Site icon CPaaSAA

The Voice Was Never the Product

Rob Kurver

Vocca’s $20 million Series A is a funding story only on the surface. Announced on 6 October and led by Norrsken VC, the round follows a year in which the Paris- and New York-based company says it grew from 2,000 to 15,000 practitioners. It now reports more than 1,500 practices, over 20 specialties and more than a million patient interactions a month, 70% of them handled end to end. By its own count it has raised $25 million in total.

Most coverage will file Vocca under Voice AI. We think that label is the wrong unit of analysis, and the more useful question for our partners is what Vocca actually sells.

Vocca sells a completed appointment, not a voice

The Vocca agent answers the call, applies the scheduling rules of the specialty and writes the appointment into the practice’s medical software. It keeps US and EU data in separate production environments, operates as a HIPAA Business Associate and runs on infrastructure with SOC 2, ISO 27001 and HDS certifications. It is already inside hospital systems: Hôpital Fondation de Rothschild uses it for intake, calls and confirmations across its 350,000 yearly patients.

The company states its thesis plainly: the conversation is the interface, and the scheduling engine behind it is the infrastructure. Its first product was a phone assistant, and it says the new funding goes to channels beyond voice, more specialties and a deeper US presence.

Vocca does not say publicly which speech and language models run underneath, and that silence is the point. Take the voice away and most of the value is still there: specialty knowledge, scheduling logic, system integration, secure handling of patient data, compliance and the booked slot. Take those away and what remains is an input that gets cheaper and better every quarter for everyone.

Investors are funding three different layers at once

The last three months of voice funding show capital going to the speech layer, the trust layer and the outcome layer in parallel. Sifted data cited by Dealroom puts European voice-focused startups at €536 million raised in the first half of 2026, up from €360 million a year earlier.

AnnouncedCompanyLayerRound
6 Oct 2026VoccaOutcome: healthcare scheduling$20M Series A, led by Norrsken VC
30 Sep 2026ElevenLabsSpeech models$300M employee share sale at a $22B valuation
28 Sep 2026ModulateTrust: emotion, intent, deepfake and compliance analysis of calls$25M, led by Future Ventures
16 Sep 2026TrebleQuality: voice simulation and testing$18M Series A extension, led by Paladin Capital Group
31 Jul 2026Smallest.aiSpeech models$13M Series A, led by Seligman Ventures
28 Jul 2026Fish AudioSpeech models$52M seed, led by Coreline Ventures and Capital Today
8 Jul 2026GradiumSpeech models$30M seed extension with Nvidia, taking the seed past $100M

ElevenLabs shows how fast the speech layer is being repriced. It raised $500 million at an $11 billion valuation in February and, on 30 September, let employees sell $300 million of shares at $22 billion, a doubling in under eight months. The second deal was a share sale co-led by Wellington and T. Rowe Price rather than new money for the company, but it is a clear signal of what investors think the leading voice model builder is worth.

Gradium’s co-founder framed the structure well: thousands of startups and enterprises now build on voice models, while fewer than a dozen players can train them at scale. A concentrated model layer feeding a very wide application layer is exactly where workflow ownership becomes the differentiator. Vocca is the clearest example so far, with 15,000 practitioners in production.

Modulate is the one we would watch most closely for our Trust pillar. Deepfake and scam detection, and compliance checks on AI agents, are the problems that appear once voice agents reach regulated industries. Modulate typically sits beside the voice stack rather than inside it, which makes it a natural complement to the infrastructure our partners already run.

Operators are placing seven different bets on the workflow layer

The operator side of this market is answering the same question in different ways, and all seven surfaced within a few weeks of CASA26, our September event on Intelligence, Trust and Transformation. TelecomTV recorded interviews there and is now publishing them, which is where the KPN and BT views below come from.

