Where to Use AI Moderators: Considering Research Phases and Timing

On July 28, 2026, Product Force officially launched "Uni-Research AI Interview."
https://prtimes.jp/main/html/rd/p/000000067.000117669.html

An AI asks questions via voice to actual participants and conducts follow-up questions based on their responses. With 30 to 100 participants, the research can be conducted in a short period without the need for individual scheduling. After the responses, recordings, transcripts, and summaries are created, and an AI report analyzing across multiple responses is also generated.
The value the company is promoting is the ability to obtain qualitative information in a quantitative manner.

While asking common questions to dozens of people, it additionally confirms the reasons and background for each individual's response. Research that was previously difficult to conduct due to constraints on the number of participants and the labor involved in fieldwork is beginning to become a realistic option.

The first scenario that came to mind where this mechanism could be easily utilized immediately was mini-interviews after CLT (Central Location Tests).

In CLT, the same stimuli—such as prototypes, tastes, scents, packaging, or advertising copy—are presented to multiple participants to obtain evaluations. Afterward, there are times when the reasons for the evaluations or points of interest are asked in a short 15 to 20-minute session.

In mini-interviews after CLT, the basic flow is to ask questions about what the participant just experienced. While linking evaluation scores to their statements, additional confirmation is made regarding "why they gave that score," "where the evaluations diverged," and "under what conditions it would be acceptable."
When human researchers ask questions in sequence over a short time, the more participants there are, the more likely it is that variations in questioning or missed points will occur, and the number of researchers and time required to conduct sufficient follow-up questions for everyone also increases.

If AI interviews are combined here, it becomes possible to conduct in-depth confirmation tailored to each participant's response while maintaining common evaluation items. It also becomes easier to analyze evaluation scores and their backgrounds across the board.

For example, even with the same "somewhat favorable" rating, the meaning for commercialization differs between those who evaluated the taste itself, those who felt it was an improvement over the current product, and those who would consider purchasing it depending on the price.
By using AI-driven follow-up questions, it is possible to obtain these evaluation conditions at a certain scale of participants. Prototypes and expressions can be modified after the research, and the question design can be improved for the next CLT.

Common stimuli, known evaluation axes, uniform questioning, and ease of retesting are all present. I believe that mini-interviews after CLT are an area with a particularly high degree of compatibility with AI moderators at this moment.

Conditions under which AI moderators function well, as seen from CLT considerations

When I organize the reasons why I thought of mini-interviews after CLT as a utilization scenario for AI moderators, several conditions become apparent.

All participants have just experienced a common stimulus, and the research side has evaluation items they want to confirm. As a result, while follow-up questions tailored to each participant's response are necessary, the starting point and scope of the questions are easy to set.
Research results can be reflected in modifications to prototypes or expressions, and in the next test.

The questions are somewhat clear, and one can ask many people under the same conditions and connect the obtained results to the next trial. I think this is the kind of situation where the uniformity, participant scale, and fieldwork speed of AI moderators shine.

When considering this condition by extending it to other types of research, the compatibility between AI and humans cannot be fully organized just by product category or the name of the research method.
Even in the same new business, the necessary way of asking questions changes between initial exploration to find customer issues or evaluation axes, and the stage of confirming reactions to a formed concept. Even with the same existing product, the need to expand questions differs between research to confirm known dissatisfactions and research to find insights that become the differentiator from competitors.

Reflecting on my own experience, the difference in this was manifested more in the progress and burden when creating reports than in the feeling during the fieldwork.

Three states visible from the experience of report creation

From the research I have been in charge of so far, the progress from fieldwork to report creation is broadly divided into three states.

1. A state where the questions are organized and can be acted upon after the research

In situations where preliminary issue organization has been done and products or measures can be modified after the research, the process from fieldwork to report creation proceeds relatively smoothly.

It is clear who is making what judgment based on the research, and the points to be confirmed are narrowed down. If some of the questions do not function sufficiently, they can be supplemented by additional research or modifications to the measures.

In a report, the information obtained can be organized according to the initial decision-making tasks and connected to the next trial.

Mini-interviews after a CLT are close to this state. There are common stimuli and evaluation criteria, and results can be reflected in revisions to prototypes or expressions. The homogeneity, sample size, and speed of AI moderators are well-suited here.

