AI in Healthcare: How Compassana Combines Innovation with Responsibility

Artificial intelligence has enormous potential in healthcare, and there is little serious doubt about its ability to create value. However, large language models in particular remain technically immature. They can present incorrect statements as established facts – a phenomenon known as hallucination – while biases can also influence decision-making without being detected.

For anyone working in healthcare, these risks are as familiar as the opportunities. For Bluespace Ventures, the company behind Compassana, the priority is therefore to create the conditions for AI to be used safely and effectively. Risks must remain manageable, and the value chain must be redesigned so that AI is deployed where it genuinely adds value.

The company’s AI strategy is highly adaptable and built around five interconnected levels. At product level, these principles are put to the test most directly, because this is where users are affected first-hand.

1. Clear rules for employees

Before deploying AI tools, Bluespace Ventures supplemented its IT security policies with an internal AI compliance policy. This sets out how employees may use AI tools and establishes clear safeguards. It allows innovation within a defined legal framework while helping to prevent confidential or personal data from being incorporated into the training datasets of external AI providers.

Briefings were also held to establish a shared understanding of AI and develop employees’ AI competence. Article 4 of the EU AI Act now explicitly requires such competence. It was already essential, however, because employees encounter AI tools in their day-to-day work.

This governance framework provides the foundation for the responsible use of further AI applications.

2. An AI team focused on effective applications

An internal AI team was established on this basis. Its task is to systematically identify where AI can create enough value along the company’s value chain to justify the effort involved. The team evaluates suitable tools, adapts them and makes them available internally once they have demonstrated their value.

In marketing and communications, specialised AI tools enable creative work that would previously have been too time-consuming. For structured workflows, the team uses agentic AI to implement processes in a targeted and secure way.

Every application is assessed consistently against its economic and technical efficiency.

3. Stringent standards for management systems

Bluespace Ventures’ management systems are a particularly important area for AI deployment. They ensure quality, data protection, information security and service management to the standard required of a professional healthcare solution such as Compassana.

Because these systems underpin the company’s reliability, the AI supporting them must meet stricter requirements than AI used in other areas.

AI systems must be factually accurate and free from hallucinations and bias. Where appropriate, locally hosted language models may also be considered. All systems must be logged throughout, with versioning that ensures their operation remains traceable and can be reviewed as part of an audit.

A marketing text can survive the occasional error. A quality or data-protection system cannot.

Careful technical implementation is therefore essential to maintaining the trust of everyone who uses Compassana, whether privately or professionally.

4. Ada as a tested medical device

Compassana integrated the AI-supported Ada symptom checker at an early stage. As a medical device, Ada was tested under particularly stringent requirements. It received regulatory approval only after specialist testing and evidence demonstrating its functionality and reliability.

These quality requirements do not apply to conventional general-purpose language models. A general-purpose LLM produces a possible answer; Ada produces an answer based on a carefully tested and traceable process.

That distinction is fundamental: a regulated medical device is not the same as a general-purpose AI assistant.

5. User autonomy remains paramount

Further enhancements to Compassana are being developed. The guiding principle is clear: users themselves decide how their personal health data is used and who may access it.

Data is not passed back to insurers or to AI systems. Applications are designed to be transparent and purpose-specific and are activated only with the consent of the individuals concerned. The same principles apply to professional users of Compassana Med and to their professional autonomy.

The legal framework is based primarily on the Swiss Federal Act on Data Protection (FADP). Article 5(f) defines profiling, while Article 5(g) defines high-risk profiling. Article 22(2) of the FADP explains how to determine whether data processing poses a high risk, taking into account the nature, extent, circumstances and purpose of the processing, particularly when new technologies are used.

Where a decision is based exclusively on automated processing and has legal consequences for, or a considerable adverse effect on, an individual, Article 21 of the FADP requires the data controller to inform the individual. On request, the individual may express their point of view and request a review of the decision by a natural person.

Further guidance is provided by the EU AI Act. Article 27 requires certain organisations deploying high-risk AI systems to assess the potential impact on fundamental rights.

What this means for Compassana

The five levels are closely interconnected. The IT security policy establishes the framework in which the AI team identifies effective applications. The management systems apply the highest quality standards because they underpin the company’s reliability– and therefore the trust placed in Compassana.

Ada illustrates the difference between a regulated medical device and a general-purpose language model. The legal safeguards surrounding user autonomy ensure that control over health data remains with the users themselves.

In practice, this means that Compassana deploys AI where it demonstrably delivers the greatest benefit at the lowest possible risk.

Before any new application goes live, it undergoes the same assessment of governance, quality and user consent.