Towards niche AI: the "last mile" gamble

While global technology news is dominated by large-scale AI platforms and debates on generative models, a discreet but meaningful trend is emerging in Switzerland: that of niche artificial intelligence, rooted in the field, useful, understandable and cooperative. It is based on a strategic principle that is now central: "last mile" innovation.

Insight: written in collaboration with Gilles Burnier, Senior Innovation Expert at Groupe Mutuel

A Swiss-style strategy

Rather than seeking to compete with the digital giants on the terrain of fundamental models, Switzerland stands out for its ability to transform existing advances into concrete, high value-added solutions. This positioning, based on specialization, adaptation and precision, enables us to focus our efforts on real impact in specific sectoral or territorial contexts.

Zurich, a real-life laboratory

The latest Fact Finding Mission of the Manufacture Thinking think tank, in which Alp ICT took part, highlighted this dynamic in Zurich. The city boasts an exceptional ecosystem, combining scientific excellence (ETH Zurich), industrial power (Google, Nvidia) and public support (Innovation Sandbox). But more than that, a Swiss model of useful AI:

• Fotokite: wired drone for urban firefighters.
• Ethon.ai: ETH spin-off dedicated to model interpretability.
• AI Sandbox: experimenting with AI in a controlled environment.

These projects embody a contextualized, sector-specific, customer-oriented AI, promoting Swiss strengths: reliability, precision, quality, confidentiality.

Read: Towards niche AI - Learning from the "last mile" in Zurich (Manufacture Thinking)

Weak signals of a silent transformation

Through this exploratory mission, several emerging dynamics come to the fore. AI is also becoming a legal issue (who owns AI productions?), an artistic issue (how to represent the effects of AI on our society?), an educational issue (how to learn in a world of intensified access to knowledge?).

We're seeing a distribution of intelligence: innovation no longer comes from a single center, but spreads via horizontal networks involving startups, researchers, local authorities and citizens. This modular, flexible and interconnected model could become a key to adaptability for territories.

Historic DNA in action

Switzerland has always known how to transform what already exists: in watchmaking, pharmaceuticals or machine tools. Today, this same spirit guides its approach to AI: take "open weights" technologies - i.e. models whose weights (trained parameters) are accessible, without the entire code or license necessarily being open - and refine them in high value-added niches (healthcare, finance, legal, logistics). This strategy is all the more relevant in that:

• Switzerland's federal structure encourages a diversity of local uses
• The Swiss Made label remains a guarantee of confidence in critical sectors
• Access to open-weights LLMs (e.g. Llama 400B) enables the development of locally-controlled but distributed AI.

Technical translation of the "last mile" in AI

The concept of the " last mile " comes from the fields of logistics and telecommunications, where it designates the last stage of a service towards its end user, "the last kilometer". Transposed to artificial intelligence, it refers to the ability of an AI solution to fit into real conditions of use, in a given context, with imperfect data and material, human or regulatory constraints.

Technically, this means:

• Contextualized AI models, trained or adapted to local, often noisy and heterogeneous datasets.
• A clear, simplified interface that can be used by non-experts (technicians, carers, civil servants).
• Seamless integration with existing systems, via plug-and-play or open APIs.
• Monitoring and explanatory capabilities, so that the user understands what the model does, on what basis, with what levels of uncertainty.

Far from generic platforms, this is a use-oriented AI, co-constructed with players in the field, and enriched by practical feedback.

3 "last mile" cases in French-speaking Switzerland

• Neosens Diagnostics (Lausanne): this startup develops AI-assisted portable diagnostic tests, capable of analyzing biological fluids on the spot to detect early markers of disease. AI enables real-time interpretation of data from miniaturized sensors, with results that can be used directly by healthcare professionals, without the need for a centralized laboratory.
• Synature (Lausanne): active in the environment and agriculture, Synature combines AI, acoustics and vision to continuously monitor biodiversity or detect pests. Intelligent sensors deployed in the field transmit locally processed data, providing operators with targeted alerts or recommendations, perfectly illustrating the integration of AI into business routines.
• WhisperIT (Prilly): this startup develops a secure AI platform for law firms, enabling dictation, transcription, case analysis and legal research. Hosted on Swiss servers, the solution guarantees data confidentiality while integrating naturally into legal practices.

ConclusionMaking French-speaking Switzerland a last mile territory

AI doesn't have to be spectacular to be strategic. It needs to be appropriated, tested in the field and open to feedback. The Last Mile is an indicator of technological maturity, but also of collective will. It sets a stimulating course for all Swiss regions. The French-speaking part of Switzerland, with its skills, diversity and proximity to the field, has everything it takes to become one of the driving forces behind it.