2. A state where there are oversights in the questions, but the situation can be adjusted after the research

When there are significant oversights in the preliminary task organization, but the situation still allows for modifications to products or measures, the burden of report creation increases. At the same time, an exploratory interest also emerges.

Through dialogue with participants, important issues different from the initially set questions are discovered. By combining the statements of multiple participants, the conditions under which products or services are chosen, the turning points where usage stops, and the discrepancies with the customer profile assumed by the client become visible.

In the report, the organization according to the initial research tasks and the construction of a new structure revealed by the actual research are performed simultaneously.

If the situation allows for moving products or measures, those findings can be reflected in the next prototype, additional research, or design changes. This state is likely to occur in the initial exploration of new businesses or in insight exploration in markets with high preference.

In this phase, a human moderator plays the role of expanding questions based on participant statements and discovering axes that were not visible before the research (a division of labor where humans discover and AI organizes is a good fit).

3. A state where there are oversights in the questions, and the situation cannot be adjusted after the research

When there are oversights in the preliminary task organization at a stage where product specifications, business policies, investment scales, and external commitments are fixed, a heavy psychological burden is placed on report creation.

It is a daily occurrence to see realities that differ from fixed premises while conducting research.
The assumed customer issues do not exist. The value thought to be emphasized is not a reason for selection. Unchangeable specifications of services or products are major barriers to use. The competitors set by the business side differ from the options that consumers are actually comparing.

The more skilled a moderator is at exploration, the more they will discover such discrepancies during the research and pursue the background deeply.

The discovered realities are essentially important enough to touch the core of the business. However, the scope that the business side can respond to narrows as decisions become fixed.
The report creator ends up (in a sense, fruitlessly) adjusting expressions and recommendations between the observed realities and the range that the organization can currently accept and act upon.

The first issue this state indicates is the timing and method of the research.

The timing for conducting exploratory research should be placed at a stage where business premises, product specifications, and investment decisions can be moved. Human moderators are appropriate at that time.
At a stage where major decisions are fixed, the role of research should be shifted to limited confirmation in accordance with the currently changeable range, and evaluation reasons and condition differences should be obtained homogeneously. In this phase, the characteristics of an AI moderator that proceeds according to pre-set questions and deep-dive intentions are effective.

Summarizing the three states, it looks like the following.

1. A state where questions are organized and the situation can be adjusted after the research
Results can be organized according to initial decision-making tasks and connected to revisions of products or measures. AI moderators are easy to place at the center, such as in mini-interviews after a CLT.

2. A state where there are oversights in the questions, but the situation can be adjusted after the research
A human moderator explores axes that were not visible before the research.
The formed hypotheses can be expanded to a large number of people through AI interviews.

3. A state where there are oversights in the questions, and the situation cannot be adjusted after the research
Reconsider the timing for conducting exploratory research. If research is conducted after major decisions are fixed, use AI moderators limited to the range that can currently be changed.

Dividing these three reveals that the compatibility between AI and humans changes not because of the superiority or inferiority of questioning techniques, but because of the state of the questions and the room to move the business after discovery.

In new business ventures, the order of exploration and validation is critical.

In the early stages of a new business, you can change the customer profile, value proposition, concept, and revenue model. This is a period when the shape of the business can be significantly altered based on research findings.

At the same time, key lines of inquiry are also open.
Even if the business issues—such as whose problems you want to solve and what those problems are—are organized, the axes to which consumers actually respond may not be visible.

For example, a sense of security during use might be valued more than the expected benefits. Connection to one's identity or relationships with others might influence usage more than functional value. The actual competition might be behaviors or habits different from the product the business had envisioned.

New axes can be discovered through words the subjects use casually, experiences that deviate from the initial theme, discrepancies between statements and actions, and unexpected comparisons with other categories.
At this stage, it is better for a human moderator to expand on questions based on the subject's responses and to form a structure through analysis after the research is conducted.

After these initial explorations reveal evaluation axes, behavioral turning points, and usage contexts, the suitability of AI interviews increases.
In other words, the formed hypotheses are converted into questions and deep-dive conditions, and then deployed to a larger number of subjects. This confirms which segments exhibit that perception and under which conditions the evaluation changes.

