The siren song of AI in marketing has been loud, luring with tales of hyper-personalization and effortless efficiency. But for many, that glittering promise is souring, as a frustrated chorus at industry conferences echoes a common complaint: AI agents are "not working." The problem often isn't the technology itself, but its core deficit: a genuine understanding of human motivation.

Avi Yashchin is the Founder and CEO of Subconscious AI, a company leveraging generative AI for investment-grade behavioral science and research. With a background steeped in the market research world, including a role as Research Assistant Professor at New York University, he believes the marketing industry's rush to adopt AI has overlooked a foundational component: the deep, experimental research that uncovers consumer decision-making.

The intern problem: "There was this great quote at one of the earlier panels, 'agents aren't working.' And the crowd erupted," Yashchin recounts from the AIx Marketing conference. "The crowd was just clapping. Yeah, we agree, agents aren't working, but every single person here is deploying marketing agents. How are those two statements true?" For Yashchin, the disconnect stems from AI agents lacking a deep understanding of what causes humans to make decisions, something only rigorous market research can provide. "If you let the language models by themselves create content or perform activities, it's like deploying interns into your company without any instructions."

Rediscovering research: "I'm actually frankly shocked at how little market research is being discussed in this industry," Yashchin admits. "People are kind of trusting the language models to generate copy at scale, which is kind of scary." He argues that without this foundational research, AI operates in a vacuum, unable to truly connect with or persuade consumers. Subconscious AI was founded to make such capabilities widely accessible, prompting an internal debate about shifting from selling to market researchers to directly engaging CMOs and CPOs.

The "say-do gap": Yashchin explains the "say-do gap"—the well-documented phenomenon where what people report they want or will do often diverges wildly from their actual behavior. "If you ask most Americans what kind of coffee they like, they say, 'I love a dark robust blend.' But if you run a randomized controlled blind trial, they all like light and sweet," he explains. "What they say and what they do reliably doesn't match." The gap, he asserts, is where marketing truly lives and breathes. "If the say-do gap is large, marketing is effective. Marketing lives in the say-do gap."

Grounding the chaos: So how can marketers ground their AI? Subconscious AI’s DIY end-to-end platform runs large-scale quantitative research quickly and cost-effectively, exposing the say-do gap. "We'll design the research, we'll design the experiment, we'll recruit for the experiment, we'll run the experiment, we'll analyze the results," Yashchin outlines. This entire process, he explains, happens in minutes, not months, and at a fraction of the traditional cost. The goal isn't to be better than human agents, but to "get 95% of the right answer with 1% of the cost." Such grounding is fundamental. "If you ground every part of the experiment—the personas, the population, the products—then you can start to get a signal that matches humans. Now your interns could be junior marketing techs."