CREATIVE INDUSTRIES
Test a work
before it meets its audience.
Streaming platforms, studios, music labels, producers, content publishers: generate coherent synthetic audiences to test your formats, pilots, music catalogs and talent decisions, before production or launch.
THE THESIS — CARDINAL PROMISE
A work is not judged in the silence of a focus group. It is judged by what happens between it and its audience: boredom, attention, recommendation, forgetting, feedback.
Our system reconstructs what happens, before the audience lives it.
CASE STUDY: GLOBAL STREAMING PLATFORM
“ How do you decide if a high-budget Korean series pilot deserves a full 10-episode order, by anticipating its reception across 6 distinct cultural markets? ”
Tested on 1.2 million synthetic viewers across 6 cultural markets · 24 typologies of relationship to genre series · 9 variants of the pilot evaluated
The Korean studio delivered a 62-minute pilot for a series blending psychological thriller and fantasy, with a cast pairing a Korean star with two emerging actors. The budget for a full order is $180 million over 3 planned seasons. Classic indicators are mixed: internal focus groups (35 people) find the series "original but slow," editorial experts are divided, and the programming slot has not yet been identified.
The platform must decide within 3 weeks: full Season 1 order, development of an alternative pilot, or drop the project. A wrong call is costly: either a flop that weighs on the region's metrics, or dropping a series that could have become a cultural phenomenon like Squid Game. Classic methods are not enough.
What the system revealed
Reception isn't driven by genre, it's driven by cultural rhythm.
The pilot is slow for an average American viewer (attention drop at 18 minutes), well-paced for a Korean or Japanese viewer (rising engagement throughout), and acceptable for a French viewer (stable attention). The problem isn't the series, it's the release strategy: releasing simultaneously in all markets, with the same cut, is a choice that optimizes for 2 markets out of 6.
Music drives 40% of reception differentiation.
Three soundtrack variants were tested on the same footage. The "atmospheric pad" variant gains +34 pts of engagement in East Asian markets but −18 pts in Western markets. The "Western orchestral score" variant does the opposite. A territory-adapted soundtrack strategy, technically feasible in post-production but never done, optimizes all 6 markets simultaneously.
Word of mouth is triggered by scene 4, not the pilot.
Converted synthetic viewers (those who actively recommend) don't convert at the end of the pilot: they convert after scene 4 of episode 2. Teasing this specific scene in the marketing campaign (instead of the current generic teaser) amplifies the recommendation effect by +42%. The real viral material is in the unedited footage.
What Imagine All The People did that no one else could have done
Our system had 6,200 synthetic viewers across 6 cultural markets and 24 typologies of relationship to genre series watch the pilot, tested 9 variants of the pilot (alternative cuts, different soundtracks, dubbing choices, marketing packaging), and projected the word-of-mouth trajectory over the following 12 weeks for each combination. The "territorialized release + adapted soundtracks + scene 4 teaser" strategy dominates 7 of the 9 variants.
No focus group can simulate 6 cultural markets in parallel. No Nielsen panel can test 9 post-production variants. No classic recommendation model can detect that scene 4 triggers viral conversion.
- GLOBAL AUDIENCE PREDICTION
- 68% precisionon internal reference cases
- RELEASE STRATEGY OPTIMIZATION
- +34%projected audience
- MISCALIBRATED ORDER COST AVOIDED
- $22 millionoverinvestment or dropping a viable series
APPLICATIONS
Three uses for the creative industries.
Development & editorial decisions
Test pilots, series concepts, synopses, formats in development. Simulate reception by differentiated audiences before committing the production budget. Inform green-light decisions with granular audience projections.
Testing finished content & post-production
Evaluate multiple variants of editing, soundtrack, casting, and marketing packaging on coherent synthetic audiences. Detect conversion scenes, drop-off moments, and word-of-mouth levers: before release.
Catalog, talent, release strategy
Anticipate the performance of an artist, a music catalog, a novel before signing or launch. Simulate the ideal release strategy by market, calibrate communication campaigns, choose release windows.
