Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Use AI for Creative Testing
Transcript
- Lucas: You know, Luna, one of the biggest pain points I hear from DTC founders is how much money they burn testing ad creative. They shoot ten video variations, run them for a week, and only one or two work. The rest is just wasted ad spend. Luna: Right, and the feedback loop is slow. By the time you know something flops, you've already spent thousands. Lucas: Exactly. So there's a growing wave of brands using AI to test creative before it ever hits a platform's auction. I'm talking about tools that use computer vision and natural language processing to predict how an ad will perform, sometimes within five percent of the actual click-through rate. Luna: That's wild. So the AI looks at the image, the headline, the call to action, and spits out a projected CTR before you spend a dime? Lucas: Pretty much. There are several players in this space now—companies like CreativeX, VidMob, and a handful of startups. But I want to talk about a specific case that really stood out to me. Helix Sleep, the DTC mattress brand, used a platform called Pattern89 to test fifteen variations of a single Facebook ad. They changed the headline, the image background, and the offer phrasing. Luna: And what did the AI predict? Lucas: The AI ranked the variations by predicted CTR. The top-ranked variation had a predicted CTR of 2.3 percent. When they ran it live, the actual CTR was 2.2 percent. The worst-ranked variation predicted 0.8 percent, actual 0.9 percent. Across all fifteen, the average error was less than half a percentage point. Luna: That's surprisingly accurate. But how does the AI actually make those predictions? Is it just pattern-matching against historical data? Lucas: It's a combination. First, computer vision analyzes the image—color palette, object placement, text overlay, even the emotional expression on a model's face. Then NLP breaks down the copy: sentence length, sentiment, power words, calls to action. Those features get fed into a model trained on millions of ad impressions from various platforms. The model learns which visual and textual elements correlate with higher engagement. Luna: So it's essentially a massive pattern-recognition system. But isn't there a risk that the AI just optimizes for what has worked before, leading to creative homogenization? Lucas: That's the big critique. If every brand uses the same AI tools trained on the same data, you could end up with ads that all look and sound the same. But the counterargument is that AI testing lets you take more risks because you can quickly validate unconventional ideas without blowing your budget. Luna: I can see that. You test a weird concept, the AI says it'll probably flop, you don't run it. But what if the AI is wrong? Lucas: It happens. Predictive models are only as good as their training data. If your target audience is niche or your creative is genuinely novel, the AI might misjudge. That's why most brands treat these tools as a first-pass filter, not a final verdict. You still run a small live test on the top three to five predictions. Luna: Makes sense. So it's more about narrowing the field than replacing human judgment entirely. Lucas: Exactly. And the cost savings can be significant. One DTC supplement brand I spoke with said they reduced their creative testing budget by forty percent in the first quarter after adopting AI pre-testing. They were spending about fifty thousand dollars a month on Facebook ad tests. They cut that to thirty thousand because they weren't running as many losing ads. Luna: That's a twenty-thousand-dollar monthly saving. But let's talk about the tools themselves. Are they accessible to smaller DTC brands, or is this primarily for the big players? Lucas: The pricing varies widely. Enterprise platforms like CreativeX can cost tens of thousands a year. But there are newer startups offering tiered plans starting around two hundred dollars a month. So it's becoming more accessible. Even a brand doing a million dollars in revenue can probably justify the investment if they're spending heavily on paid social. Luna: That's interesting. I want to circle back to the data bias issue. If the AI is trained mostly on certain verticals—say, fashion and beauty—could it mispredict for categories like home goods or pet supplies? Lucas: Absolutely. That's a real limitation. Some tools let you upload your own historical ad data to fine-tune the model. So if you're a pet brand, you can train it on your past campaigns to improve accuracy. But that requires having enough data to begin with, which is a chicken and egg problem for new brands. Luna: And what about creative fatigue? If the AI recommends the same winning formula repeatedly, brands might overuse it and see diminishing returns. Lucas: That's a valid concern. The best practice is to refresh your creative library regularly and use the AI to test new angles against established benchmarks. Some platforms actually flag when a specific creative element is showing signs of fatigue based on predicted performance decline. Luna: So the AI can also act as a creative freshness monitor. That's helpful. Lucas: Yeah, it's like having an analytics team that never sleeps. And the technology is improving fast. I've