08 – AI scribes in the heart of Berlin
AI scribes are not ready yet for large scale deployments in Germany. And why mental health companies are oftentimes a mismatch for VCs looking operating on typical return cycles.
This week's top stories:
- AI scribes are reaching Germany. Berlins main hospital, the Charité, has started trialing AI note-taking in several departments. The clinic is working with Microsoft and their Dragonfly Copilot. Why this is bad for medical transcriptions startups.
- VC is a bad funding model for mental health companies. The return cycles of VCs are not fit for mental health companies that need to build with extra rigor. A model calculation illustrates my point.
Odd Lots – News and Research of Last Week:
- Why do investors pass on technical founder startups? Alex Schubert from SciFounders answers this question with 7 different reasons in a recent blog post.
- New sycophancy benchmark. US-based researchers have released Elephant, a nuanced benchmark for agreeableness in AI.
- Talkspace is now on Amazon. The mental health provider will collaborate with Amazon Pharmacy on psychiatric medication delivery in the future.
I. Charité piloting AI scribe for medical notetaking
Their collaboration with Microsoft is one of the first in Germany and a bad sign for startups.
Charité Berlin is testing Microsoft’s Dragon Copilot to benchmark AI against human scribes in clinical workflows like the ER. This is a test, not a launch, nor a full rollout.
That alone says a lot. If one of Germany’s most prestigious hospitals is only cautiously piloting a tool from Microsoft, then the message to startups is clear: the market is not ready for you yet.
This is about more than tech. Germany’s digital infrastructure is still years behind. Electronic health records are just getting rolled out. And other leading hospitals like UKE Hamburg are still in the early stages of digital transformation, with no public plans for AI tools in the clinical workplace.
Add in concerns around data privacy, liability, and reliability, and you get a market that’s still deeply skeptical — even for basic automation, let alone AI scribes.
AI scribe startups face a tough market:
- Germany doesn’t trust startups to deliver at clinical scale. Microsoft gets the pilot. You don’t.
- AI scribes aren’t a product category yet. I see a serious risk that the category will mature so fast, that scribes will get absorbed as features into existing EHR or hospital software before they get a chance as a standalone product.
- This means: no first mover advantage. Solutions by existing players will dominate deployments. Microsoft got in because Charité is already on their cloud. Distribution is the next priority after trust.
- Denmark’s Noteless has managed to get to partnerships with clinics. This shows that the blocker isn’t EU law, it’s Germany’s system and mindset.
If you’re building in this space, calibrate your expectations. Selling AI scribes into German hospitals isn’t impossible. Just not realistic in 2025.
II. VC is the wrong funding model for mental health startups
When returns don't match reality.
The hypothesis should be clear from the headline. So let me show you with a model calculation. 👇
I will be using a fictional DiGA startup as an example. Here is a quick reminder of the criteria a DiGA needs to meet:
- Smartphone app or browser-based web application
- Supports the detection, monitoring, treatment, or alleviation of diseases, or the detection, treatment, alleviation, or compensation of injuries or disabilities
- Costs covered by statutory health insurance (GKV)
- Evidence of positive effects on healthcare outcomes
- High standards regarding safety, functionality, quality, data protection, and data security
- Classified as a digital medical device in risk class I, IIa, or IIb
- CE-certified
- Free of advertising
- Interoperability: connection to the electronic health record (ePA), integration of the health ID, etc.
This Definition is taken and translated from the DiGA Report 2024.
Okay, let's do some public math:
Mental Health Startup X: The DiGA Path
Assume a company develops a DiGA in mental health. From inception to market, the timeline is at least 18 months. Costs include:
- Product development
- Certification under ISO or MDR
- Clinical trials to prove efficacy
This brings total costs to somewhere between 600,000€ and 1M€, with no revenue until certification is complete.
Once certified, the product is locked. Any meaningful update requires recertification. For a software product, that means innovation slows down by design. You launch, then freeze.
When the product is finally ready, it enters the market as a reimbursable digital health application. Reimbursement averages 200 to 300€ per 3-month prescription. Most apps are only prescribed for a limited period, and while follow-on prescriptions are increasing, the DiGA Report 2024 shows this growth is slow. Long-term user retention is unclear.
Let’s assume a company receives 250€ per prescription. To make back 800,000€, it needs 3,200 activated prescriptions. That’s realistic within a few months, considering that 35,000 prescriptions were activated in December 2024 alone, across around 70 DiGAs, most in mental health.
