A practical matrix for testing many creative hypotheses at once without losing control of what was tested, who made it, what counted as a qualified result, and why a winner was scaled.

This is built from the campaign pattern in Dmitry's post: the team moved away from testing one offer, one GEO, or one hook at a time. Instead, they used a creator network to test 120 combinations across game x GEO x offer x hook x format in 30 days, distributed the test set through the network in 72 hours, found 3 winners, and scaled the winners to 34M views with a $19.80 cost per FTD.

Use this as a working system: define the test cells, brief creators, collect the same fields every time, score winners, and decide what gets scaled.

<aside> ⚡

Want this matrix adapted to your own product, GEOs, offers and creator network? Talk to UGC Ninja

</aside>

What this gives you

<aside> 📁

This is not a dashboard. It is the planning layer that makes the dashboard useful. If the dashboard says a clip won but the team cannot explain which offer, GEO, hook family and format it represented, the learning cannot be reused.

The core idea

Most creative tests fail because the variables are mixed together casually.

A team changes the hook, the GEO, the creator type and the offer at the same time. A video wins. Nobody knows what won.

The matrix fixes that by turning every video into a named test cell.

Layer What it means Example value Why it matters
Game / product surface The product, feature, game mode, episode, app flow or content surface being promoted Game A, Game B, Feature 1, Offer page Different products create different intent and different conversion quality
GEO The target market or language/country cluster US, CA, UK, AU, LATAM, DACH The same hook can work differently by market
Offer The incentive, promise or conversion mechanism Welcome bonus, free spins, challenge, trial, promo code Offer clarity often changes action rate more than the edit itself
Hook The first claim, scene or tension point proof, challenge, reaction, curiosity, before/after The hook decides whether the test gets enough attention to read
Format The repeatable video structure POV, screen recording, creator challenge, comment reply, green screen A format can be scaled across creators if it keeps working

The matrix fields

Create one row per test cell. One cell can produce multiple creator executions, but the cell itself should stay stable.

Field Type How to fill it Decision it supports
test_id ID T-001, T-002, etc. Lets every creator post and dashboard row map back to the test
product_surface Select The game, feature, offer page or app surface Shows which part of the product creates demand
geo_cluster Select Country, language, or regional cluster Compares market fit
offer Select The incentive or conversion mechanic Compares what makes people act
hook_family Select The hook type, not the exact line Lets the team scale a pattern, not a sentence
exact_hook Text The actual first line or first visual moment Useful for creator brief and post-mortem
format Select Repeatable structure Lets winners become production formats
creator_type Select Native creator, faceless editor, streamer, meme page, niche page Shows where the format can be produced reliably
platform Select TikTok, Instagram Reels, YouTube Shorts, X video Separates platform behavior
brief_version Text The brief sent to creators Prevents false learnings from changed instructions
views Number Verified views only Primary reach signal
qualified_action Number FTD, activation, install, signup or another agreed action Primary conversion signal
cost Number Creator payout plus direct distribution cost Economics
quality_notes Text Moderation notes, bot flags, bad-fit views, brand safety issues Keeps bad wins from scaling
decision Select Scale, retest, pause, kill The output of the read

The scoring model

Score each test cell only after it has enough data to read. If the sample is too thin, mark it retest instead of forcing a verdict.

Score area Question Input fields Score
Reach Did the format earn distribution? views, platform context, creator count 1-5
Action Did the reach create the agreed conversion signal? qualified_action, action per view 1-5
Economics Can this scale inside the target unit economics? cost, cost per qualified action 1-5
Repeatability Can more creators produce this without the original creator? creator_type, format complexity, brief clarity 1-5
Quality Are the views usable and brand-safe? quality_notes, moderation status Pass / fail