seen demos where the AI can generate entirely new ad variations by remixing elements from top-performing creatives. It'll swap the headline, change the background color, and even suggest different calls to action. Luna: That sounds like it's getting close to generative AI for ad creation, not just testing. Are brands actually using those features? Lucas: Some are. But there's a lot of caution. Brands worry about brand safety and consistency. If an AI generates a hundred variations, who's going to review all of them for tone and messaging alignment? It's still more of a hybrid workflow: AI generates, humans curate. Luna: That makes sense. I think that's the theme of this whole conversation—AI as an accelerator, not a replacement. Lucas: Exactly. And you know, it's conversations like this that make me appreciate the value of ad-free, independent podcasts. No sponsor pressure, no arbitrary content restrictions. Just genuine exploration of what's working. Luna: Totally. And that's only possible because of listeners who support the show. If these marketing deep dives have helped you save time or money in your own work, you can toss a few dollars our way at buy me a coffee dot com slash fexingo. It keeps the podcast free and independent. Lucas: And no pressure at all. We'll keep doing this regardless. But it's nice to know the option exists for those who find real value here. Luna: Alright, back to creative testing. One thing I want to dig into is the role of video. Most of the examples we've discussed are static images. Does AI testing work as well for video? Lucas: It's more complex, but yes. Video adds a temporal dimension—scene changes, pacing, audio. Some platforms analyze video frame by frame, looking at visual transitions, text overlays, and even the sentiment of the spoken audio. The accuracy tends to be a bit lower than for static ads, but it's improving. Luna: Any standout examples? Lucas: A DTC skincare brand called Ursa Major used a tool to test three different video openers for a Facebook ad. One started with a product close-up, another with a customer testimonial, and the third with a before and after comparison. The AI predicted the testimonial opener would have the highest completion rate and conversion rate. They ran it, and it outperformed the others by twenty percent on conversions. Luna: That's a meaningful difference. And it probably saved them weeks of A/B testing. Lucas: Exactly. They went from testing three variations over two weeks to testing fifteen in a single day. The speed is a huge advantage, especially during peak seasons like Black Friday. Luna: So what's the catch? I feel like there has to be a downside beyond bias and homogenization. Lucas: One downside is that the AI focuses on engagement metrics like CTR and completion rate, which don't always correlate with actual sales. You can have an ad that gets lots of clicks but low conversion. Some tools now incorporate conversion prediction, but it requires integrating with your e-commerce data. Luna: Right, so you need a solid data pipeline for that to work well. Lucas: Exactly. Another issue is that AI testing can create a false sense of certainty. Brands might over-rely on the predictions and stop testing entirely, which is dangerous because consumer preferences shift, platform algorithms change, and what worked six months ago might not work today. Luna: So it's a tool to augment testing, not replace it. I think that's a healthy perspective. Lucas: And I think the brands that use it best combine AI predictions with human intuition. The AI says 'these five variations are likely winners,' and then the creative team picks the one that also feels on-brand and emotionally resonant. Luna: That human touch is still important, especially for brand storytelling. Lucas: Yeah. And the technology is evolving so quickly that I suspect in two or three years, AI creative testing will be as standard as A/B testing is today. It'll be built into the ad platforms themselves. Luna: I could see Meta or Google offering this as a native feature. It would be a game-changer for smaller brands without dedicated analytics teams. Lucas: That's the exciting part. Right now, there's still a gap between the big players and the rest. But as the tools get cheaper and easier to use, that gap will narrow. Luna: And hopefully the creative quality will improve across the board, because brands can focus on the ideas that actually resonate. Lucas: That's the optimistic view. I'm cautiously optimistic. But I also think we'll see a new kind of creative arms race, where the winners are the brands that use AI to find insights that their competitors miss. Luna: It's a fascinating space. I'm glad we finally dug into it. This is one of those topics that's easy to dismiss as hype, but the numbers are real. Lucas: They are. And the best part is, it's all about making better ads that people actually want to watch. That's good for brands, good for platforms, and good for consumers. Luna: Couldn't agree more. Thanks for breaking it down, Lucas. Lucas: Anytime. And if you're a DTC marketer listening, I'd love to hear your experience with AI creative testing. Hit us up on social or leave a comment on our website. Luna: We'll be back next week with another deep dive. Until then, keep testing smartly.