Last Thursday evening, ElevenLabs officially launched in the Netherlands. Although the AI company has had Dutch customers for quite some time, it has now opened an office in Amsterdam. ElevenLabs’ ambitions are bold but focused. CEO Mateusz “Mati” Staniszewski says he wants to claim the interaction layer. What does that mean?
Anyone familiar with ElevenLabs will most likely think of AI-driven speech, whether it’s “cloning” one’s own voice or generating audio from text. The applications for this are already broad and diverse, especially thanks to the very recently released Eleven v4 model. But ElevenLabs wants more, such as building agents and setting up complete marketing campaigns. This is often possible through integrations with other AI providers and models, so that ElevenLabs’ high-profile AI is equipped with the necessary capabilities and tools.
The human factor
The launch of ElevenLabs in the Netherlands took place, somewhat provocatively, right in the heart of Hilversum’s Mediapark, amidst the human-made media of yesteryear. That human touch was also respectfully acknowledged with a demo inspired by *De Fabeltjeskrant*. The actual implementation of ElevenLabs technology, however, is often a bit less artistic: AI-driven customer interactions and marketing campaigns have so far proven to be the most tangible applications of speech AI.
That in itself is a major step forward, but there is definitely a human barrier. For example, German customers tend to ask for a human agent almost immediately over the phone, while Dutch customers are less likely to do so. ElevenLabs also cites research showing that certain accents lead to lower customer participation or satisfaction; thus, the customer helps determine how the AI is rolled out.
On a more practical level, ElevenLabs, together with its client and early investor KPN, highlights the customization required for success. Consider, for example, the always-tricky-to-recognize ZIP codes in the Netherlands. No matter how logical our “6724 Hendrik Johannes” may sound, unclear pronunciation and context that’s hard to decipher make it a stumbling block. Through targeted refinements, KPN and ElevenLabs now correctly recognize postal codes in 82 percent of cases, compared to 64 percent without them.
Eleven v4 in action
The quality of Eleven v4 wasn’t so much evident behind the scenes during the launch as it was continuously visible on screen. As a Dutch speaker spoke, an English translation appeared alongside, constantly refining itself. Even before the presenter was halfway through the next sentence, the translation of the previous sentence was already fully displayed on screen. In addition to speech-to-text, the reverse, text-to-speech, was also heard in all shapes and sizes, including a fine example of “Dunglish” (for the uninitiated, English with a Dutch accent as thick as syrup) and a live demo of rebooking a flight.
It should be clear that interaction and responsiveness were central. That’s also the core of the “interaction layer”: ElevenLabs aims to be the layer through which people talk to and listen to AI. This is regardless of which model is doing the thinking behind the scenes, which is usually Claude, Gemini, or ChatGPT. That focus is actually nothing new for ElevenLabs or its customers, but with the Dutch launch (Belgium had already been launched earlier last week), the platform now has a more prominent presence in the Netherlands. Ideally, that local presence should lead to more relevant certifications and local solutions that go beyond the kind of customization that only the largest players, such as KPN, can implement at scale.
It’s still early in the race among AI solutions to know what interaction with the technology will look like in the long term. A picture is already emerging, and ElevenLabs is thinking in terms of screenless, or at least multimodal, interactions. The company faces competition from various quarters and hopes to excel in areas that others have treated as secondary. For now, the strategy seems to be working, with widespread enthusiasm for the quality and speed of the tools.
Conclusion
For now, ElevenLabs is, above all, an odd one out. Its biggest competitor in speech is Google, which has all the functionality in-house but doesn’t always have the best model. OpenAI and Anthropic offer speech only to a very limited extent, but they do have models capable of all kinds of agent-based workflows. ElevenLabs falls somewhere in between: it certainly offers something that can be called a platform, but relies partly on integrations with those very same providers for its effectiveness. That is precisely why the interaction layer is so important to the company. ElevenLabs doesn’t need to provide the “brain,” nor is it directly entering that fray. Above that lies a layer that is still malleable and not yet executed perfectly by the other AI players. The opportunity for ElevenLabs is to become unbeatable on that front.
Read also: ElevenLabs, maker of AI voice tools, sparks widespread VC interest