Marketing
From 90 to 15 minutes per ad creative: AI production that keeps the brand

Mariángeles Leiva Luque
Published · 8 min

The short answer
You can cut the time each ad takes without the brand falling apart, but AI alone won't get you there. At Growth Lab in Florianópolis, Agência Tab shared that it went from 90 minutes to 15 minutes per creative with two things on top of AI: a design system with colors, typefaces and components, and a manual adjustment by a person at the end. The time was saved in production; the judgment stayed human.
For an online store, the process that works for us has five steps: a short brief, a brand kit already set up, the real product (AI changes the background or the scene, never the product), a human review with a checklist and a Meta test that compares the pieces under equal conditions.
Where the 90 to 15 minutes figure comes from
On November 12 I went to Growth Lab 3ª edição in Florianópolis. Rodrigo Chernhak, partner at Agência Tab, showed in his "logbook" talk how the agency works with AI. One of the numbers was this: each creative dropped from 90 to 15 minutes of production.
What matters is how they got there. AI speeds things up, but each piece comes out of a design system that already defines how the brand looks, and it gets a manual adjustment before it goes live. I covered it along with six other ideas in what I took home from Growth Lab.
To be clear: the 90 and the 15 minutes are Tab's numbers. They are not our measurement and not a promise for your store. Measure how long your team takes today and compare afterwards.
Why having more creatives on Meta matters
In December 2024, Meta explained on its engineering blog how Andromeda, the system that picks which ads to show, works. It said more than a million advertisers used its generative AI tools to create more than 15 million ads in a month, and that it expects the number of creatives in the system to grow significantly.
For your account, that means the algorithm works with what you give it. If you upload two pieces a month, it has little to choose from. If you produce ten good variants in the same time, you have better odds of finding the one that sells.
What to do today: count how many new creatives you launched on Meta in the last 30 days and how long each one took. That is your starting point.
The process in 5 steps
| Step | What AI does | What a person does | What stays untouched |
|---|---|---|---|
| 1\. Brief | Organizes ideas and drafts copy | Picks the angle and the offer | Real prices and terms |
| 2\. Brand kit | Applies colors, fonts and logo | Builds and maintains the kit | The approved logo and palette |
| 3\. Real product | Changes background, light or scene | Picks the best product photo | The product, its shape and its label |
| 4\. Review | Flags obvious errors | Approves or sends the piece back | Nothing ships without this sign-off |
| 5\. Meta test | Distributes delivery | Sets the hypothesis and reads the result | The same budget for each version |
1. A short brief you can repeat
The brief is what keeps AI from making things up. It should fit on half a page: product, audience, angle (for example "24-hour delivery" or "interest-free installments"), exact offer, format and call to action. The best phrases for the angle come from customer conversations, as we explain in how to build buyer personas that work.
At the agency, a person writes the brief and AI fills in copy variants. To keep it consistent, we store the template inside a Claude skill.
What to do: write a half-page brief for your best-selling product and use it for your next five pieces.
2. A brand kit set up before you generate
This is the difference between a piece that looks like yours and one that looks like anyone's. Tab works with a design system in Figma. The Figma guide to libraries explains the idea: a library gathers components, styles and variables in one file and gets reused across every design. If your team uses Canva, its Brand Kit plays a similar role with logos, colors and fonts.
If your brand is not defined yet, start there. We cover it in why to develop your branding before your website, and the same applies to ads.
What to do: load into your design tool a high-resolution logo, the palette with its codes, two typefaces and three templates per format (square, vertical and horizontal).
3. The product is always the real one
This is the rule we defend the most. AI can change the background, the light, the scene or the format. We never ask it to regenerate the product, because it changes details: a seam, the color of a fabric, the text on a label. If the ad shows something different from what arrives, you pay for it in returns and bad reviews.
In practice we start from a good product photo (we keep pushing for professional product photography), cut it out and work around it. In Magnific we use background removal and then light and background changes with Relight. Image expansion, useful to go from square to vertical, sometimes adds objects that were not there: check the edges.
Meta also offers AI background generation as part of its Advantage+ features, but its documentation notes that today it only works with catalog ads on Mobile Feed.
What to do: compare each generated piece with the original product photo, side by side, before approving it.
4. Human review with a checklist
Whoever generates does not approve. Before a piece reaches the account, another person checks it against a fixed list:
