Ember.

The problem

Burnout is no longer fringe — it's an occupational phenomenon recognized at scale across European knowledge work, especially in high-pressure roles in tech, finance, consulting, and healthcare. But the apps built to address it have a structural problem: 52% of users drop off within weeks, and 34% of wellness apps lack any clinical validation, leaving a market flooded with low-quality tools and users justifiably skeptical.

The deeper failure mode is the week-1 wow gap: when a new user opens a journal-driven app on day one with no historical data, the AI reflection feels generic and templated. That's the exact window where the 52% churn happens — if the flywheel doesn't start, nothing else matters.

The solution

Ember intervenes before stress becomes burnout. The daily loop is engineered to be effortless: a 5-emoji mood check-in, a one-line context-aware journal prompt generated by AI, a short writing window, and a tiered reflection that gets deeper as the user's history accumulates.

Two anti-churn design choices solve the week-1 wow gap: (1) onboarding-seeded reflections inject the user's onboarding answers into the day-1 reflection context so the first AI output feels personal, not templated; (2) progressive reflection depth explicitly admits limited data on day 1, then grows as patterns emerge — making the maturity of the AI a feature, not a bug.

How it works

Daily flow: home screen shows energy gauge + 7-day trend bars → mood check-in (5 emoji + streak animation) → "Generating..." orb → journal prompt revealed → user writes → reflection loading → tiered output.

Two-model Claude strategy: Haiku 3.5 for journal prompt generation — single fast call, low cost, good instruction following. Sonnet 3.7 for reflections — better reasoning, warmer tone, stronger grounding in user history (last 3 entries, weekly mood averages by day-of-week). Why Claude over GPT-4o or Gemini: Constitutional AI safety, GDPR/EU AI Act fit (mental health AI is regulated high-risk in the EU), and tone alignment.

19 test cases across three categories: 5 core journeys (first-time user, day-12 pattern showcase, both tier boundaries, alternative prompt), 7 edge cases (short entry, long entry, mood contradicts entry, no pattern, non-English, late night, long streak), and negative cases.

Who it's for

Primary persona: the European corporate knowledge worker, 25–40, in a high-pressure role (tech, finance, consulting, healthcare), iPhone user, digitally native, comfortable with app-based self-care, but hasn't sought therapy. They feel "always on" and recognize the warning signs but won't book a clinician for what doesn't yet feel like a clinical problem.

Geography: Europe-first — both because the burnout-as-occupational-phenomenon framing is more accepted there, and because Claude's regulatory positioning (Constitutional AI, GDPR alignment) is a structural moat against US-only competitors in a regulated category.

Why it matters

The AI mental health solutions market is one of the fastest-growing AI subsegments: $1.82B (2025) → $2.42B (2026) at 33% YoY, with 30–40% CAGR through 2030. But the category's structural problem (52% churn, 34% unvalidated apps) means the long-term winners will be the ones who solve the week-1 wow gap and the regulatory readiness problem at the same time.

Ember's bets: prevention beats reaction (clinical validation is easier when the intervention is earlier in the funnel); Claude over GPT is the right call in regulated mental-health (Constitutional AI, GDPR, EU AI Act compliance); and pattern-aware journaling is the product feature with the longest defensible runway — each week of data makes the AI more valuable to the same user, building switching costs naturally.

At a glance

Project

Ember

Built by

Cristina Milos

One-liner

AI burnout-prevention app for European knowledge workers — pattern-aware journaling, energy tracking, and micro-interventions before burnout hits.