The More We Rely on AI, the More We Need to Talk to People
By Rob Douglas
Art: A Chosen Soul
Long before I worked in advertising, I learned something about consumer research while restocking sweaters.
During college and afterward, I spent six or seven years managing United Colors of Benetton stores in New York City. I also bought their merchandise, choosing the items, colors, sizes and quantities from the broader Benetton collection that would ship from Italy.
Two of those stores could hardly have been more different.
At the Empire State Building, we served a constant mix of tourists and office workers. At 62nd Street and Lexington Avenue, we served an upscale neighborhood near Bloomingdale’s and the Barbizon. Brooke Shields shopped there. So did Carolina Herrera, buying for one of her daughters. Those were always interesting encounters.
Same brand. Same city. Very different customers.
Getting the merchandise right required understanding those differences. And that took more than a sales report.
The most useful question was why
I spent a lot of time watching what sold and what stayed on the shelves. When you are constantly restocking throughout the day, you develop a pretty intimate relationship with both.
I also talked to regular customers about what they liked, loved and hated. Before buying each new season’s merchandise, I held a full staff meeting devoted to what was selling, what wasn’t, and why.
The “why” was usually the most revealing part.
Lexington Avenue became our fashion-forward store, carrying trendsetting colors and silhouettes that, Benetton corporate told me, few other U.S. stores had ordered. The Empire State Building did exceptionally well with rugby shirts, backpacks, skirts and jewel-neck sweater sets.
My longtime friend Josephine, better known as Pebbles, could sell an affluent tourist a stack of 15 or more classic Benetton sweaters in one visit.
Over time, my stores consistently ranked among the strongest in sales revenue, even against Fifth Avenue locations two or four times their size.
I cannot isolate every reason for that performance. But listening shaped what I bought, and the sales results helped me judge whether those decisions were working.
Only much later did I recognize the research principle behind that routine: combine qualitative understanding with quantitative evidence.
What qualitative research actually does
Qualitative research explores people’s experiences, motivations, language and behavior through methods such as in-depth interviews, small-group discussions, shopping observations and visits to people’s homes.
It helps marketers uncover needs, interpret behavior and discover questions they had not thought to ask.
Quantitative research measures patterns: how many people hold a view, how often something happens, how groups differ, or how behavior changes. Surveys, transaction records and experiments can all contribute.
Used together, the two approaches strengthen each other. Sales data might reveal that a product underperforms. Interviews and observation can explore possible reasons. A larger survey or market test can then assess how widespread those reasons are and whether a proposed solution works.
My store conversations were informal listening, not a rigorously designed research study. Professional qualitative research adds deliberate recruitment, skilled questioning, systematic analysis and a search for evidence that challenges the initial explanation.
But the underlying habit was sound: stay close enough to customers to understand what the numbers leave unresolved.
When a customer profile becomes a substitute for understanding
Years later, inside Dentsu, I became concerned that qualitative understanding was getting pushed into the background while large data platforms took center stage.
I remember working with my colleague Diego and my now-business partner Boris on the U.S. audience for DAZN’s boxing offering. We challenged the audience picture being developed through M1 because it did not square with our firsthand knowledge of American boxing fans.
That knowledge gave us grounds to question the output. It did not make our own assumptions infallible. It made deeper investigation necessary.
A profile can tell you that someone follows sports. Understanding what a particular fighter means to that person, how a fight fits into their social life, or what might persuade them to pay for a subscription requires another level of inquiry.
The problem arises when an analyst turns an association into an explanation without enough evidence. Familiarity with a platform does not automatically confer familiarity with the people it describes.
Cultural fluency matters. So does the humility to investigate a market before explaining it.
That experience was years ago. My point is about how we used and interpreted the information then, not an assessment of Dentsu’s technology today.
It left me with a question I still ask: What did we learn from consumers that we could not have learned by reading their profiles?
AI makes that question more urgent
AI can help researchers organize evidence, identify patterns and develop hypotheses. It can also make a thinly supported explanation sound remarkably convincing.
That is why I believe marketers should give qualitative consumer research greater priority as AI becomes more embedded in planning.
A 2025 article published by Columbia Business School reported a survey of more than 170 market research practitioners and users: 45% were already using generative AI in their data and insights activities. Among current users, 62% used it to synthesize interview transcripts and other documents. The authors also highlighted uncertainty about how well models anticipate dramatic behavioral changes or discontinuous product innovations. [1]
There is a productive role for AI here. But a model-generated explanation is still something to investigate.
The Nuremberg Institute for Market Decisions has identified another concern with synthetic respondents: a tendency toward mainstream brands and opinions that can miss early adopters and niche perspectives. [2]
For challenger brands, that matters. An overlooked frustration or an emerging behavior in a small community could be the beginning of a growth opportunity. We should be especially curious about the people who do not fit comfortably inside the dominant pattern.
AI can help generate a new angle. Fresh consumer research helps establish whether that angle reflects something people actually experience and care about.
What a home visit can change
P&G offers a useful example with Swiffer PowerMop.
According to the company, home visits revealed that many consumers saw a mop and bucket as the standard for a thorough clean. That shifted the design objective toward outperforming that cleaning standard, beyond Swiffer’s established convenience benefit. P&G also used AI to test and refine advertising executions. [3]
Consumer understanding shaped the problem to solve. Technology helped develop and communicate the response.
That is a compelling model for marketers: use AI to make research and execution more effective while continuing to collect fresh evidence from people’s lives.
Put qualitative research where it can change the decision
Reprioritizing qualitative research means involving it early enough to influence the target audience, product, proposition and creative brief.
A few practical changes can help:
Start with an unresolved business question. Why do people try the product but fail to return? What makes a premium price feel justified? What keeps an interested prospect from buying?
Include people beyond loyal customers. Lapsed buyers, competitor customers and people who considered the category but walked away may reveal barriers current customers no longer notice.
Observe behavior as well as asking about it. People do not always remember, recognize or accurately explain their own habits.
Use findings to build and test hypotheses. A compelling interview is a lead to investigate. It does not establish how common a view is.
Bring frontline employees into the process. Store associates, customer service teams and salespeople can help identify questions worth taking directly to consumers.
Use AI to extend the work. Let it assist with transcripts, comparisons and analysis, then check its interpretations against the original evidence, including contradictory responses.
Qualitative research has its own vulnerabilities: leading questions, selective recruitment and the temptation to hear what we hoped to hear. Doing it well requires discipline.
It also requires willingness to change a decision.
At Left Off Madison, our 5-C process brings Consumer, Culture, Category, Competition and Company into the same strategic conversation. The Consumer dimension demands direct attention to how people live, choose and use what brands sell.
That responsibility becomes more consequential as technology lets us make and execute decisions faster. A mistaken assumption can travel a long way before anyone stops to question it.
Back at Benetton, the sales figures helped me decide what deserved more shelf space. Customers and staff helped me understand what might deserve a place there next.
I still think about those twice-yearly staff meetings. Before committing to the next season, I made time for people who could challenge what I thought I knew.
Marketers need to protect that time.
The next growth opportunity may begin with a question your dashboard has not taught you to ask.
What are your thoughts on this? Write in the comments below.
Sources
Columbia Business School, How generative AI is transforming market research, April 8, 2025. The survey describes its respondents; it is not an industry-wide adoption census.
Nuremberg Institute for Market Decisions, Generative AI in Market Research: Can Machines Simulate Human Insights?, 2024.
P&G, A Century of Curiosity: P&G Analytics & Insights Discovering the Future, 2024 series. The Swiffer example is P&G’s own account, not an independent causal evaluation.