The product is identical to the real one (shape, color, label).
Price, installments and shipping match the store today.
Logo, colors and typefaces come from the kit.
The text does not cover the product or sit in areas Instagram hides.
No odd hands, warped letters or objects that do not exist.
At the agency we add an automated check before that person, which reviews spelling, consistency and traces of AI writing. The final sign-off is still human.
Worth knowing: Meta applies an "AI info" label when its tools make a significant edit to the image or when it detects signals from third-party AI tools.
What to do: print the checklist or pin it to your task board and use it on every piece.
5. Test on Meta, under equal conditions
A faster piece is only useful if it sells. To find out, use Meta's A/B test instead of switching ads on and off by hand. According to Meta's page about A/B testing, each version is shown to a different part of the audience, nobody sees both, and you should use the same budget for each. Meta also advises against informal testing, because it leads to overlapping audiences and unreliable results.
If you are new to it, Meta recommends testing a single variable at a time to understand what moved the result. For creatives, that means the same offer and copy with a different background or scene.
What to do: set up your first test with two pieces of the same product that only change the background, and measure cost per purchase, not just clicks.
How to read the test without fooling yourself
| Signal | What to check | Decision |
|---|---|---|
| One piece wins on cost per purchase | That the margin on that sale is still positive | Scale that scene and test variants |
| Tie | Whether the test had enough purchases | Keep the one that is cheaper to produce |
| More clicks, fewer sales | The product page and the price | Fix the store before changing creatives |
| More returns or odd questions | Whether the piece shows something different from the product | Go back to the real photo |
Measure in money. A creative that brings cheap clicks and purchases with negative margin loses you money. That is why in our ad campaigns we cross Meta results with the margin of each product.
What to do: before launching the test, write down the maximum cost per purchase your margin can take. If no piece gets there, the problem is not the creative.
What not to do
Regenerating the product. Even if it looks "nicer", it stops being what you sell.
Generating without a brand kit. Each piece comes out with different colors and fonts, and the brand gets diluted.
Skipping review because the piece took 15 minutes. The time saved is gone if an old price goes live.
Copying another agency's number as your goal. The 90 to 15 minutes is Tab's; measure yours.
Testing everything at once. If you change background, copy and offer together, you won't know what worked.
Leaving creatives for the last week. For Black Friday, build your creative bank early (see our Black Friday tips).
Conclusion
AI can cut creative production time a lot, and the Agência Tab case shows it. What holds the brand together is everything around it: a repeatable brief, a kit that is set up, the real product, a person who signs off and a test that measures in money.
If you want to set up this process for your store and see which creatives are leaving you margin, our paid media for ecommerce service works on it with you. Book a 30-minute account diagnosis with us. You can also check our client success stories.
Sources
Growth Lab 3ª edição, Sympla (accessed Nov 15, 2026). The 90 to 15 minutes figure: talk by Rodrigo Chernhak (Agência Tab), Nov 12, 2026, taken from the slides.
Meta Andromeda: Supercharging Advantage+ automation, Engineering at Meta, Dec 2, 2024 (accessed Nov 15, 2026).
Generative AI features for ads, Meta for Developers (accessed Nov 15, 2026).
Expanding GenAI transparency for Meta's ads products, Meta, Feb 3, 2025, updated Jun 1, 2026 (accessed Nov 15, 2026).
About A/B testing for Meta ads and Selecting a variable for your A/B test, Meta Business Help Center (accessed Nov 15, 2026).
Guide to libraries in Figma, Figma Help Center (accessed Nov 15, 2026).
Remove Background and Image Relight, Magnific documentation (accessed Nov 15, 2026).
You might also like
An AI agent team for marketing: the writer never approvesFrequently asked questions
- How much time do AI ad creatives save? It depends on the process. Agência Tab shared at Growth Lab 2026 that it went from 90 to 15 minutes per piece, with a design system and manual adjustment. Measure your current time and compare after setting up the process.
- Can I use AI to change my product photo? For background, light and scene, yes. Keep the product intact: start from a real photo and check that shape, color and label don't change.
- Does Meta penalize ads made with AI? Meta's transparency policy describes an "AI info" label that appears when its tools make significant edits or when it detects third-party AI tools. It does not mention penalties.
- How many creatives should I test at once? To start, two pieces that change only one variable, with the same budget, in a Meta A/B test.
- Do I need Figma to have a brand kit? No. Figma, Canva or another tool works if it keeps the logo, colors, fonts and templates in one place and the whole team uses it.
About the author

Mariángeles Leiva Luque
Marketing & Comms · We make it Lab
Explore more on



