Menu
Close
.svg)
Over the past few weeks, I’ve been thinking a lot about how artificial intelligence is beginning to reshape qualitative marketing research in the life sciences industry.
A few days ago, I wrote on LinkedIN about the emergence of AI avatars and how they may fundamentally change visual aid testing by allowing us to present highly realistic, standardized stimuli to physicians. The response from colleagues across the I&A community was encouraging. Clearly, many of us are thinking about how these tech tools might fit into our research studies...and where we should still use live reps.
Today, another topic came up in a conversation with one of our field and recruitment partners that I suspect will generate even more debate: AI interview moderation.
Specifically, we were discussing whether it might be possible to run a head-to-head comparison of an AI moderator versus a human moderator in qualitative interviews.
At first glance, the question seems straightforward.
But the more we talked about it, the more complex it became.
The first challenge is deceptively simple: how do you define effective moderation?
In quantitative research, performance metrics are relatively clear. Sample size, statistical significance, confidence intervals. These are well-established benchmarks.
Qualitative research is quite different.
A skilled moderator does more than simply ask questions. In a focus group or advisory board setting, she reads the room. She picks up on hesitation, enthusiasm, or discomfort. She reads body language. She looks not just for what is said, but what isn't. She knows when to follow the guide and when to depart from it.
So if we were to compare an AI moderator with a human moderator, what exactly would we measure?
Consistency is one possibility. An AI moderator certainly could pose the same set of questions with remarkable precision, ensuring every respondent receives the same prompts in the same sequence. An AI moderator can ask the why questions too.
But consistency alone is not what makes qualitative research valuable.
The real power of qualitative research often comes from probing: the tailored follow-up questions that uncover motivations, assumptions, and reactions that weren't always anticipated in the original discussion guide.
Can an AI moderator probe effectively? And if so, how should that capability be evaluated?
Those questions alone could occupy a full conference session.
Another complexity lies in the iterative nature of qualitative research.
Many studies evolve as they progress. After the first few IDIs, patterns begin to emerge. Certain messages resonate while others fall flat. Sometimes an entirely new line of inquiry becomes obvious only after hearing respondents react in real time.
Campaign testing is a good example.
At the outset of a campaign test, the marketing and insights teams typically believe the materials are ready for evaluation. But five interviews in, it may become clear that none of the concepts are breaking through.
At that point, experienced moderators and their client teams often regroup. They refine the questions. In some cases, they introduce new stimuli or adjust the framing of the discussion.
This kind of mid-course correction relies heavily on human judgment and intuition.
Which raises a practical question: could an AI moderator recognize when a study needs to pivot?
Or would it simply continue executing the original discussion guide with pristine consistency?
This conversation reminded me of an earlier innovation from years ago when I was working at Market Measures under Elaine Riddell's leadership.
At the time, we had a syndicated research offering called Fastape, a detail recall system that used IVR/IVX technology to collect physician feedback following a sales call. Physicians would dial in and answer a structured series of questions delivered over the phone.
The system was highly efficient (not to mention that it was an incredibly popular product offering amongst industry marketers and sales leadership).
Doctors could complete the survey whenever it was convenient. There was no scheduling, no moderator coordination, and no logistical friction.
But the structure was rigid. The questions were fixed. There was no opportunity to explore unexpected responses or follow an interesting thread of conversation.
In other words, Fastape captured standardized input, but it wasn’t truly qualitative moderation.
Interestingly, some of the emerging AI moderation platforms share certain similarities with that model.
One of the most attractive aspects of AI moderation is scale and flexibility.
A physician could theoretically complete an AI-moderated interview whenever their schedule allows: early morning, late evening, or between patient appointments.
From an operational standpoint, this could dramatically accelerate qualitative research timelines. Instead of coordinating schedules around a limited number of moderators, interviews could run continuously.
Companies like Listenlabs AI have already demonstrated this approach in consumer research, and they are beginning to enter the pharmaceutical insights space as well. They had a booth at the recent PMRC meeting in Newark.
The value proposition is compelling: more interviews, completed more quickly, with fewer logistical constraints.
But speed is only one dimension of quality.
Another unknown is how physicians and other HCP types will respond to AI moderation.
Will they feel comfortable sharing candid feedback with a non-human moderator?
There are arguments on both sides.
Some respondents may actually feel less inhibited speaking with an AI system. Without the perceived judgment of another person, they may offer more candid or unfiltered opinions. After all, we already share extraordinary amounts of personal information with technology. Our smartphones know our routines, our preferences, even where we spend our time. In that context, speaking openly with an AI interviewer may not feel nearly as unusual as it once might have.
Others, however, may find the experience less engaging. A skilled human moderator creates a conversational dynamic that encourages reflection, storytelling, and nuance. There is also a matter of perception. Some physicians may question whether the industry is truly interested in their perspectives if they are being asked to share them with a machine rather than another professional. In those situations, the absence of a human presence could subtly diminish the depth -- or seriousness -- of the endeavor.
We simply don’t know yet how respondents will react in different contexts.
And that is precisely why thoughtful experimentation is needed.
There is another dimension that often gets overlooked: qualitative research often represents a learning experience for the client team.
One of the most powerful aspects of qualitative research is the ability for commercial teams and agency partners to listen in on interviews and focus groups. Observing physicians respond in real time can be transformative. For example: messages that seemed compelling on an agency's PowerPoint slide may suddenly fall flat when heard through the voice of a skeptical clinician.
These moments shape how teams think about strategy.
If AI moderation leads to a model where dozens -- or hundreds -- of interviews are conducted asynchronously, the learning experience may itself change. Instead of listening to conversations unfold live, teams may primarily engage with summarized outputs.
That shift could improve efficiency, but it might also alter how insights are absorbed and internalized.
My instinct is that the near future will not be defined by AI versus human moderation, but rather by hybrid models that combine the strengths of both.
In fact, many moderators are already working this way.
In my own work as a moderator, I frequently use AI tools as a kind of research copilot. Systems like ThinkGen's closed-model ThinkAEI platform help keep key interview topics visible during interviews and focus groups. They remind me of therapeutic context, competitor dynamics, and areas where the client may want deeper probing.
These tools do not replace the moderator.
But they enhance the moderator’s ability to stay sharp, organized, and responsive in real time.
Looking ahead, I suspect this hybrid approach will become increasingly sophisticated. AI systems could suggest follow-up questions, identify emerging themes across interviews, or flag areas where additional probing may be useful.
The human moderator remains central -- but the toolkit becomes more powerful.
At this point, AI moderation raises more questions than answers.
These are questions that deserve open discussion within the healthcare insights community.
The technology is evolving quickly. But the real opportunity lies in figuring out how to use it choicefully: preserving the depth and nuance that make qualitative research so valuable while embracing tools that can expand our capabilities.
So I’ll end with an invitation.
For those experimenting with AI or hybrid moderation in healthcare research, what have you learned so far?
I suspect the answers will shape how we conduct research for years to come.