PlatformWhere it sitsHow workflows get inStatus
Radisys V.AI (launched 10 Sep 2026)Media and application layer on the Engage Digital Platform, inside operator networksOperators deploy turnkey applications or build their own in V.AI Studio, using partner and cloud speech modelsProduction deployments reported in four regions; named operators include Deutsche Telekom (network integration of its Magenta AI Call Assistant) and Rakuten Mobile
Alianza Crux (launched 9 Sep 2026)Orchestration and intelligence control plane for operator voice networks, on Alianza and third-party infrastructureExperience providers integrate once through connectors and APIs; a business switches an experience on for the number it already hasControlled availability; broader launch planned for the first half of 2027, per Alianza
Core Cognitics IntelinteractOrchestration across voice, WhatsApp, SMS, chat and email, plus a separate edge product for operatorsEnterprises configure AI agents and workflows in a no-code studio, on any AI model, in the cloud or on their own infrastructureHealthcare scheduling and reminders are a listed use case
KPN Mobiel AI (live 1 Oct 2026)Inside the operator’s own business mobile subscriptionKPN develops the product itself and uses ElevenLabs technology for the AI functionsLive for KPN business mobile customers; AI assistant and smart menu planned
Enreach Smart Mobile (launched 24 Sep 2026)Conversational AI in the mobile core, inside the operator’s business mobile plansEnreach sells it directly; no technology partner namedAvailable in the Netherlands for partners and businesses; selected European markets to follow
Mavenir with SanasVoice AI in the mobile core, in the call pathSanas capabilities such as accent and language translation delivered through the operator’s networkPartnership announced; deployments not described
BT AI Receptionist (launched 1 Oct 2026)Sold by BT Business as a service that works across any UK fixed or mobile network, with the customer keeping their numberBT sells it directly through a self-service portal; no technology partner is namedLaunched as the first product in AI from BT

Radisys, one of our Accelerate partners, built V.AI on the Engage Digital Platform it introduced at MWC Barcelona in 2022. The five charter members are ElevenLabs, Hiya, ReadSpeaker, Sensory and Speechmatics (one of our Connect partners), covering speech recognition and synthesis, on-device AI and call identity. The platform also connects to Azure Speech, Google Cloud speech services and the OpenAI Audio API, and it supports GSMA CAMARA network APIs, so applications can draw on network identity, location and quality of service. Radisys is also the integrator Deutsche Telekom credits with the network-level integration of its Magenta AI Call Assistant, built with ElevenLabs and unveiled at MWC in March, and Rakuten Mobile has embedded the platform directly in its core network. It is a strong answer to the question of how an operator gets good voice models into its network. None of the five charter members, however, is a vertical workflow owner of Vocca’s kind, and that is the gap this piece is about.

Alianza, one of our Connect partners, takes the opposite route. Crux does not ship the applications; it lets other companies attach theirs. A business keeps its number and its operator, and turns on an experience such as an AI receptionist without forwarding or porting. Revmo AI, SoundHound and MiaRec are among the early experience partners. Alianza frames the harder question as how many experiences can join the same conversation, share context and act together, and Crux is built to orchestrate exactly that.

Core Cognitics, one of our Innovate partners, presented at CASA26 and sits closest to the Vocca pattern. Intelinteract coordinates AI agents, people, workflows and enterprise systems across channels rather than selling a voice. It spans more channels and more industries than Vocca, which has gone deep on healthcare scheduling, and it also offers an edge inference product aimed at operators.

KPN, another of our Accelerate partners, launched Mobiel AI on 1 October as an add-on to its business mobile subscriptions. KPN developed it and uses ElevenLabs technology for the AI functions, a company it also backs through KPN Ventures. The first version turns each call into a summary with key points, actions and agreed appointments, and converts voicemail to text; an AI assistant and a smart menu follow in the coming months. It is free for two months and then starts at €7.50 a month. Speaking to TelecomTV at CASA26, KPN’s Otto-Jan Woltjer argued that the operator’s edge is the backend: anyone can build an app, but not everyone has the mobile core and the numbers. He added that telling people a call is being transcribed is what earns the trust to win the business.

Enreach, one of our Connect partners, launched Smart Mobile on 24 September, a week before KPN’s Mobiel AI, with conversational AI embedded in the mobile core of its business plans. It screens calls and voicemail, summarises calls and meetings, automates tasks and reminders, and sends the output to email and the CRM, first in the Netherlands and later in selected European markets. Its CTO, Koen van Geffen, positions it as AI benefits without extra systems. Like KPN, Enreach builds and sells it itself. Two Dutch players have now put the same call-summary bundle inside a business mobile plan within a fortnight, which suggests the operator-owned workflow is a pattern, not a one-off.

