Prof. Valmed Gets Telekom Distribution, Differentiation Difficulties and Data Moats

From kma.Online (translated):

To support doctors in diagnostic and treatment decisions using artificial intelligence, Telekom has partnered with the Langen-based start-up Prof. Valmed. Its AI solution of the same name makes medical knowledge from millions of documents, such as studies and guidelines, directly available within the treatment context, the partners explain in a press release. It will be integrated into existing hospital and practice management systems.

While no concrete plans or timeline are communicated about this partnership, finding distribution and building up presence is of key importance for the startup, which up until now only runs in University Hospital Freiburg (which already uses T-Systems as their cloud solution). According to their website, which considerably changed in the last weeks, there are also integrations planned with CompuGroup Medical, medatixx and Recare.

Telekom iMedOne itself seems to have had little growth in the last 3 years (keeping at 250 customers in Germany), but this is still a good start, especially if Prof. Valmed is to be promoted as the default LLM (which we don’t know as of now).

The HIS market is getting disrupted and HIS companies are starting to use AI and agents as a differentiating feature. Dedalus does that with Corti, while Avelios, Charite and Epic do that with their own AI models (though Epic also integrates with OpenAI directly). And now Telecom gets their own AI agent too.

The question I’m asking myself is is this differentiation going to last or is it going to turn into the new baseline HIS requirement? AI companies have the inherent interest to be as widely adopted as possible, avoiding exclusive integrations with any single HIS player, which has a lot to do with their cost structure. I just made the point above that Prof. Valmed is integrating with other HIS and Corti is similarly available on multiple platforms (including CompuGroup Medical and medatixx). This means that we will see multiple agents available in each HIS, all orchestrated by a harness, so that the initial advantage will level out.

One could make the point that this will lead to HIS commoditising LLMs and making them interchangeable, which very much runs in the opposite direction of what I’ve been arguing about.

At the end there are two modes to consider: the human-to-software mode (the interface layer) and the software-to-software mode (the backend). An LLM can be used for either of those - interacting with clinicians/patients directly or providing the information that some other software/LLM then shows to the user. It is the latter that will get commoditised and the former that will become more valuable.

Prof. Valmed knows this, which is why they are working on capturing the patient interaction now too. If there is one truly valuable and ever-green moat both HIS and AI companies can acquire and use, it is patient data.