Luis Roncayolo · Creative strategist for Meta ads

AI-made Meta ads, judged on sales.

I make video and static ads with AI for a DTC apparel brand with a six-figure monthly Meta budget, and decide which ones scale.

$129Kspent on my ads
$296Krevenue from my ads
6,371purchases
2.30xROAS

75 of my ads · lifetime results · Apr–Nov 2026 · each ad competed for budget in ad sets of 6–10 creatives · brand anonymized

Top ads by spend, with lifetime ROAS

The ads that carried the budget

Selfie UGC video2.32x
Story UGC video2.69x
Stadium UGC video2.01x
Four shirts on mannequins in a studio with guitars and records
Mannequin lineup2.54x
Wall-mount video2.05x
Wall-mount video3.22x
Selected work

Three calls that moved the numbers

Four folded shirts with different designs on a dark wood shop table
Lineup 2.47x
Navy shirt on a table in a busy thrift and record shop
Thrift shop 6.26x
Charcoal shirt with Lincoln playing guitar, draped over a box of vinyl records
Record crate 4.32x
Woman in a navy shirt holding a bass guitar in an open garage
Bassist 4.07x
Charcoal 867-5309 shirt with a tag, laid on a road case by guitar amps
Road case 2.11x
Charcoal shirt with a rainbow treble clef on a wooden desk next to a concert ticket
Desk flat lay 4.35x
01 · Static ads

Sell the design, not the format

6.26x ROAS, best single-shirt static

The call
Results were scattered across every format. I stopped testing formats and started testing designs.
Result
Lineup grids won three separate reads of the data. The same design that got 0 sales as a color swatch sold once the real shirt was shown.
Now
Every batch starts from the most recognizable designs, then picks the format. A ledger of past concepts keeps them from repeating.
"yeah." unboxing
Coffee table, no words
02 · UGC video

One word outsold the whole pitch

3.46x ROAS on the one-word version

The call
Scripted pitches read like ads. Real buyers just react. I cut the script to one reaction and the product name.
Result
"yeah." returned 3.46x. Longer scripts and flat "deadpan" reads lost.
Then
I took the words out entirely: a silent coffee-table shot of the shirt became a winning template, rebuilt across the catalog with one locked setup.
Now
Four script templates running across the catalog, each checked for FTC endorsement risk before it renders.
Stack reveal
Street interview
03 · New video formats

Lock the setup, then scale it

2.89x ROAS on the locked stack-reveal setup

The call
New AI video formats look great once and break by the fifth render. I test the setup before I batch it.
Result
Only one surface gave clean stacks, and that version hit 2.89x. For the street interview I ran seven setups and kept the one that held.
Now
A written recipe per format, and one test render before any new setup goes to batch.
Cut

What I stopped making for this brand

About

Trained in design. Tested in sales.

Luis Roncayolo smiling on a cliffside overlook above the sea

I studied architecture, so I think in composition, hierarchy and where the eye lands first.

Then I led a sales team and closed $200K one conversation at a time, so I know what gets someone to say yes.

Now I build the ads myself, with AI handling production.

How I work

Weekly loop

  • Read sales data to pick designs and angles.
  • Iterate from winners and document what works. A smaller share goes to new concepts.
  • Check claims and policy before anything ships.
  • Scale winners, log losers, feed the next batch.
Tools

Stack

SeedanceNano BananaGPT ImageKlingVeofal.aiClaude CodeDaVinci ResolvePremiere ProPhotoshopWhisperffmpeg
Contact

Open to creative strategist roles and creative retainers.

Remote. Start with a paid test batch: 10 ads in 7 days.