Skip to content
See the dashboard The pilot

CadenSpark × your agency · A product for your agency to sell

The brand’s rules, in one box.

Your agency sells it and runs the media. The box produces AI-talent videos, images and copy at scale, the budget follows what wins, and every week the box learns.

Why now

Where people search has changed, and how much brands test.

USD 1B

annualized ChatGPT Ads revenue in under 200 days V

89 countries

with ChatGPT ads since October 21, 14 of them in Latin America P

5 %

of ads turn out to be winners. Volume is what finds them P

New creatives per week, by monthly Meta spend

Top quartileAverage
0204060 < USD 10k10–50k50–200k200k–1M> USD 1M 4.88.116.031.154.6 2.84.16.711.218.9

Motion, Creative Benchmarks 2026: 550,000 ads, USD 1.3B in spend, via third-party summaries P. ChatGPT Ads figures: OpenAI (Aug 31) and indexlab (Oct 21).

The product · Brand Box

Everything the brand is, loaded once.

Variablesroute · offer · date · hook · talent Skills and rulestone · legal · what it never says Templatestalking head · b-roll · image · copy Assets and AI talentlogo · product · voiced avatars
Video templatesTalking head, b-roll, image, copy
Skills and rulesHow it speaks, what it never says
AI talentAvatars with their own face, voice and style
VariablesThe only thing the brand touches

Rules that never break

Locked logo and type Price with conditions Avatar disclosed as AI Never a fake customer Difference rule: every variant changes 2 things Approval before any spend

Example · an airline

Five inputs. Six pieces, most of them video.

Try it: pick a route, offer, date, hook and talent. The box rebuilds the script, the captions and every format.

Brand Box · Ruta SurFictional brand and talent · interface mockup
{talent} AI talent · Sofía
{hook} Hook

Logo, colors and AI label: locked

Script and captions: built from the variables

Video 9:16 · talking head

RUTA SURAI AVATAR

Video 4:5 · b-roll + avatar

RUTA SURAI

Lima for USD 189

From USD 189

Banner 16:5

Image 1:1

Lima for USD 189

Travel by March 15

ChatGPT Ads

Sponsored

Ruta Sur · Lima
Miami–Lima flights from USD 189. Travel by March 15.

Help page · GEO

Organic

How much is a flight from Miami to Lima?
With Ruta Sur, from USD 189 one way, taxes included. Travel by March 15. Conditions on the site.

Avatar disclosedPrice with conditionsBrand voiceReady to approve
4routes
×
3offers
×
3dates
×
3hooks
×
5talents
×
6formats
=
3,240possible pieces

They don’t all launch: ~150 go out each week, in batches, and the ranking decides which ones continue.

The engine · how we optimize

Every week, more spend behind what wins.

~150new pieces every Monday
100%of new ads get a test floor
50 %paused at the first read
10 × 3the top 10 spawn 3 variants each
40–60AI talents at full scale
1
Casting by gapsWe add AI talent where the ad library has gaps: format, audience, region, funnel stage.
2
Monday batch~150 new pieces. Every variant changes at least 2 of hook, talent, format or scene.
3
Test floorEvery new ad gets a minimum budget for a fair read. Nothing is judged without data.
4
RankingEvery piece is scored on hook, retention, CTR and CPM, against our own measurement.
5
Spend to the winnerThe budget is split by ranking. The bottom half gets paused.
6
Variants and lookalikesThe top 10 spawn 3 variants. Winning talent spawns lookalike talent.
7
Real creatorThe top 3 videos get re-shot by a creator from your network, paid on the spend behind their video.
8
Thursday reviewYour agency and CadenSpark read the ranking and rewrite the box for the next batch.

The test floor: nothing is judged without data

test floor Ad AAd B underperforms · pausedwins · scales topped up more spend every week

The floor guarantees a fair read at the start and stops applying once the ad performs on its own. Illustrative.

The ramp: active AI talents by month

40–60 zone 510204060 Month 1Month 2Month 3Month 4After

The roster grows in stages, so each new talent has data before the next one joins.

Ranking: hook 30retention 30CTR 25CPM 15E

The dashboard · 8-week simulation

The budget follows the winner.

Every Monday a batch goes in. New ads get the floor, winners scale, the rest pause, and the best spawn variants.