METHODS
Three ways to test a decision.
Comparing what a synthetic-population simulation brings against existing methods.
| AI PERSONAS | TRADITIONAL SURVEYS | IMAGINE ALL THE PEOPLE |
|---|---|---|
| SCALE | ||
| 5 to 10 fixed personas | 500 to 2,000 respondents | Up to 10 million coherent individuals |
| DYNAMICS | ||
| Frozen characteristics | Pre-written questions, static answers | Agents that ask, probe and brainstorm |
| TIME HORIZON | ||
| Assumed current state | Snapshot of the present | Projection of future consequences |
| DEPTH | ||
| One synthetic answer | Aggregated statistical tables | Individual dialogue + collective analysis |
| DATA & SOVEREIGNTY | ||
| Limited demographic data · US public cloud | Personal data collected at scale · Public cloud | Zero personal data · Infrastructure in France · On-premise possible |
OTHER CASES
Three other decisions tested in this sector.
FILM STUDIO: GLOBAL CAMPAIGN
"How do you calibrate the global campaign for a blockbuster film across 12 distinct cultural markets?"
Tested on 480,000 synthetic listeners representative of the potential market across 4 countries · 16 profiles of relationship to musical genre · 12 tracks evaluated in pre-production
The artist is highly polarizing: 34% conversion to engaged fans, 41% indifference, 25% active rejection. The classic "median potential" is wrong. Their commercial trajectory over 3 albums is predictable, provided the label invests in the high-affinity niche instead of trying to expand into indifference. Wrong strategy = plateau at album 2. Right strategy = artist established on album 3 with a loyal base.
- COMMERCIAL TRAJECTORY PRECISION
- 71%
- SIGNING NPV DIFFERENTIAL
- +EUR 8.2M
INDEPENDENT LABEL: MUSIC LAUNCH
"How do you launch an emerging artist's second album across 6 European markets without squandering the credibility built by the first?"
2 edits evaluated on 1,800 synthetic cinephile viewers · 28 dynamic follow-ups on tipping-point moments
Edit A (2h14, strong ellipses) is perceived as "demanding but controlled": 68% active recommendation. Edit B (2h34, full narrative development) is perceived as "beautiful but long": 42% recommendation. Counterintuitively, the shorter, harder version generates more word of mouth than the longer, more accessible one. The cinephile wants to feel chosen, not accompanied.
- CINEPHILE ACTIVE RECOMMENDATION
- 68% vs 42%
- PROJECTED REVENUE
- EUR 380K vs EUR 240K
VIDEO GAME STUDIO: FRANCHISE RELAUNCH
"How do you relaunch a video game franchise dormant for 8 years without losing the historic community?"
Projection over 60 months of generational engagement evolution · 18 typologies of animation viewers by age and market
The value of an animation catalog is not linear: it tips at a precise moment (24-36 months) when the first-consumption audience ages and the new generation chooses its references. The strategy that maximizes value at 5 years isn't the one that invests in the most popular series right now, but the one that identifies the series that will become cult classics for the emerging generation. Wrong strategy = catalog worth −40% at 5 years even with heavy investment. Right strategy = catalog worth +25%.
- VALUE PREDICTION PRECISION
- 74%
- CATALOG NPV DIFFERENTIAL
- +$180M
METHOD — CREATIVE INDUSTRIES
A fifteen-minute study.
How a test runs on our system, from the brief to the delivered study.
1
You formulate your decision
A question in plain language, phrased the way you would phrase it in the executive committee. No formalism required.
2
The system builds the populations
Target segments are reconstructed from public aggregates and industry data. No personally identifying data enters the system.
3
The agents interview, discuss, brainstorm
The system runs the interviews itself, probes friction points, and proposes alternative scenarios you had not considered.
4
You explore the study
You talk with individual voices, challenge the reasoning, change the decision and watch the impact in real time.
Your next decision
Which decision do you want to explore?
Describe your need. We can point you to the right level of support.
