by Don’t ask me why, but I kept an article from The New York Times from February 2002 on a “groundbreaking” method of market research at the time called ZMET, or the Zaltman Metaphor Elicitation Technique.
I say don’t ask because I genuinely don’t know, which made rediscovering it a bit like seeing a message in a bottle washed upon my shore, which is to use a metaphor and at the same time to quote Sting, which is always a good thing.
ZMET disappeared, or at least it seemed to based on a little Googling, and there might be really good reasons for this. But, as a premise, the approach still has merit. And here’s the juicy part: As we adopt AI in all sorts of new, even unexpected ways, ZMET might just make a comeback. Its weaknesses would seem to be addressed directly and even wondrously by exactly what AI is designed to do.
First, though, a word about metaphor, the central concept for ZMET and its focus group method of mapping the deeper psychological structures that drive purchase decisions.
Metaphors can be powerful because by understanding and communicating experience in terms of objects and activities, they allow us to pick out parts of our experience and treat them as discrete entities. Once we have made our experiences concrete in some way, they can be referred to, compared, classified, quantified and reasoned about. And these are the very activities AI is best at: comparing, classifying, quantifying and summarizing.
Perhaps most importantly, as metaphor these experiences can be visualized.
Metaphors are not merely language, therefore, but ways of seeing and even understanding. Once expressed in language, metaphors are able to structure thoughts, attitudes and even actions. Consider “the high wall of separation” between church and state, or intellectual “property,” to borrow two from law. These metaphors have shaped policy and guided jurisprudence, sometimes terribly, which is simply to emphasize that metaphor can be powerful.
Because experience is diffuse, fragmented and isolated, a good analogy reveals connections between domains that we wouldn’t have thought had anything to do with one another. By elevating us into a broader expressive context, metaphor and analogy allow us to see a phenomenon in the light of another.
The Core Innovation
Developed by Gerald Zaltman at Harvard Business School in the early 1990s, ZMET reached its peak influence in the early 2000s, which explains the Times picking up on it in 2001. The technique was revolutionary because it challenged the assumption that traditional market research could access the genuine levers of consumer behavior. As Zaltman put it, “A lot goes on in our minds that we’re not aware of. Most of what influences what we say and do occurs below the level of awareness.” Anyone who has sold home furnishings knows this all too well.
Zaltman believed, and I believe, as well, that images and metaphors can transcend language barriers to reveal all sorts of things we can’t otherwise articulate. I teach a course called Visual Rhetoric, which explores how we communicate and persuade using visual information. The fancy word for this is semiotics. Consider that when you came into this world, before you acquired language, you saw the world and, therefore, you thought only in images.
The ZMET technique asks focus group participants to collect images that represent their thoughts and feelings about a topic, then uses intensive interviews with trained facilitators to explore the deeper meanings behind their choices. One of the better known case studies for this approach is the launch of Febreze by Procter & Gamble. ZMET helped the company better understand the psychological relationship people have with odors and freshness. The “cleanliness is next to godliness” myth is a powerful motivator, which is why so many cleaning products use halos, heavenly lights, gold and silver in their advertising.
Another success story for ZMET was Oticon, a Danish hearing aid manufacturer, which used the method to discover that people associated hearing aids with being seen as old and flawed. Other companies that have used ZMET include Coca-Cola, Mercedes and Audi.
The Critique
With that customer list, what happened? Like any good focus group research, ZMET requires competent interview skills and significant time for both interviews and analysis. Another limitation, and this is where AI thunders into the picture, ZMET’s results were deemed by some to be difficult to interpret and non-generalizable, which are common critiques of focus group research in general. So, subjectivity is both ZMET’s strength and its weakness.
Another barrier to wider adoption of ZMET was its commercial licensing model. Confusion about just what kinds of permissions were needed and the requirement to use “trained and certified ZMET interviewers” were problems.
Back to the future, AI would seem to be the ideal complement for ZMET, because AI is a “highly skilled human interpreter.” The GenAI large language models train on psychological literature, marketing research and metaphorical analysis, as well as everything else, right? Thus, AI could reasonably be expected to:
- Analyze visual metaphors at scale.
- Identify patterns across thousands of images.
- Reduce interpreter bias.
- Eliminate the need for specialized training.
But don’t take my word for it, take Claude’s. I’ve been using (and even paying to use!) Anthropic’s Claude all summer for a book project. I asked whether these are the sorts of tasks AI could bring to the focus group table. Claude said absolutely and then gave me detailed breakdowns of just what that could look like.
I’ll spare you the granular details, but suffice to say here that AI image analysis could transform how we process visual information, because without breaking a sweat (metaphor!), AI can analyze compositional elements and symbolic content, apply color theory and psychology, cross-reference images against vast databases of cultural and psychological associations already mapped, and identify patterns among participant responses that human analysts might miss. To this latter task, we know already that AI is proficient at this in medical test analysis, including radiology.
Agentic AI
AI could also conduct preliminary interviews, guide participants through image selection and generate initial metaphorical frameworks, with human researchers focusing on the higher-level interpretation and strategic application. This would be classic AI-human collaboration, the sort already normal for customer service chatbots, for example.
AI-assisted analysis could provide more consistent, reproducible results while still preserving the nuanced insights that made ZMET valuable in the first place.
So, just maybe ZMET was slightly ahead of its methodological moment, which isn’t to dismiss the success it did have. ZMET 2.0 could be a rather ideal test case for how AI can augment human insight rather than simply replacing it.
Regardless, we know visual metaphors reveal unconscious mental structures. The question remains how to tap into these structures, or “pick people’s brains,” to apply one of my least favorite analogies of all time. Keep your brain picks to yourselves!
The original ZMET had participants curate eight to 12 physical images to represent their feelings about a topic. AI could completely transform this, because even I know that AI can create metaphorical images based on text prompts. I’ve also been experimenting with Midjourney, Claude, ChatGPT and Google’s Gemini to generate images. “Sky + fire = sunset” is exactly how they like to be prompted. Instead of asking participants to hunt for existing images, AI could generate personalized visual metaphors based on participants’ verbal descriptions, creating entirely new metaphorical vocabularies.
And because AI’s pattern recognition can identify and distinguish objects, faces and other visual patterns in what would be a metaphorical analysis, AI could automatically identify recurring visual elements across metaphors, detect emotional associations through color patterns across participant groups, and identify culturally specific metaphors. I can see how AI’s pattern recognition could allow marketers to map how metaphors cluster around specific brand attributes or consumer needs, to throw out just one example.
Yes, metaphor and AI would seem to be a powerful combination for the future of market and consumer research. Like taking candy from a baby. A piece of cake. A walk in the park.
