Trace Launch Research Notes

Working evidence file for the 2025/2026 consumer app launch research goal. This is intentionally rough: source links, numbers, caveats, and synthesis fragments that feed LAUNCH_RESEARCH_REPORT.md.

Research Rules

• Prioritize 2025/2026 examples and benchmark reports.
• Use 2023/2024 cases only when still strategically relevant.
• Separate verified facts, estimates, and self-reported claims.
• Translate patterns into Trace-specific decisions.
• Avoid importing tactics that conflict with Trace's product philosophy: no per-meal verdicts, no shame-based diet language, no unsafe medical claims.

Source Map

Market Benchmarks

Case Studies

Asia Quality Review (2026-06-16)

Main finding: the report was much stronger after the 2026-06-15 codebase pass, but still read like a US-iOS launch memo. The weak spots were:

• Channel bias toward TikTok/Meta/Apple Search Ads; underweighted Kakao, Naver, LINE, Zalo, WhatsApp, and Google Play custom listings.
• Platform bias toward iOS despite Android-first project history and APAC Android reach.
• LTV/CAC bias from North American subscription benchmarks; APAC payer LTV can be real but lower and more uneven.
• Food-data bias: no enough focus on hawker meals, Korean stews/banchan, sauces, soups, shared plates, rice/noodle bases, and drinks.
• Country grouping bias: "Asia" was treated as one market instead of Singapore, Korea, Taiwan, Vietnam, India, and mainland China having different launch mechanics.
• Regulatory/distribution gap: zh-CN was not clearly separated from mainland China readiness.

Working recommendation after this pass:

• Singapore = first APAC proof market.
• Korea = second only after Korean food-scan trust and Korean-native store copy.
• Taiwan = later zh-TW/LINE test.
• Vietnam = content/community learning before paid subscription scale.
• India = useful case-study market but not a first launch without Indian food coverage and localized pricing.
• Mainland China = explicitly out of first-launch scope.

Verified Product & Cost Facts (grounded in the repo, 2026-06-15)

These are checked against the codebase, not inferred — they anchor the report's economics and funnel sections. The earlier draft was written without reading the product and got the funnel and measurement sections wrong.

Unit economics (firm)

Key takeaway: cost is knowable to the cent; the only thing that moves the CAC ceiling is churn, so measure renewal early and stop refining cost guesses.

Funnel / product grounding (firm)

• Shipped first-session order: onboarding → Plan Loading → Food Lens Reveal → Sample Meal Picker → Sample Meal Result (running totals, FREE, pre-paywall) → Paywall → First Scan Prompt (post-payment; OS-notif prompt fires here). SampleMealResultScreen.tsx renders macros + per-condition lens readings with no payment. The earlier draft's "paywall-first" recommendation was a regression.
• Analytics events already exist (apps/mobile/src/utils/analytics.ts): first_scan_completed, scan_result_viewed, close_gaps_viewed, second_scan_completed, paywall_viewed, condition_selected, cancellation_* with days_since_trial_start. Don't tell James to "add" these.
• Attribution is the real gap: source hardcoded 'organic' (IntroCarouselScreen.tsx: "paid + referral sources will hook in when attribution lands"); no MMP. Creator CAC is unmeasurable until this ships.
• GLP-1 companion is free forever (D93) → wedge audience == free tier == lowest-converting segment. Acquisition ≠ monetization segment.
• Platform/locale: ships iOS + Android (project notes: Android-first), 5 locales EN/zh-CN/zh-TW/KO/VI. Report assumed US-iOS — flag the scope.
• Prod AI provider = Gemini 3 Flash Preview (verified 2026-06-16). .do/app.yaml sets AI_PROVIDER=gemini + GEMINI_MODEL_OVERRIDE=gemini-3-flash-preview. The committed apps/api/.env says anthropic but that's the LOCAL default — the DO deployment config is the prod source of truth. (Corrects an earlier in-session flag that read .env and worried prod was on Claude.) → the §5 Gemini COGS rate is correct, and the Asian-food advantage IS live.
• Asian-food bench evidence (D66): Gemini 3 Flash Preview chosen because Haiku failed on Asian foods; 6-photo bench cleared Korean gimbap-with-Spam (prior "salmon" miss) + banchan; "lumping appeared on multi-bowl photos" (strict-decomp prompt rule added). Breadth beyond Korean/Western dishes UNVALIDATED — spot-test laksa/cai fan/pho/jjigae/dim sum/bubble tea per launch market. ai.ts:8 comment still says "Gemini 2.5 Flash" (doc drift vs the 3-Flash prod override).