In new business, the sequence of human-led exploration followed by AI-led expansion, and then re-exploration based on the results, is assumed to be the most convincing and efficient approach.

Research for existing businesses is divided by the business growth phase.

In research concerning existing products and services, it is easier to judge the suitability of an AI moderator by looking at the growth phase of the market and the company's business rather than the product category.

When the entire market has room for growth and the company can increase customers within existing evaluation axes, it is easy to set the points that need to be validated.

Confirming reasons for selection and dissatisfaction with usage along judgment axes already shared within the category—such as taste, aroma, capacity, ease of use, function, price, purchase location, and awareness channels—is important in this situation.
In this phase, it is easy to place an AI moderator at the center of the research.

While maintaining common questions, you can additionally confirm reasons and usage contexts for each subject. The greater the market's room for growth, the more likely it is that improvements along known judgment axes and the expansion of customer segments will directly lead to business results.

At the same time, by combining this with DI or GI sessions for small groups (about 3-6 people) led by a human moderator, you can check if there are any significant oversights in the evaluation axes set by the research team.

You can grasp numerous trends through AI interviews and read characteristic cases or reactions outside the established axes through human interviews. Alternatively, after checking the question design through small-group human research, you can deploy AI interviews to a larger number of people.
Through this combination, you can supplement conditions that were not visible during design while leveraging the homogeneity and scale of AI.

In situations where the market is maturing and the company's share or sales have plateaued, the role required of research changes.

Even if you accumulate improvements along existing axes—such as slightly improving the taste, making the container easier to use, or making advertising expressions easier to understand—the difference in selection compared to competitors may not widen. In this phase, what is needed is not the task of confirming superiority or inferiority within current evaluation items, but the task of re-grasping the framework through which consumers choose products.

Even for food and daily necessities, if share has plateaued and room for growth is hard to find through category-level improvements alone, projective techniques and indirect questions become necessary.
By introducing judgment axes considered natural in other categories, selection criteria that were difficult to see within that category may emerge.

A human moderator can rearrange comparison targets and the direction of questions while observing how subjects choose their words, where they pause in their explanations, and the discrepancies between their statements and actions. From there, they can form the next differentiation axis the business can take.
After new evaluation axes or hypotheses are formed through this exploration, AI interviews can be utilized to confirm with a large number of people which segments accept the new axes, under which conditions the evaluation changes, and how reactions differ between existing and new customers.

The proper use of AI in existing businesses can be summarized as follows.

Mini-interviews after CLT
Use AI immediately after evaluating common stimuli to uniformly obtain reasons for evaluation, acceptance conditions, and discomfort from all participants.

Existing businesses with room for growth
Use small-scale DI or GI to refine question design, then deploy to a large audience using AI. Alternatively, you can have AI lead the process and have humans verify characteristic cases.

Exploring stagnation factors and selection criteria
Explore selection criteria that humans may have overlooked, then use AI to verify the hypotheses formed.

In this framework, even for low-unit-price, high-frequency purchase categories like food, beverages, and daily necessities, human moderators are better suited when market share has plateaued and new selection criteria are needed.

Even for categories with high preference or relatively high unit prices, such as cosmetics, supplements, and entertainment, AI moderators can be utilized in research aimed at improvement or customer expansion along known evaluation axes, provided the overall market is growing.

In phases where growth is possible along known axes, use AI to ask questions broadly.

In phases where growth has stalled along known axes, have humans explore the framework of selection.

In phases where newly formed axes need to be verified, use AI to expand the scope again.

I believe this is the most efficient approach that best leverages the respective strengths of each.

Suitability of AI Moderators vs. Human Moderators (Summary)

The following shows the suitability of each as a primary research method for different research phases.

