Designing an AI-assisted cognitive health product under real-world constraints

“Am I okay, and what should I do next?”

Users weren’t asking for diagnoses. Raw scores and medical language created anxiety without clarity, while clinics lack capacity for proactive screening.

The product question: how might we deliver early cognitive insight that feels trustworthy, understandable and safe, without overstepping into diagnosis or medical authority?

What the product could not do

• No diagnosis or medical claims (regulatory risk)
• No black-box AI decisions (trust risk)
• No overwhelming data or alarming language (emotional risk)
• Must work for non-technical older adults
• Must be viable as a direct-to-consumer product

Assessment first, AI second

I grounded insights in validated cognitive tasks and used AI only for explanation and synthesis. AI alone can sound confident, but confidence without grounding is dangerous in health contexts.

The tradeoff: slower to build than pure AI analysis and less “wow” upfront, in exchange for much higher interpretability and safety. AI explains and contextualizes results; it does not decide them.

Insight framing over raw scores

Results are framed by cognitive domain (memory, attention, executive function), relative comparisons and plain-language summaries. No red/yellow/green risk labels, and no terms like “decline” or “impairment.” Results are signals, not verdicts.

Intentional cut lines for the MVP

Each of these would have introduced regulatory complexity, trust erosion or false authority, so they stayed out of scope:

• Longitudinal prediction
• Risk labeling
• Automated care plans
• Clinical escalation

From kit to clinician-reviewed service

Orena moved to a fully digital service using licensed BrainCheck Assess. The clinician-in-the-loop model that was on the V1 roadmap is now part of the product: every assessment includes a 1:1 results consultation with a neurologist.

I built the acquisition, onboarding, payments and clinician-review workflows, and designed the current $299 cash-pay offer.

What I measured, and what I deliberately didn’t

Early metrics focused on clarity and completion. Predictive accuracy and long-term outcomes belong in later phases.

• Assessment completion rate
• Drop-off points within tasks
• Time-to-insight
• Qualitative feedback on confusion vs. reassurance
• User questions after results, as a signal of trust gaps