Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Are Using AI-Generated Video Ads That Convert
Transcript
- Lucas: So there's this assumption that great video ads require a full production crew — studio, lighting, actors, the works. But over the past nine months, a really different model has been gaining traction, especially among DTC brands on Shopify. Luna: You're talking about ai generated video. Like, fully synthetic. I've seen some that are surprisingly good. Lucas: Right. And not just good — in some cases, they're outperforming the real thing. I want to focus on one specific example: a skincare brand called Dewy Glow Co. They launched in 2023, bootstrapped, about two million in annual revenue. They were spending heavily on ugc style video ads shot on iPhones, which worked okay, but scaling them meant either hiring more creators or recycling the same clips until viewers tuned out. Luna: Creative fatigue. That's the enemy for any DTC brand running a seven-figure ad budget. Lucas: Exactly. So in February, they tested a new approach: using an AI avatar platform — Synthesia is the big one, but there's also HeyGen and Colossyan — to generate product demo videos. They created forty different versions in one afternoon. Different scripts, different AI actors, different backgrounds. Cost per video? About thirty dollars, compared to maybe four hundred for a real UGC clip. Luna: And what happened to the numbers? Lucas: Over a ten-week test, the ai generated ads had a thirty-four percent lower cost per mille — that's CPM — and a twenty-two percent higher conversion rate than their best-performing UGC ads. Now, you have to take that with a grain of salt: they were A/B testing against their existing control, which had been running for months. Some of that performance lift is novelty. But still, a twenty-two percent bump in conversion is substantial. Luna: Wait — the AI avatars were actually converting better than real people? That seems counterintuitive. I'd think consumers would spot the fakeness and bounce. Lucas: That's the fascinating part. Dewy Glow ran a follow-up survey among purchasers who saw the AI ad. Only twelve percent realized the presenter wasn't a real person. And of that twelve percent, most said they didn't care — the product information was clear, the demo was helpful. The authenticity threshold, for product demos at least, appears to be lower than we assumed. Luna: But I wonder if that holds for testimonial-style ads. If an AI avatar says 'I love this moisturizer,' that's different than a demo showing how the pump works. Lucas: You're right, and that distinction matters. Dewy Glow only used AI for demo and explainer videos. Their testimonials still use real customers. They're treating AI video as a complement, not a replacement. The real value is speed and scale: they can spin up a version for a new product launch in hours, test multiple hooks simultaneously, and kill the losers fast. Luna: And that addresses the creative fatigue problem directly — you can refresh ads weekly without blowing your production budget. Lucas: Exactly. The other tool in play here is Runway's Gen-3, which lets you generate short video clips from text prompts. Some brands are using it for lifestyle B-roll — someone walking on a beach, pouring coffee — that can be layered with product shots. The quality has improved dramatically in the last six months. It's not cinema-grade, but for a Facebook in-feed ad it's totally passable. Luna: If today was actually useful to you, the way we keep these episodes ad-free is through listener support. It's buy me a coffee dot com slash fexingo — just a couple of dollars helps keep the research time funded and the conversations independent. Lucas: Yeah, that's genuinely appreciated. And it's what lets us dig into specifics like the Dewy Glow numbers instead of staying surface-level. So back to the tech — there's a concern I hear from marketers: if everyone uses the same AI avatars, won't all video ads start looking the same? Luna: I've actually seen that on my feed. There's a certain 'Synthesia look' — the slight uncanny valley, the same gestures. It's becoming recognizable. Lucas: It is, and that's a real risk. But the brands that are succeeding with AI video are customizing heavily: choosing less common avatars, using their own branded backgrounds, adjusting the voice tone, and — this is key — combining AI footage with real product close-ups. So the AI is the presenter, but the product shots are real. That blend seems to preserve trust while getting the efficiency gains. Luna: What about localization? That seems like a natural use case — one AI video, dubbed into ten languages, with the avatar lip-syncing. Lucas: That's actually where some of the best ROI is showing up. A DTC apparel brand called Stride Activewear — about five million in revenue, sells mostly in the U.S. and Canada — used HeyGen to create French and Spanish versions of their