GLP-1 ad-policy constraint (channel-defining)

• Meta: prohibits prescription-drug-implying ads + restricts weight-loss creative.
• TikTok: prohibits weight-management / Rx-adjacent ads; Spark Ads inherit review.
• ⇒ GLP-1 = organic/seeding/ASO/owned-web only. No paid amplification. Paid spend belongs to non-GLP-1 wedges. Verify current platform policy at launch.

Trial-copy debt (verified, already tracked)

• Dead "3 lifetime AI meal scans / no credit card" model still in PRODUCT.md, flagged PROGRESS.md:1863; orphaned logMeal.trial* locale keys. Reconcile before launch.

Verification Results (web-checked 2026-06-15)

New sources added to the report this pass: trial-to-paid by length (25.5% ≤4d vs 37.4% 5–9d, RevenueCat/Adapty); AI apps +41% RPU / +30% churn (RevenueCat); 95% of annual cancellers never return (RevenueCat SoSA 2026); Adapty H&F benchmarks; ChatGPT $2.91/install RPI (Appfigures).

Tag convention now used inline in the report: [audited] / [self-reported] / [estimated].

Competitor Teardown Data (web-checked 2026-06-15)

Pricing/positioning (annual / monthly): Trace $79.99/$9.99 · Cronometer ~$55/$10.99 · MyFitnessPal $79.99/$19.99 · Cal AI $29.99/$2.99 · MacroFactor $71.99/$11.99 · Lose It! $39.99/— · Lifesum ~$45/$9.99–12.99 · Shotsy $49.99/$9.99 · Noom subscription+own-Rx $119–129.

• Trace sits at the TOP of the price band (tied with MFP $79.99/yr) — only the condition-lens interpretation justifies the premium; the scan alone can't.
• Cronometer = the one philosophical neighbor; measurement instrument (84 nutrients, manual targets) vs Trace's interpretation instrument. ~15M users (self-reported). No automated condition lens.
• Calorie-first pack (MFP/Cal AI/Lose It/Lifesum/MacroFactor) = wrong philosophy, but proves photo-scan is a commoditized front door + Cal AI is the GTM lesson (~15M dl, ~$35M ARR; TikTok creator-led + sequenced paid).
• Whitespace (condition-lens + running-totals + no-judgment) = real but narrow: positioning moat not technical; niche-of-niche TAM; distribution (demoing interpretation) is the hard problem.

GLP-1 Field Data (web-checked 2026-06-15)

• Shotsy — direct precedent for free-wedge→paid; ~100K dl (measured; its "1M+" is self-reported, unreliable). Paywalls the level curves Trace gives FREE; upsells dose tracking, not nutrition → Trace's nutrition bridge is differentiated.
• Noom — the one to watch: GLP-1 Companion + protein logging + Muscle Defense™; but weight-loss-framed, coaching-heavy, drug-bundled ($119–129 own Rx).
• Caloria — closest philosophical match (endocrinologist-built, metabolic steadiness) but ~20K dl (self-reported).
• Telehealth-bundled (Embla ~€150/mo+Rx · Voy £94–239/mo · Found $349–699/mo) = app is a retention wrapper around a prescription; different business.
• "Nutrition steadiness on GLP-1, interpreted, non-weight-loss" = genuinely open lane; Noom is the credible mover toward it.
• TAM: anchor on ~25–30M US adults who've tried GLP-1 (KFF-derived). No credible GLP-1-app market size exists; reject vendor "$Xbn / 280M users" figures (those reports measure supplements, not software).

Ad-policy (the organic-only thesis — confirmed + worse)

• Meta: Rx weight-loss ads need prior written permission + geo-restriction (Health & Wellness ad standard); 35-state-AG coalition (Dec 2025) pushing tighter.
• TikTok: prohibits weight-management / Rx ads; bans "GLP" in product branding; Spark Ads inherit review.
• NEW + load-bearing: TikTok May 2026 guidelines suppress weight-loss- medication content ORGANICALLY, not just paid → GLP-1 is doubly constrained. Frame GLP-1 around nutrition/protein/companion, never weight loss.
• Nuance: Trace advertises a free nutrition companion, not a drug — a nutrition-steadiness ad that never names the med/weight-loss may clear review; worth a small policy-probe, don't bet the launch on it.

