Companies invest significant budgets in market research. Yet in practice, a familiar problem repeatedly emerges: The insights gained end up in presentations, reports, or databases and are only partially utilized. High investments in understanding target audiences. Different approaches to target audience segmentation… Yet segmentations often remain superficial and difficult for marketing management to use effectively.
This raises a central question for many marketing and innovation teams: How can research findings be made accessible more quickly, easier to understand, and consistently actionable?
One possible answer lies in the stronger integration of artificial intelligence, particularly through AI-supported persona avatars.
This article shows how source-informed AI avatars can help bridge the gap between statistical analysis and deep customer understanding—and where their limitations lie.
From Static Personas to Interactive AI Avatars
Classical personas have been part of the standard toolkit of market research for many years. They serve as model representations of typical characteristics within a segment, making complex target groups more structured and manageable.
However, they often remain static and PowerPoint-based, typically consisting of profiles, photos, and a few typical statements.
The idea behind so-called QUPAs (Quelleninformierte Persona-Avatare – Source-Informed Persona Avatars) takes this concept one step further. Instead of being presented as a document, the persona becomes an interactive conversation partner. Rather than relying on freely generated AI content, the avatar is based on quantitative survey data and qualitative interview data from a real person.
The key difference from a standard chatbot lies in the specific AI architecture known as Retrieval-Augmented Generation (RAG).
For Insights Managers, this creates a new way of accessing research findings. Instead of having to search for information in reports or databases, they can ask questions in natural language and receive immediate, source-based answers.
Why RAG Architecture Can Increase Trust in AI
RAG architecture describes a system design in which the language model (LLM) and the knowledge source are functionally separated. The system does not generate answers based on the model’s internal knowledge, but on external, controlled data sources. This means that the information output is consistently tied to verifiable sources.
The technical process follows a clear principle: When a user submits a query, the system identifies relevant documents and incorporates them into the generation process. Based on these sources, the language model produces a reliable answer.
In a corporate context, this architecture creates greater transparency. Statements can be traced directly back to the underlying documents, allowing users to better understand and evaluate decisions.
At the same time, this architecture significantly reduces the risk of so-called hallucinations, as generation is systematically linked to available information. However, the quality of the results remains dependent on the quality and structure of the underlying data. Michael Klesel and Felix Wittmann from Frankfurt University of Applied Sciences have described this relationship in their study on RAG.
A particularly relevant practical effect concerns missing information: If no corresponding data is available, the AI does not generate a hypothetical answer. Instead, it transparently communicates that the information is not available.
5 Steps to Creating Source-Informed Persona AI Avatars
Creating a QUPA follows a structured process that systematically combines quantitative breadth with qualitative depth.
- As part of the quantitative data collection, a standardized online survey is first conducted with a larger sample. Behavioral patterns, preferences, and attitudes are captured on a statistically robust basis.
- The next step is target audience segmentation, in which statistical methods divide the overall data into homogeneous segments. The objective is to identify clearly distinguishable groups with similar patterns.
- Building on this, representative individuals are selected from the respective segments. These individuals best embody the characteristics identified in the previous step.
- The next stage involves qualitative in-depth interviews with the selected individuals. The interviews are used to explore motives, decision-making logic, and contextual information in detail and are fully transcribed.
- Finally, the data is consolidated at the individual level by combining the quantitative survey data and qualitative interview data of each individual. This creates an individual data profile (n = 1), which forms the basis for the respective AI avatar.
The AI persona avatars generated in this way evolve from mere data points into digital conversation partners. They are based exclusively on this closed data environment. They do not operate with general world knowledge, but solely with the stored, person-specific information.
5 Tips for Using AI Personas in Market Research
To ensure successful implementation, it is advisable to structure the process around the following principles:
- Primary data as the foundation. A QUPA is only as good as its source data. Invest in methodologically sound qualitative in-depth interviews with sufficient interview duration (at least 60 minutes). Short, superficial interviews produce thin avatars—and thin avatars give thin answers.
- Segmentation first, avatars second. The statistically grounded segmentation of the overall data (Step 2 in the process) is the strategic foundation. Only by knowing which segments are relevant can you select the right representatives and build avatars that reflect the market structure.
- Multiple representatives per segment. Each segment is internally diverse. Deliberately select two to three people per group who embody different expressions of the segment profile. In the joint project of Manuel Helpenstein (HARIBO GmbH & Co. KG) and Stefanie Sonnenschein (Interrogare GmbH), several avatars were deliberately created for certain segments in order to generate breadth rather than false precision [2].
- Hallucination protection as a design principle. Explicitly define from the outset which subject areas the AI avatars should not answer. A clearly defined knowledge base protects against unwanted AI-generated fabrications and preserves internal stakeholders’ trust in the system.
- Internal democratization. The real leverage of QUPAs lies not in the data collection phase, but in the usage phase. Give brand managers, product teams, and senior management direct chat access to the avatars. In this way, market research is not perceived merely as a data collection instrument, but is established as an interactive knowledge resource available to the entire organization at any time.
Conclusion
The real innovation of AI-supported persona avatars lies not in artificial intelligence itself, but in the way research findings are made accessible and actionable.
Current developments show that combining quantitative data, qualitative interviews, and modern RAG technology can create a new way of accessing target audience insights. The technology does not replace research; instead, it significantly expands its practical applicability.
However, AI avatars do not replace conventional concept or innovation testing. For new products, concepts, or communication ideas, there is no empirical data available for the avatar to draw upon.
The future of market research therefore does not lie solely in collecting additional data. Equally important will be the ability to activate existing insights more efficiently and integrate them consistently into everyday business practice.
This is precisely where Insights Activation comes in.
The central question is therefore not whether companies need more data, but how they can make more effective use of the knowledge they already possess. The Insights Activation Workshop by BESTVISO supports you in translating your research findings into concrete actions.
Sources:
-
- Klesel, M., & Wittmann, H. F. (2025). Retrieval-Augmented Generation (RAG). Business & Information Systems Engineering, 67(4), 551–561
- Sonnenschein, S., & Helpenstein, M. (2026). Zielgruppe zum Anfassen – mit HARIBO und KI [Konferenzpräsentation SUCCEET 2026]. Interrogare GmbH & HARIBO GmbH & Co. KG.