top-performing ad. They spent maybe two hundred dollars total. Within two weeks, the French version had a forty-percent higher ROAS than the English original. Not because it was better, but because the Quebec market had never seen that ad in their language. Luna: So the barrier to entering new language markets just got way lower. That's huge for DTC brands that are still domestic-only. Lucas: Absolutely. And it's not just language — you can swap the avatar's ethnicity, age, gender to match the target demographic. One brand we spoke to runs four different avatar versions of the same script: one for Gen Z women, one for millennial men, one for older women, one for Gen Z men. Each sees different engagement patterns. They're effectively personalizing the presenter at the audience level. Luna: That feels like a double-edged sword. If a brand tries to target too granularly with AI, they risk seeming manipulative. There's a fine line between personalization and deception. Lucas: I think the key is transparency. We're not seeing brands hide the fact that it's AI — at least the smart ones aren't. They just present it as a demo. But if an AI avatar pretends to be a real customer giving a genuine testimonial, that's deceptive and probably violates FTC guidelines. The brand needs to clearly label that as simulated. Luna: Have any brands run into trouble with that yet? Lucas: Not major enforcement, but there have been a few warning letters from the BBB National Programs. One supplement brand used AI avatars that looked like doctors. That got flagged. So the takeaway: keep AI video for demonstrations, tutorials, and explainers. Keep authenticity for testimonials. That line is still holding. Luna: Let's talk about costs. You mentioned thirty dollars per AI video. What's the full stack? Lucas: Synthesia's starter plan is about thirty dollars a month, which gives you ten minutes of video. HeyGen is similar. Runway's Gen-3 is about fifteen dollars a month for standard generation. So a brand could run a full testing cycle for under a hundred dollars. Compare that to producing a single UGC video — even a low-cost one runs two hundred to five hundred dollars, and that's if you're working with micro-influencers. The math is compelling. Luna: But there's a hidden cost: the time to write scripts and iterate on what works. AI video doesn't fix bad copy. Lucas: No, it doesn't. And that's the trap some brands fall into — they think the AI does everything. But the brands seeing results are the ones that treat the AI as a rendering engine for strategically written scripts. The same copywriting skills matter. The difference is you can test fifteen hooks instead of three. Luna: What about measuring success? Are they using different metrics for AI video vs. traditional video? Lucas: The core metrics are the same: CTR, CPM, ROAS, conversion rate. But one interesting thing is that AI video tends to have higher early engagement — first-second retention — but lower mid-roll retention. Viewers either bounce quickly or they stay. The hook matters even more. The best-performing AI ads Dewy Glow ran had a very specific, problem-first hook. 'Tired of moisturizers that pill under makeup?' That kind of thing. Luna: So the playbook is: use AI to generate volume, test hooks rapidly, kill underperformers early, and scale the winners with slight variations. That's basically what we've seen with static image creatives, but now applied to video. Lucas: Exactly. And video has historically been the hardest asset to test because it's expensive and slow. AI removes those friction points. I think within two years, most DTC brands will be generating their video ad creatives this way, at least for prospecting campaigns. Retargeting might still use real footage, but for cold audiences, AI video is going to be the default. Luna: What about the risk of platform policy changes? If Meta or TikTok decide to deprecate ai generated content, brands could lose that library overnight. Lucas: That's a real risk. Right now, Meta requires disclosure for political or social issue ads, but not for standard product ads. TikTok has a similar stance. But if misuse grows — especially deceptive testimonials — regulators could force platforms to restrict or label all AI video. The smart brands are keeping their production workflows flexible: they can switch back to UGC quickly if needed. But for the time being, the cost advantage is too big to ignore. Luna: Any final takeaway for a DTC founder listening right now who's never tried AI video? Lucas: Start with one product. Pick a simple demo script — under sixty seconds. Generate five variations using different avatars or backgrounds. Run them as a small test with maybe fifty dollars each. Compare against your current best-performing video. You'll have data in three days. That test will cost you less than a single dinner out, and it might change how you think about video production forever.