Unveiling the realities of AI in market research: Our journey with an AI-driven, in-depth Interview platform

Follow this journey into using AI in market research, using an interview platform designed to streamline qualitative studies. It highlights the need for ongoing development and clear communication to optimize AI integration in market research.

5 min read
unveiling the realities of ai in market research

We recently embarked on a venture into the realm of artificial intelligence (AI). This journey began with the development of an AI-powered interview platform. This platform aims to make qualitative studies more efficient and versatile. To lay the groundwork for our exploration, let’s start with what the tool is and what it does.

Our web application, known internally as Automoderator, facilitates interactions between users and AI through a chat-like interface, adhering to specific guidelines and scripts. These guidelines dictate factors such as conversational tone and the avoidance of biases while the script outlines the study's questions. Automoderator orchestrates the conversation, ensuring the user feels engaged while also managing question sequencing and response quality. Additionally, internal checks track question completion and trigger interface alterations as needed, such as displaying media or collecting metadata. Once all questions are answered, the platform can summarize responses using AI or provide raw data downloads for study directors.

architecture diagram


Initial success, challenges, and applications
Initial testing revealed the tool's efficacy in handling unstructured data, crafting discussion guides, moderating in-depth interviews, converting audio to text, and analyzing textual content. Starting with interview moderation, the platform evolved into a dynamic chat that can engage in personalized conversations based on predefined questions. And even accommodate multimedia elements for evaluation purposes.

The more we started using Automoderator, the more challenges surfaced. This, in turn, reshaped our expectations. Clients, enticed by the term “artificial intelligence,” envisioned a system capable of fluent conversation with human-like behaviors. Explaining the nuanced nature of the system – more intelligent than a chatbot but not a substitute for a psychologist – became an initial source of frustration.

We conducted two pilot studies using Automoderator with clients experiencing real challenges:

  1. One was with a beverage company interested in evaluating two animatics for an ad campaign, which were usually tested with quantitative methodologies and wanted to understand the reasons for open-ended questions more deeply. This involved 60 surveys among randomly selected participants from our client's own panel (average interview length 10 mins). Additionally, our client conducted simultaneous online quantitative sampling with 200 cases.

  2. The other study was for a community services company that was interested in getting to know young adolescent users of their services at a qualitative level. This involved 36 x 1 hour interviews recruited by referrals. 

Technical limitations arose throughout the stages of these projects, which we describe below.

  1. Subjectivity in Question Depth: The system's depth of questioning relies on participant responses, posing challenges due to the impracticality of setting parameters for all possible terms.

  2. Conditional Jumps Between Questions: Programming conditional jumps or branching logic proved unattainable. The system strictly adheres to a predetermined sequence of questions stored in an auxiliary file, making the customization of pathways too extensive.

  3. Avoidance of Confrontation: AI systems record participant changes of opinion without delving into the underlying reasons. While this approach ensures neutrality, it falls short in exploring the motivations behind shifts in perspective.

  4. Misinformation Topics: Engaging in conversations about topics prone to misinformation, such as climate change or vaccines, poses challenges for AI in providing accurate information. This limitation, though less relevant in typical market research discussions, becomes critical in health-related studies where misinformation can impact integrity

  5. Branding Confusion: Even after specifying the brand, the AI can be confused by statements using the brand word in another sense, potentially misinterpreting the context of the conversation. 

  6. User Experience: Users interact with the platform through various devices, and interruptions in the connection between users and AI present challenges, necessitating anticipation and addressing scenarios like internet signal disruptions or AI overload to ensure a seamless user experience. Additionally, managing user expectations regarding the AI's capabilities, such as inquiries about ending the conversation, is crucial.  

Managing Complexity and Enhancing AI Integration:
Ongoing development and refinement is crucial when navigating these kinds of technical limitations. Tackling misinformation challenges requires continuous updates to the knowledge base, ensuring the AI remains well-informed on evolving topics. Addressing branding confusion involves refining the system's contextual understanding, distinguishing between brand-related discussions and general language use.

Moreover, optimizing user experience necessitates proactive measures, including robust error handling for connectivity issues and clear communication to users about the AI's capabilities and limitations. As technology evolves, the integration of user feedback becomes invaluable, guiding developers in enhancing the platform's responsiveness and adaptability.

mobile

Conclusion:
This exploration of AI in market research has not only uncovered the potential of innovative technologies but has also shed light on the nuanced challenges that accompany their integration. In this example, our AI-driven interview platform helped provide a broader perspective and uncover insights that might be elusive in a purely quantitative approach.

However, it's essential to acknowledge that while AI plays a pivotal role in enhancing efficiency and versatility, it does not replace the depth of understanding and empathetic connection that skilled human moderators, especially psychologists, bring to qualitative studies. The intricacies of human behavior, emotional nuances, and the ability to adapt dynamically to participant responses remain unparalleled strengths of human-led interactions.

In the evolving landscape of market research, embracing AI is not about replacing human expertise but rather integrating it seamlessly. AI’s ability to handle certain tasks efficiently makes it a valuable ally, particularly in scenarios where scalability, cost-effectiveness, and processing vast amounts of data are paramount. Yet, it's equally crucial to recognize its limitations and position it as a complement to, not a substitute for, the irreplaceable depth that psychologists bring to qualitative interviews. As we navigate this dynamic intersection of technology and human insight, the synthesis of AI and human expertise emerges as a powerful approach. 

Related