But that’s only the start. If you budget for:
- Cashflow until break-even
- Marketing and Sales (!)
- Failure risks or pivots
- Potential delays in trials or approval
…you are realistically looking at a 2–3M€ ticket size for a company to have a shot at getting to Series A.
And it might take 2 to 3 years before the company generates revenue, depending on how clinical trials and BfArM listing go. With further research or setbacks, this stretches to 4+ years.
So, a DiGA is not a fast track. It’s a narrow path with regulatory cliffs on either side.
Of course, the DiGA path is not the only one available, and it likely won’t be the default choice for every health company going forward. Many teams already avoid it to maintain flexibility. But the reality is that, if you aim to deliver clinically valid, high-trust mental health products, you’ll still need certifications, trials, and regulatory clarity. That means you’ll face many of the same costs and delays that come with a DiGA anyway — just without the reimbursement upside.
VC Fund Y: Standard Economics, Misaligned Incentives
Now contrast this with a simplified venture capital fund model I made up:
- Fund size: 100M€
- Fund duration: 10 years
- Management fees: 2% annually = 20M€
- Deployable capital: 80M€
- Initial check per company: 1.5M€
- Number of companies: ~30
- Follow-on reserve: 35M€ for 10–15 companies
Let’s say returns follow a typical power law distribution:
- 18 fail = 0€
- 7 return 1.5x = 15.75M€
- 3 return 3x = 13.5M€
- 1 returns 10x = 15M€
- 1 returns 25x = 37.5M€
Total fund return: 81.75M€, below breakeven, and certainly not a success.
To get to a 3x fund return, VCs need:
- One 50x exit, or
- Two 25x exits, or
- A handful of solid 10x wins
That’s hard in any industry. In mental health, it’s near impossible. DiGA companies are not built to scale fast. They are not allowed to iterate fast and can not be designed to hit hypergrowth.
Yet the VC clock runs on a 5–7 year return window. Most funds need clear growth signals by Year 3 or 4 to justify a Series B (or generally further support of the Startup) or prepare for exit planning.
A DiGA company that spends two years in trials and certification is, by default, already “slow”. If it then optimizes for quality over speed — which is what mental health demands — it is often seen by investors as a “walking dead” company.
What This Means for Funding
There’s a fundamental mismatch between the structure of venture capital and the reality of mental health innovation in regulated markets like Germany.
- VCs will push for scale and early proof of returns. That means sales pressure before trust is built, marketing before clinical stability.
- DiGAs require rigor and patience, not speed.
- Founders will be forced to promise the moon to get funding, even if the business needs a decade of focused work.
- Dilution over long timelines makes outcomes for founders less attractive. Many will burn out before a meaningful exit is even possible.
In theory, VCs want “impact”. In practice, they need “hypergrowth”. In theory, DiGAs enable access. In practice, they freeze iteration. And in between, founders are expected to build slow medicine on fast money.
Mental health startups are being pulled in two directions. On one hand, the DiGA system demands medical-grade stability. On the other, venture capital demands tech-level growth curves. Both structures were built for different outcomes.
Trying to reconcile the two often ends in disappointment.
There are ways to build sustainable businesses in this space. But the combination of DiGA and VC money? It’s a bad fit — structurally, financially, and ethically.
Odd Lots – News and Research.
What happened last week?
I. Reasons to pass on technical startups
https://alexschubert.xyz/2025/05/02/why-we-have-to-pass-on-startups/
In this article, one of the cofounders of SciFounders outlines the main reasons they pass on investments. The fund is not mental health tech focussed but rather on businesses with technical founders and research heavy products.
II. Using Reddit to Benchmark LLMs friendliness.
Rhiannon Williams
Sycophancy is a major issue for any health related AI applications. A new benchmark, Elephant, helps in identifying the issue. But solutions to the problem are still unclear. The benchmark is partially based on the “Am I The Asshole?” Subreddit and more nuanced in its assessment of extreme agreeableness than previous tests.
III. Getting your medicine from Amazon
Talkspace formed an agreement with Amazon Pharmacy for fulfillment of psychiatric medication treatment. The company has one of the biggest mental health platforms in the US, that is accessible to 140 Million people through their insurance plans.
Alright, that's it for the week!
Best
Friederich
Get the weekly newsletter for professionals in mental health tech. Sign up for free 👇