plateau: they automate a fixed percentage of interactions, and human agents handle the rest. Omilia’s self-learning architecture is designed to continuously expand that containment rate as the model learns from each additional call. For high-volume operations, that improvement compounds quickly.

Scale Numbers That Matter

The scale figures in today’s announcement are worth examining directly.

Deployment	Daily Volume	Notes
Tier 1 US bank	1,000,000+ calls/day	Single client
Multinational insurer	600,000+ calls/day	Agent-assist + automation
Concurrent sessions	50,000+	Single client, sub-second latency
Taco Bell	1,000+ drive-thrus	38 US states
Total deployments	200+	Enterprise production

Fifty thousand concurrent voice interactions for a single client, at sub-second latency, without degradation is an infrastructure claim that few platforms in this category can match. Most general-purpose voice AI stacks, built on external API chains, introduce compounding latency at each handoff: ASR to LLM to TTS, with network round-trips at each step. Omilia eliminates those handoffs by running the full stack inside its own infrastructure.

Pricing Model and the Token Cost Question

One of the most commercially significant aspects of Omilia’s platform is its pricing structure: zero token cost passthrough.

Most enterprise AI platforms today carry a hidden variable cost. Every customer interaction generates tokens billed at the rate of the underlying LLM provider. For a contact center running 600,000 calls per day, that cost is not trivial. It also cannot be precisely forecast, because it depends on call length, model version, and pricing decisions made by a third-party lab that the enterprise cannot control.

Omilia prices per resolved call. The unit of cost aligns with the unit of business outcome. Finance teams can model costs against call volume forecasts rather than token throughput estimates. The per-resolved-call model also creates alignment between Omilia and its customers: Omilia earns more when it contains more calls autonomously.

This structural difference is becoming a strategic consideration for enterprise AI buyers as broader market consolidation increases token prices. Platforms that own their inference stack are insulated from those fluctuations. Platforms that do not are exposed to them.

The GTM Build-Out

Expedition Growth Capital’s $67 million will fund two primary priorities: US market expansion and commercial organization scale.

Omilia is opening its first US office in the second half of 2026. The company has operated primarily from its Athens and Cyprus bases and expanded through a global channel and partner network. A US office changes the enterprise sales motion, allowing direct engagement with the decision-makers at Tier 1 banks and health insurance carriers who represent Omilia’s primary buyer profile.

The commercial leadership team assembled over the past several months reflects that ambition. Nick Delis as Chief Revenue Officer, Ryan Kam as Chief Marketing Officer, Armando Trivellato as EVP Latin America and Iberia, and Dave Ogden as VP Revenue Operations all previously worked at Five9, where they collectively scaled ARR from $100 million to over $1 billion. That is a specific pattern: people who know how to take a specialized enterprise software company through the $100M to $1B ARR inflection point.

Oliver Thomas, Founder and Managing Partner at Expedition Growth Capital, framed the investment thesis around structural moat: “Omilia’s Self-Learning Agentic CX is delivering a step change in call containment and operational improvement, with glass-box auditability and cost predictability that is structurally difficult for other vendors to compete with.”

What This Signals for Enterprise AI Buyers

For enterprises evaluating agentic voice AI, Omilia’s funding round is relevant context for three reasons.

First, it confirms that domain-specialized, compliance-first platforms are winning the regulated enterprise market. General-purpose LLM orchestration is not sufficient for Tier 1 banking or government mandates. The compliance stack (FedRAMP, PCI Level 1, HIPAA, GDPR simultaneously) is itself a multi-year investment that creates durable competitive separation.

Second, the pricing model question is becoming a strategic decision, not just a procurement detail. Enterprises that sign long-term contracts with token-cost-passthrough platforms are taking on pricing risk they cannot control. The per-resolved-call model that Omilia uses is worth examining as a structural alternative.

Third, the self-learning pipeline represents a meaningful differentiation that compounds over time. Platforms that improve automatically from production traffic are structurally different from platforms that require manual retraining cycles. In high-volume contact center environments, the compounding improvement rates translate directly into containment economics.

For teams building agentic AI strategies in regulated industries, the Omilia story is a useful data point: the platforms that own their full stack, hold the compliance certifications, and align pricing with outcomes are the ones that enterprise buyers in banking, insurance, and healthcare are choosing at scale. See also our coverage of PolyAI’s Dialog-RSN-1, another enterprise voice AI architecture story from earlier this year, for a broader picture of how audio-native models are reshaping the contact center market.

Omilia’s platform is available at omilia.com. For enterprises exploring how agentic AI applies to your GTM and customer operations, see how Enera approaches AI-native transformation or explore autonomous GTM systems.

Sources: Omilia official press release via Business Wire, CityBiz, Omilia product page, FinSMEs