Last week, Oura announced its first proprietary large language model (LLM) built specifically for women’s health. It was a notable departure from the "GPT-wrapper" trend, signaling a belief that generic AI isn't yet nuanced enough to handle the complexities of female biometrics.
But this move toward purpose-built intelligence isn't just happening on the consumer wrist. Earlier this month, Dama Health launched Dama Assist, an AI-driven clinical decision support tool designed to help doctors navigate hormone care—from contraception to menopause.
Together, these launches mark a strategic pivot in the ecosystem. We are moving away from "bolting on" third-party APIs and toward building intent-driven infrastructure designed specifically for female physiology.
The Case for Specificity: The 67% Accuracy Gap
The risk of relying on general-purpose LLMs (like standard ChatGPT or Gemini) in women's health is becoming a documented clinical concern. Research has shown that even top-performing general models achieved only around 67% accuracy when tested on menopause-related clinical questions.
If a model isn't built with specific intent, the historical "data gap" in women’s research simply replicates itself in the digital age. As Dama Health co-founder Elena Rueda puts it: "If these tools aren't built with intent, they’ll get good at general medicine, but they won’t get great for women’s health."
Two Sides of the Same Coin: Consumer vs. Clinical
We are seeing a "pincer movement" attempting to upgrade both the patient and the provider experience:
- The Consumer Side (Oura): Oura is tuning its model to be non-dismissive and emotionally supportive. By grounding the AI in longitudinal biometric data (sleep, temperature, cycle), they are attempting to provide the "context" that women often find missing in a standard 15-minute appointment.
- The Clinical Side (Dama Health): Dama Assist is addressing a massive knowledge gap. With only 31% of OB/GYNs reporting formal menopause training during their residency, many clinicians are uncomfortable dosing HRT. Dama provides a "clinical co-pilot" trained on medical consensus documents, allowing doctors to synthesize patient data in real-time.
The Legacy Dilemma: Re-tooling vs. Starting Fresh
This shift creates a difficult tension for the cohort of companies built in the previous market cycle. Many of these startups were designed for a "pre-AI" world—focused on manual tracking, community forums, or telehealth marketplaces.
As the "Native AI" wave hits, these teams are facing a first-principles moment. For some, AI is a natural extension of their DNA. For others, the opportunity cost of re-tooling a legacy system is starting to outweigh the benefits.
We see three distinct approaches emerging:
- The Native Build: Teams building proprietary models to ensure sex-disaggregated data is a first-order variable from day one.
- The Hard Pivot: Founders who realize their existing architecture is a mismatch for the current frontier. As we've seen recently, some are making the hard call to start fresh, recognizing that building "AI-native" from zero is often more effective than forcing it into an old legacy product.
- The Feature Bolt-On: Companies using generic APIs to power simple "chat" features. While these are fast to ship, they risk becoming obsolete as soon as specialized, clinician-vetted models become the expected standard for safety and accuracy.
The Bottom Line
We are moving from Femtech 1.0 (Tracking) to Femtech 2.0 (Contextual Intelligence). The recent shifts at Oura and Dama Health underscore that in women’s health, "general" intelligence may not be enough to drive clinical outcomes.
Whether these proprietary builds will ultimately outperform the generalists depends on their ability to prove clinical superiority over time. But for now, the most ambitious founders have stopped asking "How do we add AI?" and are instead asking: "Is our current business the best vehicle for the AI-native future?"
