The Problem With Averages in Women's Health
Most health technology is built on population averages. Average heart rate ranges. Average sleep cycles. Average symptom severity. The user is benchmarked against an aggregate, and deviations from the average are flagged as anomalies.
For women's health, this approach fails in fundamental ways. Hormonal variation across the menstrual cycle creates physiological changes that would look like anomalies in any system built on averages. Conditions like PCOS, endometriosis, and perimenopause create profiles that diverge sharply from population norms — not because something is wrong, but because these conditions define a different normal.
What Bioindividuality Means for AI Design
Bioindividuality is the recognition that each person's biological systems operate optimally at values that are unique to them. For AI design in FemTech, this means shifting from population-referenced benchmarks to individual baselines.
An AI system designed with bioindividuality in mind:
- Establishes each user's personal baseline before comparing or recommending
- Learns how that baseline changes across hormonal phases
- Flags deviations from the individual's norm, not the population's norm
- Adjusts recommendations based on where the user is in their cycle, not just their aggregate profile
The Design Implications
Building AI systems this way requires more data collection at onboarding — and more thoughtful design of that onboarding experience. Users need to understand why they're being asked for detailed symptom and cycle data, and they need to trust that the data is being used to personalise their experience, not to categorise them against a norm.
We design FemTech onboarding around a simple principle: every question asked should have a visible payoff in the product. If the user logs their sleep quality and nothing changes in their experience as a result, that data collection erodes trust rather than building it.
The AI Architecture That Makes It Work
Bioindividual AI in FemTech requires a different model architecture than population-level health AI. Rather than training one model on aggregate user data, you build a hierarchical system that:
- Maintains a global model trained on population data for cold-start recommendations
- Adapts to individual user data as it accumulates
- Transitions from population-referenced to individually-referenced benchmarks as the individual model matures
The transition timing — when to stop leaning on the population model and start trusting the individual model — is one of the most important design decisions in this architecture.
Why It Matters
FemTech that treats users as individuals rather than data points builds stronger retention, more trust, and better health outcomes. The market is increasingly demanding it, and the AI tools to deliver it exist. The gap is in the product design thinking that connects the two.
