Stop Selling Gigabytes: How KPN Built Mobiel AI with ElevenLabs
Small businesses no longer want to talk about gigabytes. They want to stop missing calls. Otto-Jan Woltjer and Leander Zapheriou on how KPN and ElevenLabs built Mobiel AI, launched this month, and why the way it is built matters as much as what it does.
Otto-Jan Woltjer of KPN and Leander Zapheriou of ElevenLabs at CASA26, a few weeks before Mobiel AI went live
Every mobile conversation with a small business customer had become the same conversation. I have five gigabytes, maybe I want seven, and I do not want to pay more. Otto-Jan Woltjer, who looks after KPN’s mobile and voice propositions for small and medium businesses, described it plainly at CASA26: the value of the subscription only goes one way.
So KPN asked its business customers a different question: what is your main problem? The answer was reachability. A missed call can be a missed job worth thousands of euros. And every customer wants something different: some want to be reachable from eight to nine, some around the clock, all want control, and all of it on the mobile, because that is where their customers call.
KPN’s research found something else. Small businesses want AI but do not know what or how. Adoption is low and scattered across tools, most use stops at rewriting emails and LinkedIn posts, and it rarely touches the business processes where the money is. And they worry about what happens to their data.
The fourth transition
KPN sees this as its next big shift: from fixed voice to mobile voice, from mobile voice to mobile data, and now AI integrated into voice. Mobiel AI, which KPN launched on 1 October and describes as a first for a Dutch operator in business mobile, is the first product of that shift. At launch it brings smart reachability you set per moment, transcripts and summaries of calls that can go straight to email or CRM, voicemail greetings written with AI for wherever you are that day, and voicemail transcripts ranked by priority and sentiment.
The live demo made the point best. Otto-Jan typed where he would be, accepted a suggested greeting, and set it for the busy hours. Then he told the app which topics mattered to him, and a voicemail it had filed as spam moved up the list.
Built in pieces, on purpose
The part that will matter most to other operators is how it is built. KPN used to build one product with everything inside. Mobiel AI is one product for the customer, built from separate components. If ElevenLabs releases a better model, KPN swaps it in without changing the product. If a use case requires data to stay in the Netherlands rather than Europe, that can be done for that use case alone.
On top of that, KPN is building a platform in three layers: interfaces (the app, but also web, APIs and workflow tools), an orchestration layer where KPN connects partners to end users, and a partner layer where KPN’s own services, such as SMS, sit next to specialists like ElevenLabs, all checked against KPN’s standards on security and data. Business partners will be able to build their own flows, for example integrating a niche CRM that KPN would never prioritise. Agents and email follow next.
Otto-Jan’s view of partnering was refreshingly unsentimental. KPN is very good at SMS. It is not going to develop speech technology as fast as ElevenLabs. So partner with the best, and keep control of the orchestration.
Why voice is coming back
Leander Zapheriou of ElevenLabs set the wider frame. For years we bent human behaviour around screens. Now technology can bend back towards voice, the most natural way people interact, and use all the data around a conversation to improve it. He sees six areas where operators can put AI to work, from inbound support and new offerings to proactive outreach, contact centre tooling, bundling and internal agents. The smart move is to start low on the maturity curve. Transcription and summaries are easy to adopt; full agentic experiences are a much bigger leap. That is why Mobiel AI starts with voicemail and call summaries.
The collaboration runs deeper than one app. KPN and ElevenLabs worked together on speech-to-text for Dutch, where postcodes, email addresses and street addresses are hard, and Leander said the work improved ElevenLabs’ accuracy considerably. Text-to-speech, agents and real-time translation are next.
The lesson for the rest of the industry is in where KPN started, not where it ended up. KPN did not start with a model. It started by asking what its customers were losing. Our thanks to Otto-Jan and Leander for showing it to the room before it shipped.
More from CASA26
CPaaSAA members have every CASA26 session in full in the member community.
Not a member yet? Tell us you are interested and we will be in touch.
Following the series? New sessions appear here every week until Mobile World Congress. Subscribe on YouTube to catch each one.
CASA26 ran 21–23 September 2026 at the Tobacco Theater in Amsterdam, hosted by CPaaSAA. This is part of our Monday series on Transformation.