Mavenir and Sanas make the case for putting voice AI in the mobile core. Announced as a partnership at TelecomTV’s AI-Native Telco Forum in Düsseldorf, the pitch is that AI in the call path keeps data inside the operator’s network and keeps latency low. Mavenir’s Nilesh Parikh says the pair already offer language and accent translation and fraud detection, with in-call automation next, so that a subscriber could complete a task mid-call from the native dialler.

BT, a Grow partner through BT International, saw its BT Business unit launch AI Receptionist on 1 October, the first product in a new AI from BT portfolio. It answers calls around the clock, books appointments into Outlook and Google Calendar, transcribes calls and offers customisable voices and accents. BT says it works across any UK fixed or mobile network, so a business keeps its number and phone system, and that customer data stays private and is not used to train models. The release names no technology partner and does not say where in the network the service runs. It does list healthcare and professional practices among its target sectors. At CASA26, BT International’s Richard Atherton framed the operator’s job differently: carry the call accurately, govern the data at the point it is handed to AI, and capture a share of the value that data creates.

One name recurs across these bets. ElevenLabs supplies the voice AI in Deutsche Telekom’s Magenta AI Call Assistant and in KPN’s Mobiel AI, and it is a V.AI charter member. The voice layer is converging; what operators are choosing is where the workflow sits and who owns the customer.

The unresolved question is who owns the customer and the outcome. In Vocca’s model it is the vertical player, selling directly to the practice. In the operator-side models it is the operator or the platform, and KPN’s Mobiel AI is the clearest case: there the operator owns the customer, the number and the price. BT’s AI Receptionist sharpens the point, because healthcare practices are among its target sectors: an operator product and a vertical specialist like Vocca may soon be pitching the same front desk. Both can work, but they split the value very differently.

Radisys’s owner wants to be known as a technology exporter

Radisys is a wholly owned subsidiary of Jio Platforms, and Jio is doing something operators have talked about for decades and rarely achieved. Its FY26 annual report says Jio would act as a managed services provider, taking its proprietary network technology to select international markets in partnership with local operators. The portfolio named there includes cloud-native RAN, 5G core, OSS/BSS platforms and UBR-based fixed wireless access, built on a base of more than 524 million customers, more than 268 million of them on 5G, at March 2026.

The timing matters. Jio Platforms has filed for an IPO that Reuters reports could raise around $3.8 billion, with most of the proceeds going to repay debt in the telecom business. Media reports point to an opening on 21 October, and Jio has not formally confirmed the timetable. For public investors the pitch is no longer only Indian subscriber growth. It is an operator that built its own stack, proved it at scale and now sells it.

We found nothing that ties V.AI to that export plan, so we treat the two as context for each other rather than one roadmap. But the structure is worth noticing. A company with two decades of real-time media processing, a footprint inside live operator networks and a parent that is about to be valued as a technology company is leading one of the more serious operator-side attempts at voice AI. The same logic applies to any company that already sits in the operator’s core, such as in IMS and session border control: the shortest path to the call belongs to whoever is already connected to it. Alianza is a case in point, since Crux builds on a portfolio that already includes an IMS core and a session border controller.

Sit in the workflow or under it: the question for every partner

We should be honest that much of the AI Voice conversation, ours included, has centred on the speech layer. This month’s evidence says that is only part of where value lands. Excellent media and telephony infrastructure remains necessary, but it is priced per minute unless it becomes part of a workflow that ends in an outcome. Each group in our community faces a slightly different version of that question.

We will keep following this under our Intelligence and Trust pillars, and we want the vertical workflow companies in the room as well as the model builders and the networks. If you run an agent that ends in a booked appointment, a settled claim or a resolved case, we would like to hear from you.

The question we would put to every CPaaS leader is simple: are you part of the agentic workflow, or the line item underneath it?

Sources

Figures for Vocca and for Radisys deployments are company-reported.

Exit mobile version