Week 4 of 8
AI talent
New this week
With spend
Spend in top 10%
Cost index
AdRankingShare of spendStatus

Where the spend is

top 10%
rest

Cost per purchase, index (week 1 = 100)

Engine rules: a test floor for every new ad · anything under its batch’s median pauses · the top 10 spawn 3 variants each · from week 3, the top 3 get re-shot by a real creator from your network.

Illustrative simulation with weekly batches of 150 ads. Not real brand data.

After the box · Your agency

Media spend goes behind what works.

Content at scale video · image · copy Meta · TikTok · AppLovin40% · split by ranking Social and influencers20% · your creator network ChatGPT Ads15% · where available Audiences and reach15% Retail media · CTV · podcasts10% GEO · organicno spend

Example mix. The client approves it; within each channel, the ranking allocates.

More budget to the winnerSpend follows the piece that converts
Creators re-shoot the winnersYour network earns on the spend behind its videos
Our own measurementWhat the platform reports versus what the site records

AI search · GEO + ChatGPT Ads

One screen, two routes: the answer and the ad.

Real sample: which brands ChatGPT names across 11 fintech questions in one market

The brandCompetitor ACompetitor BCompetitor CCompetitor DCompetitor E 9 of 1173322 011
Wins on “card”First in 3 of 4 questions
Loses on “account”Competitor A comes first in 4 of 4

CadenSpark, one ChatGPT run, Oct 30 2026, 11 unbranded questions. Brand names withheld. A sample of the method, not a measurement: the full audit repeats each question 3 times on 3 engines.

Ranking · budget · next batch

What wins goes back into the box.

BrandBox Mondaybatch Spend byranking Readand A/B Thursday:learn every week, a better box

What goes back into the box every Thursday

The winning hookA new template in the box
The winning AI talentLookalike talent for the next batch
Questions that convert in ChatGPT AdsContent for GEO
The format that performsMore budget in the mix
The winning AI videoA real creator re-shoots it, paid on its spend
Anything too similarDoesn’t launch: difference rule

How it works · Your agency × CadenSpark

A product you sell, split 50/50 on net.

1Start with a pilotProve it on a real account, then agree how to move forwardBoth
2You sell the boxTo brands and agencies in your marketYour agency
3The brand createsPicks variables and talent; the box builds the piecesClient or agency
4Monday batch~150 pieces with a test floor and the difference ruleCadenSpark
5Spend by rankingThe winner scales; the best videos go to real creatorsYour agency
6Net and splitMedia, creators, tags and dev come out. The net splits 50/50Both

Your agency

Sells and runs media

  • Sells the Brand Box
  • Your media buyer allocates spend by ranking
  • Your creators re-shoot the winners
  • Owns the relationship and invoices

Both

Decide together

  • How to move forward after the pilot
  • Fees for each client
  • Which costs get netted
  • The Thursday review
  • Net split 50/50

CadenSpark

Builds and measures

  • Builds the box and the AI talent roster
  • Weekly batches with the difference rule
  • Ranking: hook, retention, CTR, CPM
  • GEO and ChatGPT Ads: pixel, catalog, measurement

How it earns

Two fees. One split on net.

1
SaaS feeFor using the Brand Box. Set by the number of campaigns the box is set up for. more campaigns, higher fee
2
Media investmentWhat the brand invests in media, in the mix it approves.

Costs come out

  • Media
  • Creators
  • Tags and dev

What’s left

Net
Your agency50%
CadenSpark50%

We start with a pilot. With results in hand, we agree how to move forward.

The pilot · first 14 days

Two weeks to learn whether it sells.

1234567 891011121314 Product meeting and account mapBoth Flow: creative, media, performance, reportingBoth AI audits, a demo box and 5 AI talentsCadenSpark Pitches with a live demoYour agency Pilot live: first batch on the test floorBoth First Thursday review✓
30sample audits
1010-minute calls
≥ 2brands in pilot
1 in 2continue after the pilot
Months 1–2

Pilot

Your agency, the mapped accounts and 5 AI talents

Goal: 2+ brands in pilot within 14 days

After the pilot

Decide

We review the results together and agree how to move forward

Same split: 50/50 on net

If it works

Scale

More accounts, 40 to 60 AI talents, real creators on the winners, new channels

Measured by brands still active at 6 months