◎: Particularly suitable
○: Suitable
△: Suitable for complementary use
×: Less suitable as a primary method

Mini-interviews after CLT

AI: ◎ Human: △
Immediately after evaluating a common stimulus, collect reasons for evaluation, acceptance conditions, and any sense of discomfort uniformly from all participants.
Examples of research formats:
Short interviews after CLT, individual interviews after tasting/trial, confirmation after presenting advertisements or packaging

Confirmation and improvement along known evaluation axes

AI: ○ Human: ○
At the stage where confirmation items such as taste, scent, price, ease of use, and functionality are clear, collect reasons for selection and conditions for improvement. Gathering information from many people via AI and then verifying missing evaluation axes or characteristic cases through small-scale human research increases the accuracy of the results.
Examples of research formats:
AI interview + small-scale DI, AI interview + GI, post-usage interviews, CLT

Exploration of unknown evaluation axes or selection criteria

AI: × Human: ◎
Discover evaluation axes not anticipated by the business side or selection criteria that the subjects themselves cannot clearly explain. Use projective techniques, indirect questioning, comparisons with other categories, and verification of discrepancies between statements and actions, while adjusting the direction of questions during the research.

Examples of research formats:
DI, GI, HV (Home Visits), behavioral observation, interviews after diary studies
This applies to initial exploration for new businesses, searching for differentiation axes in mature market categories, and exploring stagnation factors for businesses where market share has plateaued.

Verification of hypotheses formed after exploration

AI: ◎ Human: △
Expand and verify the evaluation axes, psychological conditions, and behavioral turning points discovered through human exploration across a larger number of people.
Examples of research formats:
AI interviews after DI/GI/HV, AI interviews after concept presentation, additional confirmation after exploratory research

Comparing Multiple Concepts

AI: Yes, Human: Yes
Compare the reasons for evaluation and acceptance criteria for multiple proposals using a common framework. AI confirms the breadth of reactions, while humans delve into unexpected responses and decision-making processes.
Examples of research formats:
Concept CLT, AI interviews, DI, GI

Differentiating by Business Phase

When a market has room for growth and you aim to grow using known evaluation criteria, it is efficient to center the research on AI interviews combined with a small number of DI or GI sessions.
When a market is mature and share or sales have plateaued, the direction for deep dives is not fixed, so human-led exploration should come first. The new evaluation criteria and differentiation hypotheses formed there can then be expanded to a larger audience via AI interviews.
The early stages of a new business should also begin with human-led exploration. Only after customer issues, usage contexts, actual competitors, and evaluation criteria have been formed should you move to AI-based confirmation.

Differentiating by Decision-Making Timing

Before major decisions are finalized, it is better to conduct human-led exploration first. Unexpected discoveries can be reflected in customer personas, value propositions, product specifications, and business policies.
After major decisions are finalized, narrow the scope of confirmation to currently changeable specifications, communication, sales methods, and post-launch monitoring items. At this stage, uniform confirmation by an AI moderator functions well.

What is Required of Research Agencies and Clients

With the spread of AI interviews, I predict that setting issues before research and sharing an understanding of the current situation will influence the value of research results more than ever before.

Who is the research for, and what are they trying to decide? How much can the product, measures, or business policy be changed based on the research results? Do you want to discover unknown evaluation criteria, or do you want to expand already formed hypotheses to a large number of people? The order in which AI and humans are deployed changes depending on these conditions.
Research agencies are required not only to choose between AI and human moderators but also to design the research process, including the order of exploration, hypothesis formation, expansion, and confirmation.

Those using research are required to place exploratory research at a time when unexpected discoveries can be reflected in the business, and to clarify the scope of what can currently be changed after major decisions have been finalized.

Product Force's AI interviews are making the acquisition of qualitative information from large numbers of people a reality.

In phases where known questions are expanded to many people, the scope that AI moderators can handle will continue to grow.
In phases where unknown axes are discovered and business premises are restructured, the exploratory power of human moderators will remain important.

Rather than choosing between AI or humans on a project-by-project basis, design who asks what and at which stage. I believe that design capability is the condition for connecting AI interviews to business decision-making.

When I was a new graduate, the arrival of Windows 95 and the spread of the internet changed the way we worked and the nature of our work significantly within a few years.
I feel that today, that same dynamism is arriving at a scale hundreds of times larger and at a speed dozens of times faster.
I think it is a very interesting era, but as I enter the latter half of my life, I also have a different feeling about how far I can keep up, or how much I want to keep up.

I am recording that feeling and my thoughts at this very moment.

(End)