Groups do what interviews can't: react together.
Two moderated groups (US/EU wearable owners online; China office workers offline) map the moments of need in the participants' own words, rate three blind audio stimuli before discussing them, walk through the Lilt concept and demo, and react to the measurement idea and the price. Method: Krueger & Casey; reflexive thematic analysis; findings enter the decision model as qualitative likelihood ratios with shrinkage.
Source: research/FOCUS_GROUP_KIT.md (protocol, guide, stimuli, coding) · research/focus-groups/SYNTHESIS.md (synthesis)
Design
US/EU wearable owners · online · 82 min · Apple Watch 4 · Oura 2 · GAD-2 ≥ 3: 4
China office / hybrid · Shanghai · 78 min · Apple Watch 2 · Huawei/Xiaomi 2 · GAD-2 ≥ 3: 3
survey G4 first (pre-screened) · ≥ weekly need moments · incentive $40 / ¥150
scripted moderator + note-taker · founder absent · recorded, anonymised FGx-Py
Need mapping → blind stimuli → concept & demo → measurement → price
Three clips, loudness-matched, order rotated
Known song, instrumental, 60–66 bpm, no percussion, −14 LUFS, gentle low-pass, no sudden dynamics.
Filtered noise + slow pad, no pitch centre, no rhythm, spectral centroid drifting 1.2 kHz → 600 Hz over 60 s, nothing below 60 Hz (the "hum" complaint), no transients > 6 dB.
Calm voice, 4-sentence grounding script over a soft pad. Incumbent-style control.
| Clip | "Would help me settle" (1–5) | "Would irritate me" (1–5) | Fits acute-at-night (n) | Fits wind-down (n) | Neither (n) |
|---|---|---|---|---|---|
| S-A calm real song | 3.4 | 2.1 | 3 | 9 | 1 |
| S-B neutral adaptive | 3.9 | 1.7 | 10 | 2 | 1 |
| S-C guided voice | 2.6 | 3.2 | 1 | 4 | 8 |
Five moments, not one "anxiety"
Both groups sorted their own notes into the same clusters. Each is a distinct job with a distinct texture — the product needs at least two modes (acute / wind-down) and possibly a third (pre-performance: short, sharp).
Rumination after the day
Home, head still running the meeting; replaying what was said. Body: tight chest, restless. Wants: something that stops the loop without asking anything.
Sleep-onset — can't blank the mind
Body exhausted, mind won't switch off; by 1 am the anxiety is about still being awake. Wants: a "soft wall", no melody, no voice, no numbers to watch.
Anticipatory / performance
Exams, investor calls, reviews. Sharp, time-boxed: "I don't need to sleep, I need to not shake for 20 minutes." Wants: 5–10 min, fast entrainment, discreet.
Notification spike
A sound → chest tightens (as in interviews I-03/I-04). Wants: instant, one tap, at the desk, no one notices.
Background tension
"Not anxious, just never off." Daytime; tolerant of real music; identity and taste matter here.
Modes, not a playlist
Acute / night → neutral adaptive, numbers hidden · wind-down → real music, artist identity · pre-performance → short sharp entrainment. Survey B3/B6 quantify the split; the pilot A/B tests it.
Eight themes — who raised them, what they imply
| # | Theme | FG-1 (n/7) | FG-2 (n/6) | Node |
|---|---|---|---|---|
| T1 | Five moments, not one "anxiety" — rumination · sleep-onset · anticipatory · notification spike · background tension | 7 | 6 | need map · personas |
| T2 | State-dependent sound — neutral for acute/night, real music for wind-down; lyrics/beat/sharp sounds make acute worse; two participants disliked S-B as "hospital air-conditioning" (engineering quality is decisive) | 6 | 5 | A1a ↓ · A1b ↑ |
| T3 | The number cuts both ways — the after-result is valued ("92 → 68, I'd screenshot that"); a live heart-rate line "makes it a test I can fail" | 5 | 4 | A3 · design: adapt silently, show after, hide-numbers default |
| T4 | One tap or nothing — no onboarding, no mood/duration/genre pick at 1 am; the watch should trigger it | 6 | 6 | A3 · HMW zero-effort |
| T5 | "Prove it, but don't call it therapy" — measured result is the credibility hook; wellness words welcomed, medical words rejected | 5 | 3 | A3 · positioning |
| T6 | Subscription fatigue → bundled preference; China anchored on 潮汐 ¥218/年 — ¥30 "expensive", ¥15 "试试可以" | 4 | 6 | A4 · form C2 · F9 |
| T7 | Wearable "action gap" — stress scores read "like the weather"; nothing done beyond the watch's breathe prompt | 5 | 3 | A3 opportunity |
| T8 | Real artists matter for wind-down identity, not for acute relief — "at 1 am I don't want anyone in the room" | 4 | 3 | A1b · C3 |
Adapts to me without asking 14 · Shows me the after-result 11 · Works from the watch/ring with one tap 9 · Neutral sound engineered "clean" 3 · Real songs by artists I like 2
S-B rejected by 2/13 as clinical; guided voice preferred by 1 for daytime; 3 would pay standalone if the result is real; one non-owner would not buy a wearable "for an app".
Qualitative, shrunk — can move a belief, cannot cross a gate
| Node | Direction | Reasoning | LR (k = .5, correlated with interviews → ×.6 in log) |
|---|---|---|---|
| A1a | ↓ | 10/13 chose neutral for acute; melody/lyrics rejected again | 0.6 |
| A1b | ↑ | 9/13 chose the real song for wind-down; artist identity valued | 1.4 |
| A3 | ↑ (design condition) | result valued; connect willingness high among owners; live number is a risk → post-hoc result, hide-numbers default | 1.3 |
| A4 | ↔ / ↓ standalone · ↑ bundled | subscription fatigue; bundled/employer preference | 0.9 (B2C) · C2 weight ↑ |
The size of the state-dependent split (C3) and whether GAD-2 ≥ 3 sharpens it · whether "hide the live number" changes D1/D4 (add a probe to the think-aloud) · whether neutral-sound quality is a spec problem or a segment problem (the pilot logs per-participant reactions to S-B).