Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Use Emotional AI to Test Ad Creative
Transcript
- Lucas: Earlier this year, Unilever tested two versions of a Hellmann's mayonnaise ad using an AI platform that reads micro-expressions from viewers' webcams. The results shifted their media spend by millions. Luna: Millions? Based on tiny facial twitches? Lucas: Exactly. The platform — it's called Realeyes, there's also Affectiva — uses computer vision to track seven core emotions across thousands of viewers. The winning ad produced a stronger 'joy' spike during the key moment, and Unilever allocated more budget to that cut. Luna: So they're basically replacing traditional focus groups with AI watching faces. What's the evidence that it actually works better? Lucas: There's a growing body of research. One meta-analysis from the Journal of Advertising Research found that emotional response measured by facial coding predicts in-market sales lift about 30 percent better than standard survey recall data. But the real selling point is scale and speed. Luna: Speed being you can test fifty versions in a day instead of three. Lucas: Right. A traditional focus group takes two weeks to recruit, moderate, transcribe, analyze. Emotional AI can generate results within hours. And you're not limited to a single city or demographic — you can recruit a nationally representative panel via your own app or a research platform. Luna: But there's a creepy factor, right? Asking people to turn on their webcam and let an algorithm judge their emotional state. Lucas: Transparency is critical. Companies like Realeyes require explicit opt-in, and they don't store raw video — they only keep the numerical emotion scores. Still, you're right that it raises privacy questions. Some European regulators have flagged it under GDPR if the data is considered biometric. Luna: Let's stick with the Unilever example. What specifically did Hellmann's learn? Lucas: They tested two approaches: one focusing on the product's creamy texture, the other on a family dinner scene. The emotional AI showed that the family scene drove a 40 percent stronger 'happiness' response among millennials, while the texture ad performed better with empty nesters. They ended up running separate versions to different audience segments. Luna: That's basically dynamic creative optimization powered by emotion scores. Makes sense for a brand with a big media budget. Lucas: And it's not just Unilever. Procter & Gamble, Coca-Cola, even some pharmaceutical brands have started using emotional AI for TV and digital ads. The technology has become affordable enough that mid-sized brands can access it through platforms like Kantar's Link or Ipsos's Emotion AI. Luna: What about the risk of over-optimizing for a single emotion? If you always chase the 'joy' peak, you might end up with ads that all feel the same. Lucas: That's a real concern. Some practitioners talk about an 'emotional valley' — the idea that a successful ad often includes a moment of tension or sadness before the resolution. If the AI is only trained to maximize positive spikes, you might flatten that narrative arc. The best CMOs use the tool as a diagnostic, not a dictator. Luna: So the human creative director still has a job. Lucas: Absolutely. The AI tells you which cut elicits which response, but it doesn't tell you why. That interpretation still requires creative judgment. And the technology has limitations — it's less reliable for subtle emotions like surprise or disgust, and it can be thrown off by glasses, masks, or cultural differences in expression. Luna: What about audio? Voice tone analysis is another layer. Lucas: Yes, some platforms now combine facial coding with voice sentiment analysis. The tone of a viewer's response — are they speaking with excitement or boredom? — adds another dimension. A few automotive brands have used this to test radio ads, where there's no visual to analyze. Luna: Interesting. Has any academic research validated the link between these AI emotion scores and actual purchase behavior? Lucas: There's a 2022 study from the University of South Carolina that correlated facial emotion responses to in-market sales for snack food ads. They found that a one standard deviation increase in positive emotion score predicted a 12 percent lift in sales. But the strongest predictors came from combining facial, vocal, and survey data together. Luna: So the smart play might be a blended approach — AI for scale, human for nuance. Lucas: Exactly. The CMOs I've spoken with say they use emotional AI for pre-testing, then run traditional brand lift studies during the campaign to confirm. It's not either/or. And they're pushing for more transparent metrics — some even demand that their ad agencies provide emotion score benchmarks as part of the creative brief. Luna: Can we take a quick step back? I'm wondering if this technology is more applicable to certain categories than others. Lucas: Good question. It works best for emotional categories — food, beverage, personal care, entertainment. For B2B or high-consideration purchases like insurance, the emotional response is often muted. But even then, some brands use it to test brand recall or trust signals. Luna: What's the cost? Is this only for the big players? Lucas: Pricing has come down significantly. A typical test might cost between $5,000 and $15,000 per creative concept, depending on sample size and number of questions. That's comparable to a focus group but with faster turnaround and more granular data. Some platforms offer self-service tiers for under $2,000. Luna: So a midsized brand with a quarterly campaign budget of $50,000 could afford to test two or three concepts. Lucas: And they should. The ROI math is pretty clear: if emotional AI helps you avoid running a dud ad that would have wasted 20 percent of your media budget, the tool pays for itself many times over. One packaged goods CMO told me their conversion rate on test-optimized ads was consistently 15 to 20 percent higher than non-tested controls. Luna: Let's talk about the ethical side. Are there regulations coming that could restrict this? Lucas: Yes. The EU's AI Act classifies emotion recognition as 'limited risk', but it requires transparency about when the technology is being used. Some US states like Illinois and Washington have biometric privacy laws that could apply. A smart CMO works with legal early to ensure consent language is clear. Luna: And what about the potential for bias? If the AI is trained mostly on Caucasian faces, it might misread expressions from other ethnicities. Lucas: That's been documented. Affectiva and Realeyes have both released updated models with more diverse training data, but it's an ongoing challenge. Another issue is that people from different cultures express emotions differently — a smile in Japan can mean politeness, not happiness. If you're testing globally, you need culturally calibrated models. Luna: So emotional AI is powerful, but it's not a magic bullet. It requires careful implementation. Lucas: Exactly right. The best CMOs treat it as one tool in a broader testing toolkit, alongside surveys, biometrics like eye tracking, and good old-fashioned judgment. Luna: You know, these marketing conversations — tools like emotional AI that actually change how campaigns work — they're exactly why I love this show. And I know we keep this ad-free, which is rare for a business podcast. If you've found these insights useful, and you want to keep the show independent, you can support us at buy me a coffee dot com slash fexingo. Every little bit helps us avoid running ads. Lucas: Yeah, we really appreciate that. It keeps the focus on the content, not on sponsors. Luna: So, where do you see emotional AI heading in the next five years? Lucas: I think it becomes standard in every major campaign pre-test, just like copy testing is today. We'll see integration with ad servers so that emotional scores feed into real-time bidding — imagine a platform that buys more inventory for ads that scored high on engagement. But there's also a counter-movement: some creatives are pushing back, arguing that ads optimized for algorithmically measured emotion lose their artistic edge. Luna: That tension between data and art isn't going away. Lucas: No, it isn't. But the CMOs who navigate it well will be the ones who use emotional AI to inform, not dictate. They'll ask: does this data point tell me something I didn't already suspect? And if it does, how do I use it to make the creative stronger without flattening it? Luna: That feels like a good note to end on. Thanks, Lucas. Lucas: Thanks, Luna. See you next time.