GTM Benchmarks (web-checked 2026-06-15)

• Creator seeding: gift+affiliate (no upfront) converts nano/micro only; Cal AI ran 150+ retained creators ~4 posts/mo (the paid version). UGC $150–300/asset; health affiliate 5–30%; paid CPI N.America $2.50–5.00. FTC #ad mandatory, enforcement up on small creators 2025–26; brief experience not outcomes.
• ASO: condition nouns OK; verbs (manage/lower/diagnose) trip Apple 1.4.1; Google requires "not a medical device / consult a professional" disclaimer. NEVER drug brand names in metadata (Apple 5.2.1; Novo 130+ suits 2025) — use "GLP-1". Screenshots = top free lever (+9pp); store-metadata localization +26–30% (Trace already has in-app locales); CPP +156% on referred traffic (AppTweak sober +5.9–8.6%) — needs inbound; ASA popularity = free keyword tool.
• Web-to-app: H&F top performers 23%+ download-to-trial; Noom quiz = 113 screens, email gate ~33%, quiz IS the demo. Trace's free sample-demo is a cheaper show-don't-tell alternative — don't blindly copy Noom.
• Lifecycle: 55% of 3-day cancels Day 0; 82% of H&F trials start on download day; push+in-app carry it, email secondary (open ~48%/click ~1.5%); pause>cancel; 95% of annual cancellers never return → one win-back email only; ~5-min TTV rule.
• Referral: two-sided give-get (91% of successful programs); WhatsApp+SMS ~90% of sharing; mature 20–35% of installs, referred +37% D30 — but compounds AFTER a base; free month > cash.
• SEO: Trace's 976×12 data is a programmatic moat BUT YMYL needs a credentialed (RD/MD) reviewer byline (Sept 2025 update penalized missing author creds); ~30–40% of permutations have volume; 12–24 mo to mature; organic CAC ~3× < paid once mature.
• Highest-EV experiment: test 5–9 day trial (≈+10–15 pts trial-to-paid).

Key source URLs (for the report's source appendix)

• RevenueCat SoSA 2026: https://www.revenuecat.com/state-of-subscription-apps/
• Sensor Tower SoM 2026: https://sensortower.com/blog/state-of-mobile-2026
• AppsFlyer install spend: https://www.appsflyer.com/blog/trends-insights/app-install-ad-spend/
• AI apps +41%/−churn: https://ppc.land/ai-apps-earn-41-more-per-user-but-churn-30-faster-revenuecat-finds/
• 95% cancellers: https://9to5mac.com/2026/05/27/new-report-shows-annual-app-subscribers-rarely-return-after-they-cancel/
• Cal AI growth: https://growthcurve.co/three-engines-and-an-exit-the-cal-ai-growth-playbook
• Shotsy funding: https://siliconflorist.com/2025/02/20/shotsy-lands-2-million-for-weight-loss-shot-tracking-app/
• Noom Muscle Defense: https://www.noom.com/in-the-news/noom-adds-muscle-defense-to-its-glp-1-companion-program-to-help-prevent-muscle-mass-loss-while-taking-glp-1s/
• Meta health/wellness ad standard: https://transparency.meta.com/policies/ad-standards/restricted-goods-services/health-wellness/
• TikTok weight-mgmt ad policy: https://ads.tiktok.com/help/article/tiktok-ads-policy-weight-management
• TikTok May 2026 organic suppression: https://rollingstone.com/culture/culture-features/ozempic-influencers-tiktok-weight-loss-guidelines-1235029891/amp
• Apple CPP / ASO benchmarks: https://www.apptweak.com/en/aso-blog/aso-app-store-trends-benchmarks-report
• Google Play Health Content policy: https://support.google.com/googleplay/android-developer/answer/16679511

Known gaps (next pass)

• No clean public quiz-step visit→install→trial benchmark.
• No sourced health-app-specific organic-vs-paid dollar CAC (only the ~3× ratio).
• KFF GLP-1 user figure cited via aggregator — cite KFF primary directly.
• Non-AI infra cost/MAU still an estimate — pull real numbers for the §5.2 COGS line.

Open Questions For Draft

• Which first launch wedge gives Trace the best blend of pain intensity, creator ecosystem, low claims risk, and paid conversion potential?
• Is "GLP-1 nutrition companion" the best acquisition wedge, or should the free GLP-1 companion remain a lead magnet while paid nutrition is positioned more broadly? (Sharpened: GLP-1 is free forever — what is the explicit bridge from free companion to paid food-logging, and how is it measured?)
• How aggressive can Trace be with hard paywalling when the core food logging path is Pro-gated, without suppressing first value too early? (Note: the free sample-meal demo already provides pre-paywall proof; the real test is a "few free real scans" variant, which COGS makes economically free to run.)
• What CAC threshold is sane before six-month retention is known? (Derived: ≤ $20–25 payer CAC target; the binding unknown is renewal, not cost.)
• What is the platform launch order (Android-first per project notes?) and the geo/language scope, given 5 shipped locales incl. VI?
• Is the competitive set complete without Cronometer (closest philosophical comparator) and the GLP-1 field (Shotsy, Embla, Found, Simple/Noom pivots)?