We used AI for our ad creatives for six months — what we saved, and where we messed up

really honest write-up thanks for sharing. we also went all-in on AI during the initial hype. then we realized the chefs in our ads had six fingers on each hand and loaves of bread were floating mid-air above the cutting board... the human eye catches that weirdness immediately, unfortunately.

We also tried escaping a 50,000 ruble monthly design budget by jumping on AI. Our cost per click jumped from 18 rubles to 26 rubles. People are just sick of seeing the same glossy AI kitchens with identical filters. Authentic product photos still clearly drive way better sales.

Don't use AI for ad copy or banner layouts. Generate the background scene in AI, then drop in your product and campaign copy yourself using a free graphic editor. That completely gets rid of the messed-up typography issues.

If you need an exact 1:1 match of the physical product, plain text prompts won't cut it. You have to use inpainting and ControlNet setups to lock the original product shot in place while only generating the surrounding scene, which requires basic technical know-how.

When customers see the AI image their expectations go through the roof. Once the package arrives and the wood tone doesn't match that fantastic lighting in the render, they instantly open a return request. If the ad visual doesn't match reality, it can seriously drive up your return rates.

instead of firing the designer completely u gotta find a designer who uses ai but u wouldn't pay the same rate settle for half the price but a human does the final check. otherwise your brand image gets trashed with weird distorted products.

How did the CPC and conversion rates of the ads change over these six months though? Did the 41.000 rubles you saved on design make up for any potential drop in ROAS?

One time, in an ad we generated with AI, it placed an object that looked like a red tennis ball next to the cheese board instead of a tomato and we launched it without noticing. Half the comments weren't even about the board, they were asking what kind of tomato that was.

Let me write how it's done in practice. Any unwritten clause becomes a point of disagreement later, as both sides remember it differently.

If you have questions, write them; I'll answer as best I can.

Saved.

I agree. The first thing to break as you grow is communication, not the numbers.

Solutions that work at a small scale collapse when you grow; I learned this late. Correct me if I'm wrong.

let me summarize what's been said so far. i mean any unwritten clause becomes a point of disagreement later, as both sides remember it differently.

im also curious if anyone does it differently.

Let me clarify the technical side. Trying to do this alone is the most expensive way.

I'm also curious if anyone does it differently.

Three different views emerged, they all complement each other. Any unwritten clause becomes a point of disagreement later, as both sides remember it differently.

Correct me if I'm wrong.

I have no experience with visual creation AI, so I'm asking. The reason for failure isn't one big mistake but the accumulation of small decisions.

Let me speak from the other side; I'm on the supplier side. Trying to do this alone is the most expensive way.

The longer you delay cutting the team the more people it affects. anyway just leaving this note it might be useful.

Let me speak from the other side; I'm on the supplier side. Looking profitable on paper but having no cash in the bank is very common.

Mistakes made on the visual creation AI side are usually reversible but expensive. I mean correct me if Im wrong.

I didn't know that. The biggest time-waster for us was not knowing who had the final say.

Just leaving this note, it might be useful.