#214
JP · BRAND S · D+14
「成分表を読む30秒」
TRUSTHigh confidence
「成分表まで出してくれるブランド、初めて見た」
“First brand I’ve seen show the full ingredient list” — saves rose before shares. Verdict: ingredient trust is the purchase trigger.
We deploy hundreds of short-form videos into real markets and learn behavioral dynamics from the reactions.
All figures are illustrative. Replaced with sourced data after the first measurement.
We deploy hundreds of AI short-form videos into real markets, observe the reactions, and learn consumer behavior dynamics.
For each hypothesis, dozens of AI short-form variants go to TikTok · YouTube Shorts · Reels in the same market and window, on organic reach. Paid boosting is excluded.
Views · likes · comments · shares · saves · conversions are observed hourly; comments are read per language. No personal data is collected.
From the pattern of reaction changes we judge four dynamics — Propagation · Trust · Habit · Exploration — and propose the next hypothesis. Judgments are sealed before deployment and checked against D+14 data.
Variants are deployed under identical conditions. With paid boosting excluded, only organic reactions remain.
14-day window after deployment. No personal data is collected; small groups (below the minimum cell size) are not reported.
Before deployment we record and seal which creative will move which dynamic. At D+14 we compare against the data and publish the hit rate.
Judgment rules are updated per experiment, and the baseline (natural variance of a redeploy) is re-measured with them.
The overlap of the two curves is the agreement. A redeploy of the same creatives overlaps only 91%, so 91% is the ceiling. SAMPLE DATA
「成分表まで出してくれるブランド、初めて見た」
“First brand I’ve seen show the full ingredient list” — saves rose before shares. Verdict: ingredient trust is the purchase trigger.
「今のが合ってるから変える気はない」
“What I use works, not switching” — 42% of comments name a current product. Verdict: high switching resistance; comparison messaging needed.
「wait is this the one on amazon with zero reviews?? ordering to test」
Zero reviews worked as an exploration trigger — the only creative where shares led saves. Verdict: target the first-reviewer cohort.
As a seismograph measures the ground, SEISMO measures the market. The gap between the sealed judgment and the D+14 data trains the next experiment, so the rules sharpen as experiments accumulate.
Shares and reposts grow before views do
Saves and trust-signaling comments rise first
Mentions of current products; delayed switching
Question comments and off-platform search spike
320 videos · 14 days · 11 creatives where saves rose before shares were judged Trust. Next hypothesis: lead with the ingredient list.
318 videos · 21 days · 7 creatives turned “zero reviews” into an exploration trigger. Next hypothesis: comparison-led messaging for the first-reviewer cohort.
Quant at Robust Investment, digital marketing lead at LINE Vietnam, RPA developer and CEO at Creatory. The roles changed; the seat was always data. In 2024 he spun Creatory’s AI unit out as Digicon.
Team members with MS/PhD-level degrees
Singapore · Ho Chi Minh City · Seoul
Headcount · to be confirmed PLACEHOLDER
A first answer, from real deployment data, to the question the simulation camp raised.
What 23 creatives that broke the sealed judgment revealed.
Why we chose a 14-day window, and how 7- and 21-day windows differ.
Pick one hypothesis and one market; the judgment report arrives 14 days later. You can also start with just the sample report.